doi
stringlengths
10
10
chunk-id
stringlengths
1
4
chunk
stringlengths
1
1.66k
id
stringlengths
10
10
title
stringlengths
19
148
summary
stringlengths
345
1.92k
source
stringlengths
31
31
authors
sequence
categories
sequence
comment
stringlengths
4
284
journal_ref
stringclasses
14 values
primary_category
stringclasses
16 values
published
stringlengths
8
8
updated
stringlengths
8
8
references
list
2209.07686
63
COT CC OT (ours) C OT CC OT (ours) C OT CC OT (ours) C OT CC OT (ours) GSM-8 K(Table 33) 27.4 % 29.1% 46.9% 52.4% 65.6% 62.5% 53.2% 56.2% DATE(Table 34) 44.7 % 51.3% 54.1% 61.1% 69.2% 70.0% 65.3% 69.1% SPORTS (Table 35) 93.7 % 94.6% 63.5% 74.5% 98.2% 98.4% 95.4% 97.4% SORTING (table 41) 55.3 % 60.2% 26.8% 99.8% 27.2% 100% 71.2% 88.6% CCOT outperforms C OT while employing prompts with fewer tokens. The task solve rate of CC OT remains relatively high as we scale the model to PaLM-540 B, highlighting the efficiency of CC OT. AppendixTable 36 compares the average number of input/output tokens between C OT and CC OT. On average, CC OT (our approach) reduces the number of input (1.39 ) and output tokens (1.58 ). 9. R ELATED WORK
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
64
(our approach) reduces the number of input (1.39 ) and output tokens (1.58 ). 9. R ELATED WORK Broadly, this paper intersects with a growing body of work on prompting and large language model reasoning (Brown et al., 2020; Chowdhery et al., 2022). Below, we review the most relevant work in these directions. Exploring the role of examples in few-shot setup With the growing interest in few-shot prompting, several works have explored the role that in-context examples play in the success of few-shot prompting. Notably, Min et al. (2022) find that label correctness is not crucial for the success of the models, and even random labels might lead to competitive performance. Building on this work, Kim et al. (2022) find that the role of the correctness of the labels might be task-dependent. Our findings concur with these methods on label correctness—for GSM-8 K, label correctness is not material, whereas it plays a larger role for S PORTS . Surprisingly, we find cases where wrong examples can improve performance by being better indicators of the end task. Our work goes beyond label correctness explored by these methods and teases apart the
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
65
Surprisingly, we find cases where wrong examples can improve performance by being better indicators of the end task. Our work goes beyond label correctness explored by these methods and teases apart the role of placeholder symbols, patterns, and text in the success of few-shot models. Finally, in addition to comparing the final results (outcome), we also focus on the mechanism (attention patterns), allowing us to reveal instances where model reasoning is identical. Our results also resonate with the work of Reynolds & McDonell (2021), who found that one of the key roles played by the prompt is to remind the model of the underlying task. Razeghi et al. (2022) find that pre-training term frequencies can somewhat explain the success of few-shot methods. In line with their work, our experiments on S PORTS also show that D IRECT prompting method is most suited for easy questions (involving personalities and activities found on the web). Finally, Xie et al. (2021) show that in-context learning enables a large model to infer a shared concept between the examples, possibly leading to better task understanding. Our studies on the role of prompt, especially examples where wrong examples lead to better output (e.g., for S ORTING ), add more empirical evidence to this finding.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
66
wrong examples lead to better output (e.g., for S ORTING ), add more empirical evidence to this finding. Further, we show that a symbiotic relationship between text and patterns allows a more efficient inference of task instruction. Least to most prompting. Zhou et al. (2022) help the model generate a chain of thought by first asking the model to generate the sub-questions for the given problem. Next, the model is asked to answer the subquestions, and finally, the sub-questions, along with sub-answers, are combined to generate the final result. This work is closely related to Kojima et al. (2022), the latter distinguished by generating the rationale from 13 a large language model directly. We posit that Zhou et al. (2022) derives its key strengths from its ability to generate useful sub-steps. This resonates with our finding that the key contribution of CoT is the extraction of meaningful sub-steps. Rationale generation as an intermediate step The idea of generating rationales as an intermediate output for reasoning and structured generation tasks has shown promising results for fine-tuned models (Ling et al.,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
67
for reasoning and structured generation tasks has shown promising results for fine-tuned models (Ling et al., 2017; Sun et al., 2019; Rajani et al., 2019; Shwartz et al., 2020; Madaan et al., 2021; Nye et al., 2021; Gu et al., 2022). Recently, Wei et al. (2022) proposed chain-of-thought prompting, which shows that few-shot setups can also be improved by making the model first generate an understanding of the output. As a natural extension to C OT, Wang et al. (2022b) seek to improve C OT using over-generation using selfconsistency . They sample multiple outputs, and take a plurality vote (i.e. most frequently generated answer) to arrive at the final answer. This general idea of enforcing the model outputs to be consistent has also been explored for symbolic-commonsense reasoning (Kassner et al., 2021). The efficacy of this approach is corroborated by Wang et al. (2022a), who report that taking multiple samples helps a model become robust to settings in a few-shot setup. Our work looks at understanding the efficacy of C OT in the standard setup of generating a single output per input.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
68
to settings in a few-shot setup. Our work looks at understanding the efficacy of C OT in the standard setup of generating a single output per input. Prompt selection. Several works have recently explored the design of the prompt—a process often called “prompt engineering” (Le Scao & Rush, 2021; Liu et al., 2021c). The methods include dynamically creating prompts based on the question (Liu et al., 2021a; Rubin et al., 2021; Poesia et al., 2021), formatting the prompt as a list or questions (Mishra et al., 2021; Rubin et al., 2021), improving order of examples in the prompt (Lu et al., 2022), and providing instructions in the task (Ouyang et al., 2022). Unlike these techniques, C OT is relatively robust to minor changes in the prompt design. Thus, the findings of our work might be more generally applicable. Explaining model behavior using counterfactual prompts and attention. As noted by Jacovi & Goldberg (2020), an explanation of a deep learning system typically serves two different purposes: i) plausibility, which aims to provide an interpretation of system outputs that is convincing for humans, and ii) faithfulness, which aims to capture the actual reasoning process of a model. Our study requires both and uses different
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
69
which aims to capture the actual reasoning process of a model. Our study requires both and uses different means to achieve them. We utilize counterfactual prompts to interpret the system outputs to aid human understanding. This is similar to using posthoc analysis tools (Ribeiro et al., 2016; Lundberg & Lee, 2017; Liu et al., 2021b), which also focus on analyzing outputs without concern for the details of the model. To get a glimpse of the model’s inner workings, we leverage attention (Vaswani et al., 2017), a ubiquitous mechanism in NLP. While the broader question on the utility of attention for posthoc analysis is still open (Jain & Wallace, 2019; Pruthi et al., 2020), there is some evidence to show that attention can act as an explanation (Wiegreffe & Pinter, 2019). Finally, the utility of any explanation mechanism is closely tied to the users and application domain (Kaur et al., 2020; Burkart & Huber, 2021). As our analysis shows, attention adds intuition and insights to the empirical findings. Counterfactual explanations seek to explain the behavior of a model by performing a what if analysis on
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
70
intuition and insights to the empirical findings. Counterfactual explanations seek to explain the behavior of a model by performing a what if analysis on examples (Mothilal et al., 2020; Stepin et al., 2021; Verma et al., 2020; Poyiadzi et al., 2020; Goyal et al., 2019). While counterfactuals can be misleading due to artifacts (e.g., see (Laugel et al., 2019; Slack et al., 2021)), they offer a tractable solution for probing large models like PaLM and GPT-3. Notably, unlike finetuned methods, the most important examples for generating the model output are readily available. Thus, counterfactual inputs that show a consistent and systematic change in the model performance are more likely to reflect the model’s behavior. 10. C ONCLUSIONS This work evaluates the capacity of C OT to elevate complex reasoning in three state-of-the-arts LLMs, PaLM, GPT-3, and C ODEX . We systematically assembled a series of controlled counterfactual experiments. 14 Our results show the initial inklings of connection between text, patterns, and reasoning in LLMs. Our study indicates that the symbiosis of text and patterns bears more weight in the chain of thought reasoning process.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
71
indicates that the symbiosis of text and patterns bears more weight in the chain of thought reasoning process. In addition, we assert that text is a channel to extract semantic patterns, unlocking the ability of these models to mold correct answers. ACKNOWLEDGEMENTS We would like to extend our gratitude towards Kathy Meier-Hellstern, Denny Zhou, Victor Veitch, Saleem Abdulrasool, Shruthi Sukumar, Milad Hashemi, Douglas Eck, Christian Szegedy, and Stella Aslibekyan. We also thank the PaLM team and our extended team at Google Research, Brain Team who enabled this research and helped us conduct our experiments. 15 REPRODUCIBILITY STATEMENT We take the following steps to enable the reproducibility of our work. Controlling for randomness due to the order of examples. We run each experiment with multiple random seeds to control for randomness due to the order of examples in the prompt. We report the average and standard deviation of the results across all the random seeds. Additionally, we conduct statistical significance tests (McNemar’s test (McNemar, 1947)) to compare the results across different prompts. Finally, we evaluate the agreement in output generated by different models using Cohen’s kappa ( ) metric.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
72
evaluate the agreement in output generated by different models using Cohen’s kappa ( ) metric. Reproducing results. We open sourced the code at https://github.com/google-research/ google-research/tree/master/l2da/learned2design . In addition, we have provided scripts for one-click reproduction of the results for the publicly available models in the paper. Experiments with publicly available models. We experiment with three different language models: PaLM, GPT-3 ( text-davinci-002 ), and C ODEX (code-davinci-002 ). PaLM is not publicly available as of submission time, but the provided source code is compatible with OpenAI API v0.23.0, and can work with any OpenAI models. Finally, C ODEX is free to use as of submission time that further helps with the reproducibility of the results. Prompts and outputs. All the prompts are included in the prompts/ directory. The generated outputs from GPT-3 and C ODEX are provided in the outputs/ directory. Each output file follows a standard naming convention: task _name _model _name _sseed:jsonl . Each line of the output file is a json with three
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
73
fields: 1. prompt + test question ( “question” ), 2. generated answer ( “generated_answer” ), 3. true answer (“answer” ) . A shortened output example is as follows: 1{ 2 " question ": "Q: Is the following sentence plausible ? ’Jonas Valanciunas beat the buzzer .’\ nA: Jonas Valanciunas is a basketball player . Beating the buzzer is part of basketball . The answer is yes ... Q: Is the following sentence plausible ? ’ Malcolm Brogdon banked the shot in .’\ nA: Malcolm Brogdon is a basketball player . Banking the shot in is part of basketball . The answer is yes .\n\n\nQ: Is the following sentence plausible ? ’Sam Darnold passed the puck .’\ nA: Sam Darnold is an American football player . Passing the puck is part of hockey , not American football . The answer is no .\n\n\nQ: Yes or no: Is the following sentence plausible ? \" Javi Martinez launched the desperation heave .\"\ nA :", 3 " generated_answer ": " Javi Martinez is a soccer player . Launching a
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
74
desperation heave .\"\ nA :", 3 " generated_answer ": " Javi Martinez is a soccer player . Launching a desperation heave is part of basketball , not soccer . The answer is no .", 4 " answer ": "no", 5 " is_correct ": 1 6} 16 ETHICS STATEMENT Disseminating reasoning into machines has numerous benefits and applications, from algorithmic reasoning (Li et al., 2022) to code generation (Chen et al., 2021b; Poesia et al., 2021) and formal verification (Wu et al., 2022). While this research does not directly enhance the reasoning capabilities of large language models, it identifies several systematic behavioral patterns in the functioning of few-shot models. Similar to any technological advances, this work has risks of detrimental societal impact. However, anticipating potential future downsides of such methods is challenging. More than ever, the research community’s utmost responsibility is to acknowledge these risks candidly and reflect on practices and strategies to prevent potential harm. Environmental impact. Training large language models devour a nontrivial amount of compute resources,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
75
potential harm. Environmental impact. Training large language models devour a nontrivial amount of compute resources, a limiting factor for frequent training. Few-shot prompting is an appealing solution for mitigating the unfavorable environmental impact of large language models by evading additional iterations of training and dataset collection. Ours and similar studies may lead to more effective prompting techniques and bring technological innovation to the architecture of large language models, especially regarding their reasoning capabilities. Therefore, we hope that the significant compute used in this work can help promote positive environmental outcomes. Finally, aligned with the credible concerns of the research community, we recognize the longer-term risk of Excellence in artificial intelligence, primarily when it boils down to human reasoning. While distilling comparable human reasoning to machines offers many benefits, undisciplined and uncontrolled progress in this area could be alarming, especially in the presence of bad actors. Effective reasoning in machines as a result of our study, even though not directly, can lead to algorithmic advances that may facilitate bad actors in developing malicious software and systems with human-level capabilities. We also want to acknowledge a large body of researchers that has greeted innovations in large language
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
76
in developing malicious software and systems with human-level capabilities. We also want to acknowledge a large body of researchers that has greeted innovations in large language models and steady scaling of models with skepticism, questioning the connection between human reasoning and large language models (Han et al., 2022; Binz & Schulz, 2022). Additionally, recent efforts have started exploring the relationship between the structure of contemporary language models and the human mind (Schrimpf et al., 2021; Tang & Ha, 2021; Whittington et al., 2021). Despite their progress, the rationale behind C OT’s mimicking human reasoning, and any potential connection with linguistics is so far an uncharted territory. BIBLIOGRAPHY Michael Ahn, Anthony Brohan, Noah Brown, Yevgen Chebotar, Omar Cortes, Byron David, Chelsea Finn, Chuyuan Fu, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Daniel Ho, Jasmine Hsu, Julian Ibarz, Brian Ichter, Alex Irpan, Eric Jang, Rosario Jauregui Ruano, Kyle Jeffrey, Sally Jesmonth, Nikhil J Joshi, Ryan Julian, Dmitry Kalashnikov, Yuheng Kuang, Kuang-Huei Lee, Sergey Levine, Yao Lu, Linda
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
77
Joshi, Ryan Julian, Dmitry Kalashnikov, Yuheng Kuang, Kuang-Huei Lee, Sergey Levine, Yao Lu, Linda Luu, Carolina Parada, Peter Pastor, Jornell Quiambao, Kanishka Rao, Jarek Rettinghouse, Diego Reyes, Pierre Sermanet, Nicolas Sievers, Clayton Tan, Alexander Toshev, Vincent Vanhoucke, Fei Xia, Ted Xiao, Peng Xu, Sichun Xu, Mengyuan Yan, and Andy Zeng. Do as I Can, not as I Say: Grounding Language in Robotic Affordances. arXiv preprint arXiv:2204.01691 , 2022. Aida Amini, Saadia Gabriel, Shanchuan Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh Hajishirzi. MathQA: Towards Interpretable Math Word Problem Solving with Operation-Based Formalisms. In ACL, 2019. BIG-bench Collaboration. Beyond the Imitation Game: Quantifying and Extrapolating the Capabilities of Language Models. arXiv preprint arXiv:2206.04615 , 2022. 17 Marcel Binz and Eric Schulz. Using Cognitive Psychology to Understand GPT-3. arXiv preprint
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
78
17 Marcel Binz and Eric Schulz. Using Cognitive Psychology to Understand GPT-3. arXiv preprint arXiv:2206.14576 , 2022. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-V oss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, and Dario Amodei. Language Models are Few-Shot Learners. In NeurIPS , 2020. Nadia Burkart and Marco F Huber. A Survey on the Explainability of Supervised Machine Learning. JAIR , 2021. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
79
2021. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, Nick Ryder, Mikhail Pavlov, Alethea Power, Lukasz Kaiser, Mohammad Bavarian, Clemens Winter, Philippe Tillet, Felipe Petroski Such, Dave Cummings, Matthias Plappert, Fotios Chantzis, Elizabeth Barnes, Ariel Herbert-V oss, William Hebgen Guss, Alex Nichol, Alex Paino, Nikolas Tezak, Jie Tang, Igor Babuschkin, Suchir Balaji, Shantanu Jain, William Saunders, Christopher Hesse, Andrew N. Carr, Jan Leike, Josh Achiam, Vedant Misra, Evan Morikawa, Alec Radford, Matthew Knight, Miles Brundage, Mira Murati,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
80
Achiam, Vedant Misra, Evan Morikawa, Alec Radford, Matthew Knight, Miles Brundage, Mira Murati, Katie Mayer, Peter Welinder, Bob McGrew, Dario Amodei, Sam McCandlish, Ilya Sutskever, and Wojciech Zaremba. Evaluating Large Language Models Trained on Code. arXiv preprint arXiv:2107.03374 , 2021a. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, Nick Ryder, Mikhail Pavlov, Alethea Power, Lukasz Kaiser, Mohammad Bavarian, Clemens Winter, Philippe Tillet, Felipe Petroski Such, Dave Cummings, Matthias Plappert, Fotios Chantzis, Elizabeth Barnes, Ariel Herbert-V oss, William Hebgen Guss, Alex Nichol, Alex Paino, Nikolas Tezak, Jie Tang, Igor Babuschkin,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
81
Herbert-V oss, William Hebgen Guss, Alex Nichol, Alex Paino, Nikolas Tezak, Jie Tang, Igor Babuschkin, Suchir Balaji, Shantanu Jain, William Saunders, Christopher Hesse, Andrew N. Carr, Jan Leike, Josh Achiam, Vedant Misra, Evan Morikawa, Alec Radford, Matthew Knight, Miles Brundage, Mira Murati, Katie Mayer, Peter Welinder, Bob McGrew, Dario Amodei, Sam McCandlish, Ilya Sutskever, and Wojciech Zaremba. Evaluating Large Language Models Trained on Code. arXiv:2107.03374 [cs] , 2021b. arXiv: 2107.03374. Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, Kensen Shi, Sasha
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
82
Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, Kensen Shi, Sasha Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker Barnes, Yi Tay, Noam Shazeer, Vinodkumar Prabhakaran, Emily Reif, Nan Du, Ben Hutchinson, Reiner Pope, James Bradbury, Jacob Austin, Michael Isard, Guy Gur-Ari, Pengcheng Yin, Toju Duke, Anselm Levskaya, Sanjay Ghemawat, Sunipa Dev, Henryk Michalewski, Xavier Garcia, Vedant Misra, Kevin Robinson, Liam Fedus, Denny Zhou, Daphne Ippolito, David Luan, Hyeontaek Lim, Barret Zoph, Alexander Spiridonov, Ryan Sepassi, David Dohan, Shivani Agrawal, Mark Omernick, Andrew M. Dai, Thanumalayan Sankaranarayana Pillai, Marie Pellat, Aitor Lewkowycz, Erica Moreira, Rewon Child, Oleksandr Polozov, Katherine Lee, Zongwei Zhou, Xuezhi
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
83
Lewkowycz, Erica Moreira, Rewon Child, Oleksandr Polozov, Katherine Lee, Zongwei Zhou, Xuezhi Wang, Brennan Saeta, Mark Diaz, Orhan Firat, Michele Catasta, Jason Wei, Kathy Meier-Hellstern, Douglas Eck, Jeff Dean, Slav Petrov, and Noah Fiedel. PaLM: Scaling Language Modeling with Pathways. arXiv preprint arXiv:2204.02311 , 2022. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. Training Verifiers to Solve Math Word Problems. arXiv preprint arXiv:2110.14168 , 2021. Jacob Cohen. A Coefficient of Agreement for Nominal Scales. Educational and psychological measurement , 1960. 18 Elaine Espindola. A Systemic Functional Translation Analysis of Thematic Structure: Directing Attention to Yoda’s Linguistic Manifestation. Word , 2016. Amir Feder, Katherine A. Keith, Emaad Manzoor, Reid Pryzant, Dhanya Sridhar, Zach Wood-Doughty,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
84
Amir Feder, Katherine A. Keith, Emaad Manzoor, Reid Pryzant, Dhanya Sridhar, Zach Wood-Doughty, Jacob Eisenstein, Justin Grimmer, Roi Reichart, Margaret E. Roberts, Brandon M. Stewart, Victor Veitch, and Diyi Yang. Causal Inference in Natural Language Processing: Estimation, Prediction, Interpretation and Beyond. arXiv preprint arXiv:2109.00725 , 2021. Sebastian Gehrmann, Tosin Adewumi, Karmanya Aggarwal, Pawan Sasanka Ammanamanchi, Aremu Anuoluwapo, Antoine Bosselut, Khyathi Raghavi Chandu, Miruna Clinciu, Dipanjan Das, Kaustubh D. Dhole, Wanyu Du, Esin Durmus, Ond ˇrej Dušek, Chris Emezue, Varun Gangal, Cristina Garbacea, Tatsunori Hashimoto, Yufang Hou, Yacine Jernite, Harsh Jhamtani, Yangfeng Ji, Shailza Jolly, Mihir Kale,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
85
Dhruv Kumar, Faisal Ladhak, Aman Madaan, Mounica Maddela, Khyati Mahajan, Saad Mahamood, Bodhisattwa Prasad Majumder, Pedro Henrique Martins, Angelina McMillan-Major, Simon Mille, Emiel van Miltenburg, Moin Nadeem, Shashi Narayan, Vitaly Nikolaev, Rubungo Andre Niyongabo, Salomey Osei, Ankur Parikh, Laura Perez-Beltrachini, Niranjan Ramesh Rao, Vikas Raunak, Juan Diego Rodriguez, Sashank Santhanam, João Sedoc, Thibault Sellam, Samira Shaikh, Anastasia Shimorina, Marco Antonio Sobrevilla Cabezudo, Hendrik Strobelt, Nishant Subramani, Wei Xu, Diyi Yang, Akhila Yerukola, and Jiawei Zhou. The GEM Benchmark: Natural Language Generation, its Evaluation and Metrics. arXiv preprint arXiv:2102.01672 , 2021. Yash Goyal, Ziyan Wu, Jan Ernst, Dhruv Batra, Devi Parikh, and Stefan Lee. Counterfactual Visual Explanations. In ICML , 2019.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
86
Yash Goyal, Ziyan Wu, Jan Ernst, Dhruv Batra, Devi Parikh, and Stefan Lee. Counterfactual Visual Explanations. In ICML , 2019. Yuling Gu, Bhavana Dalvi, and Peter Clark. DREAM: Improving Situational QA by First Elaborating the Situation. In NAACL , 2022. Simon Jerome Han, Keith Ransom, Andrew Perfors, and Charles Kemp. Human-like Property Induction is a Challenge for Large Language Models. PsyArXiv , 2022. Micha Heilbron, Kristijan Armeni, Jan-Mathijs Schoffelen, Peter Hagoort, and Floris P De Lange. A Hierarchy of Linguistic Predictions During Natural Language Comprehension. Proceedings of the National Academy of Sciences , 2022. Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. The Curious Case of Neural Text Degeneration. In ICLR , 2019. Curtis Honeycutt. Correct Grammar, Yoda’s Speech Is? Correct Grammar, Yoda’s Speech Is?, 2019. Accessed: 2022-08-15. IMDB. Star Wars: Episode V - The Empire Strikes Back. Star Wars: Episode V - The Empire Strikes Back,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
87
Accessed: 2022-08-15. IMDB. Star Wars: Episode V - The Empire Strikes Back. Star Wars: Episode V - The Empire Strikes Back, 1980. Accessed: 2022-08-15. Alon Jacovi and Yoav Goldberg. Towards Faithfully Interpretable NLP Systems: How Should We Define and Evaluate Faithfulness? In ACL, 2020. Sarthak Jain and Byron C. Wallace. Attention is not Explanation. In NAACL , 2019. Norman P. Jouppi, Cliff Young, Nishant Patil, David Patterson, Gaurav Agrawal, Raminder Bajwa, Sarah Bates, Suresh Bhatia, Nan Boden, Al Borchers, Rick Boyle, Pierre-luc Cantin, Clifford Chao, Chris Clark, Jeremy Coriell, Mike Daley, Matt Dau, Jeffrey Dean, Ben Gelb, Tara Vazir Ghaemmaghami, Rajendra Gottipati, William Gulland, Robert Hagmann, C. Richard Ho, Doug Hogberg, John Hu, Robert Hundt, Dan Hurt, Julian Ibarz, Aaron Jaffey, Alek Jaworski, Alexander Kaplan, Harshit Khaitan, Daniel
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
88
Hundt, Dan Hurt, Julian Ibarz, Aaron Jaffey, Alek Jaworski, Alexander Kaplan, Harshit Khaitan, Daniel Killebrew, Andy Koch, Naveen Kumar, Steve Lacy, James Laudon, James Law, Diemthu Le, Chris Leary, Zhuyuan Liu, Kyle Lucke, Alan Lundin, Gordon MacKean, Adriana Maggiore, Maire Mahony, Kieran Miller, Rahul Nagarajan, Ravi Narayanaswami, Ray Ni, Kathy Nix, Thomas Norrie, Mark Omernick, Narayana Penukonda, Andy Phelps, Jonathan Ross, Matt Ross, Amir Salek, Emad Samadiani, Chris Severn, Gregory Sizikov, Matthew Snelham, Jed Souter, Dan Steinberg, Andy Swing, Mercedes Tan, Gregory Thorson, Bo Tian, Horia Toma, Erick Tuttle, Vijay Vasudevan, Richard Walter, Walter Wang, Eric Wilcox, and Doe Hyun Yoon. In-Datacenter Performance Analysis of a Tensor Processing Unit. In ISCA , 2017. Norman P. Jouppi, Doe Hyun Yoon, Matthew Ashcraft, Mark Gottscho, Thomas B. Jablin, George Kurian,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
89
Norman P. Jouppi, Doe Hyun Yoon, Matthew Ashcraft, Mark Gottscho, Thomas B. Jablin, George Kurian, James Laudon, Sheng Li, Peter Ma, Xiaoyu Ma, Thomas Norrie, Nishant Patil, Sushma Prasad, Cliff 19 Young, Zongwei Zhou, and David Patterson. Ten Lessons from Three Generations Shaped Google’s TPUv4i: Industrial Product. In ISCA , 2021. Michael Kaminski. Yoda-Speak: A Study of Yoda’s Speaking Pattern and Their Frequencies. The Secret History of Star Wars , 2011. Nora Kassner, Oyvind Tafjord, Hinrich Schütze, and Peter Clark. BeliefBank: Adding Memory to a PreTrained Language Model for a Systematic Notion of Belief. In EMNLP , 2021. Harmanpreet Kaur, Harsha Nori, Samuel Jenkins, Rich Caruana, Hanna M. Wallach, and Jennifer Wortman Vaughan. Interpreting Interpretability: Understanding Data Scientists’ Use of Interpretability Tools for Machine Learning. In CHI, 2020.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
90
Vaughan. Interpreting Interpretability: Understanding Data Scientists’ Use of Interpretability Tools for Machine Learning. In CHI, 2020. Junyeob Kim, Hyuhng Joon Kim, Hyunsoo Cho, Hwiyeol Jo, Sang-Woo Lee, Sang-goo Lee, Kang Min Yoo, and Taeuk Kim. Ground-Truth Labels Matter: A Deeper Look into Input-Label Demonstrations. arXiv preprint arXiv:2205.12685 , 2022. Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large Language Models are Zero-Shot Reasoners. arXiv preprint arXiv:2205.11916 , 2022. Taku Kudo and John Richardson. SentencePiece: A Simple and Language Independent Subword Tokenizer and Detokenizer for Neural Text Processing. In EMNLP-Demo Track , 2018. Brenden M Lake, Tomer D Ullman, Joshua B Tenenbaum, and Samuel J Gershman. Building Machines that Learn and Think Like People. Behavioral and brain sciences , 2017. Thibault Laugel, Marie-Jeanne Lesot, Christophe Marsala, Xavier Renard, and Marcin Detyniecki. The
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
91
Thibault Laugel, Marie-Jeanne Lesot, Christophe Marsala, Xavier Renard, and Marcin Detyniecki. The Dangers of Post-hoc Interpretability: Unjustified Counterfactual Explanations. In IJCAI , 2019. Teven Le Scao and Alexander M Rush. How Many Data Points is a Prompt Worth? In NAACL , 2021. Aitor Lewkowycz, Anders Andreassen, David Dohan, Ethan Dyer, Henryk Michalewski, Vinay Ramasesh, Ambrose Slone, Cem Anil, Imanol Schlag, Theo Gutman-Solo, Yuhuai Wu, Behnam Neyshabur, Guy Gur-Ari, and Vedant Misra. Solving Quantitative Reasoning Problems with Language Models. arXiv preprint arXiv:2206.14858 , 2022. Yujia Li, David Choi, Junyoung Chung, Nate Kushman, Julian Schrittwieser, Rémi Leblond, Tom Eccles, James Keeling, Felix Gimeno, Agustin Dal Lago, Thomas Hubert, Peter Choy, Cyprien de Masson d’Autume, Igor Babuschkin, Xinyun Chen, Po-Sen Huang, Johannes Welbl, Sven Gowal, Alexey
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
92
Cherepanov, James Molloy, Daniel J. Mankowitz, Esme Sutherland Robson, Pushmeet Kohli, Nando de Freitas, Koray Kavukcuoglu, and Oriol Vinyals. Competition-Level Code Generation with AlphaCode. arXiv preprint arXiv:2203.07814 , 2022. Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. Program Induction by Rationale Generation: Learning to Solve and Explain Algebraic Word Problems. arXiv preprint arXiv:1705.04146 , 2017. Jiachang Liu, Dinghan Shen, Yizhe Zhang, Bill Dolan, Lawrence Carin, and Weizhu Chen. What Makes Good In-Context Examples for GPT-3? arXiv:2101.06804 [cs] , 2021a. arXiv: 2101.06804. Pengfei Liu, Jinlan Fu, Yang Xiao, Weizhe Yuan, Shuaichen Chang, Junqi Dai, Yixin Liu, Zihuiwen Ye, and Graham Neubig. ExplainaBoard: An Explainable Leaderboard for NLP. In IJCNLP , 2021b.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
93
Graham Neubig. ExplainaBoard: An Explainable Leaderboard for NLP. In IJCNLP , 2021b. Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hiroaki Hayashi, and Graham Neubig. Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing. arXiv preprint arXiv:2107.13586 , 2021c. Yao Lu, Max Bartolo, Alastair Moore, Sebastian Riedel, and Pontus Stenetorp. Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity. In ACL, 2022. Scott M. Lundberg and Su-In Lee. A Unified Approach to Interpreting Model Predictions. In NeurIPS , 2017. Aman Madaan, Niket Tandon, Dheeraj Rajagopal, Peter Clark, Yiming Yang, and Eduard Hovy. Think about it! Improving Defeasible Reasoning by First Modeling the Question Scenario. In EMNLP , 2021. Quinn McNemar. Note on the Sampling Error of the Difference between Correlated Proportions or Percentages. Psychometrika , 1947.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
94
Quinn McNemar. Note on the Sampling Error of the Difference between Correlated Proportions or Percentages. Psychometrika , 1947. Sewon Min, Xinxi Lyu, Ari Holtzman, Mikel Artetxe, Mike Lewis, Hannaneh Hajishirzi, and Luke Zettlemoyer. Rethinking the Role of Demonstrations: What Makes In-Context Learning Work? arXiv preprint 20 arXiv:2202.12837 , 2022. Swaroop Mishra, Daniel Khashabi, Chitta Baral, Yejin Choi, and Hannaneh Hajishirzi. Reframing Instructional Prompts to GPTk’s Language. arXiv preprint arXiv:2109.07830 , 2021. Ramaravind K Mothilal, Amit Sharma, and Chenhao Tan. Explaining Machine Learning Classifiers Through Diverse Counterfactual Explanations. In FAT, 2020. Guoshun Nan, Jiaqi Zeng, Rui Qiao, Zhijiang Guo, and Wei Lu. Uncovering Main Causalities for Longtailed Information Extraction. In EMNLP , 2021. Maxwell Nye, Anders Johan Andreassen, Guy Gur-Ari, Henryk Michalewski, Jacob Austin, David Bieber,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
95
Maxwell Nye, Anders Johan Andreassen, Guy Gur-Ari, Henryk Michalewski, Jacob Austin, David Bieber, David Dohan, Aitor Lewkowycz, Maarten Bosma, David Luan, Charles Sutton, and Augustus Odena. Show your Work: Scratchpads for Intermediate Computation with Language Models. arXiv preprint arXiv:2112.00114 , 2021. Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. Training Language Models to Follow Instructions with Human Feedback. arXiv preprint arXiv:2203.02155 , 2022. Arkil Patel, Satwik Bhattamishra, and Navin Goyal. Are NLP Models Really Able to Solve Simple Math Word Problems? arXiv preprint arXiv:2103.07191 , 2021.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
96
Word Problems? arXiv preprint arXiv:2103.07191 , 2021. Gabriel Poesia, Alex Polozov, Vu Le, Ashish Tiwari, Gustavo Soares, Christopher Meek, and Sumit Gulwani. Synchromesh: Reliable Code Generation from Pre-trained Language Models. In ICLR , 2021. Rafael Poyiadzi, Kacper Sokol, Raul Santos-Rodriguez, Tijl De Bie, and Peter Flach. FACE: Feasible and Actionable Counterfactual Explanations. In AAAI , 2020. Danish Pruthi, Mansi Gupta, Bhuwan Dhingra, Graham Neubig, and Zachary C. Lipton. Learning to Deceive with Attention-Based Explanations. In ACL, 2020. Geoffrey K. Pullum. YODA’S Syntax the Tribune Analyzes; Supply more Details I Will! YODA’S Syntax the Tribune Analyzes; Supply more Details I Will!, 2005. Accessed: 2022-08-15. Nazneen Fatema Rajani, Bryan McCann, Caiming Xiong, and Richard Socher. Explain Yourself! Leveraging Language Models for Commonsense Reasoning. In ACL, 2019.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
97
Yasaman Razeghi, Robert L Logan IV , Matt Gardner, and Sameer Singh. Impact of Pretraining Term Frequencies on Few-shot Reasoning. arXiv preprint arXiv:2202.07206 , 2022. Emily Reif, Daphne Ippolito, Ann Yuan, Andy Coenen, Chris Callison-Burch, and Jason Wei. A Recipe for Arbitrary Text Style Transfer with Large Language Models. arXiv preprint arXiv:2109.03910 , 2021. Laria Reynolds and Kyle McDonell. Prompt Programming for Large Language Models: Beyond the Fewshot Paradigm. In Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems , 2021. Marco Túlio Ribeiro, Sameer Singh, and Carlos Guestrin. “Why Should I Trust You?”: Explaining the Predictions of Any Classifier. In SIGKDD , 2016. Ohad Rubin, Jonathan Herzig, and Jonathan Berant. Learning to Retrieve Prompts for In-context Learning. arXiv preprint arXiv:2112.08633 , 2021. Victor Sanh, Albert Webson, Colin Raffel, Stephen H. Bach, Lintang Sutawika, Zaid Alyafeai, Antoine
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
98
Victor Sanh, Albert Webson, Colin Raffel, Stephen H. Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Teven Le Scao, Arun Raja, Manan Dey, M Saiful Bari, Canwen Xu, Urmish Thakker, Shanya Sharma Sharma, Eliza Szczechla, Taewoon Kim, Gunjan Chhablani, Nihal Nayak, Debajyoti Datta, Jonathan Chang, Mike Tian-Jian Jiang, Han Wang, Matteo Manica, Sheng Shen, Zheng Xin Yong, Harshit Pandey, Rachel Bawden, Thomas Wang, Trishala Neeraj, Jos Rozen, Abheesht Sharma, Andrea Santilli, Thibault Fevry, Jason Alan Fries, Ryan Teehan, Stella Biderman, Leo Gao, Tali Bers, Thomas Wolf, and Alexander M. Rush. Multitask Prompted Training Enables Zero-Shot Task Generalization, 2021. Martin Schrimpf, Idan Asher Blank, Greta Tuckute, Carina Kauf, Eghbal A Hosseini, Nancy Kanwisher,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
99
Martin Schrimpf, Idan Asher Blank, Greta Tuckute, Carina Kauf, Eghbal A Hosseini, Nancy Kanwisher, Joshua B Tenenbaum, and Evelina Fedorenko. The Neural Architecture of Language: Integrative Modeling Converges on Predictive Processing. Proceedings of the National Academy of Sciences , 2021. 21 Vered Shwartz, Peter West, Ronan Le Bras, Chandra Bhagavatula, and Yejin Choi. Unsupervised Commonsense Question Answering with Self-Talk. In EMNLP , 2020. Dylan Slack, Anna Hilgard, Himabindu Lakkaraju, and Sameer Singh. Counterfactual Explanations can be Manipulated. NeurIPS , 2021. Ilia Stepin, Jose M Alonso, Alejandro Catala, and Martín Pereira-Fariña. A Survey of Contrastive and Counterfactual Explanation Generation Methods for Explainable Artificial Intelligence. IEEE Access , 2021. Emma Strubell, Ananya Ganesh, and Andrew McCallum. Energy and Policy Considerations for Deep Learning in NLP. In ACL, 2019. Kai Sun, Dian Yu, Dong Yu, and Claire Cardie. Improving Machine Reading Comprehension with General Reading Strategies. In NAACL , 2019.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
100
Kai Sun, Dian Yu, Dong Yu, and Claire Cardie. Improving Machine Reading Comprehension with General Reading Strategies. In NAACL , 2019. Yujin Tang and David Ha. The Sensory Neuron As a Transformer: Permutation-invariant Neural Networks for Reinforcement Learning. NeurIPS , 2021. Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, YaGuang Li, Hongrae Lee, Huaixiu Steven Zheng, Amin Ghafouri, Marcelo Menegali, Yanping Huang, Maxim Krikun, Dmitry Lepikhin, James Qin, Dehao Chen, Yuanzhong Xu, Zhifeng Chen, Adam Roberts, Maarten Bosma, Vincent Zhao, Yanqi Zhou, Chung-Ching Chang, Igor Krivokon, Will Rusch, Marc Pickett, Pranesh Srinivasan, Laichee Man, Kathleen MeierHellstern, Meredith Ringel Morris, Tulsee Doshi, Renelito Delos Santos, Toju Duke, Johnny Soraker, Ben Zevenbergen, Vinodkumar Prabhakaran, Mark Diaz, Ben Hutchinson, Kristen Olson, Alejandra Molina,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
101
Zevenbergen, Vinodkumar Prabhakaran, Mark Diaz, Ben Hutchinson, Kristen Olson, Alejandra Molina, Erin Hoffman-John, Josh Lee, Lora Aroyo, Ravi Rajakumar, Alena Butryna, Matthew Lamm, Viktoriya Kuzmina, Joe Fenton, Aaron Cohen, Rachel Bernstein, Ray Kurzweil, Blaise Aguera-Arcas, Claire Cui, Marian Croak, Ed Chi, and Quoc Le. LaMDA: Language Models for Dialog Applications. arXiv preprint arXiv:2201.08239 , 2022. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is All you Need. In NeurIPS , 2017. Sahil Verma, John Dickerson, and Keegan Hines. Counterfactual Explanations for Machine Learning: A Review. arXiv preprint arXiv:2010.10596 , 2020. George Walkden. English VP-fronting and the Syntax of Yoda. talk given at LinguistMix , 2012.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
102
George Walkden. English VP-fronting and the Syntax of Yoda. talk given at LinguistMix , 2012. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, and Denny Zhou. Rationale-Augmented Ensembles in Language Models. arXiv preprints arXiv:2207.00747 , 2022a. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, and Denny Zhou. Self-Consistency Improves Chain of Thought Reasoning in Language Models. arXiv preprint arXiv:2203.11171 , 2022b. Yaqing Wang, Quanming Yao, James T Kwok, and Lionel M Ni. Generalizing from a few Examples: A Survey on Few-shot Learning. ACM Computing Surveys , 2020. Jason Wei, Maarten Bosma, Vincent Y Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M Dai, and Quoc V Le. Finetuned Language Models are Zero-shot Learners. arXiv preprint arXiv:2109.01652 , 2021. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed Chi, Quoc Le, and Denny Zhou. Chain
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
103
Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed Chi, Quoc Le, and Denny Zhou. Chain of Thought Prompting Elicits Reasoning in Large Language Models. arXiv preprint arXiv:2201.11903 , 2022. James CR Whittington, Joseph Warren, and Tim EJ Behrens. Relating Transformers to Models and Neural Representations of the Hippocampal Formation. In ICLR , 2021. Sarah Wiegreffe and Yuval Pinter. Attention is not not Explanation. In EMNLP-IJCNLP , 2019. Wookieepedia. Yoda. Yoda, 2022. Accessed: 2022-08-15. Yuhuai Wu, Albert Q Jiang, Wenda Li, Markus N Rabe, Charles Staats, Mateja Jamnik, and Christian Szegedy. Autoformalization with Large Language Models. arXiv preprint arXiv:2205.12615 , 2022. Sang Michael Xie, Aditi Raghunathan, Percy Liang, and Tengyu Ma. An Explanation of In-context Learning as Implicit Bayesian Inference. In ICLR , 2021.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
104
as Implicit Bayesian Inference. In ICLR , 2021. Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, 22 Olivier Bousquet, Quoc Le, and Ed Chi. Least-to-Most Prompting Enables Complex Reasoning in Large Language Models. arXiv preprint arXiv:2205.10625 , 2022. 23 Appendix Table of Contents A Extended Background 25 A.1 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 B Details on Studied Reasoning Tasks 26 C Computational Resources and Models 26 D Attention Analysis 26 D.1 Per-layer Attention Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 D.2 Specialized Attention Heads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
105
E Results on C ODEX , GPT-3, PaLM-540 Band Statistical Significance Test 33 E.1 Results on C ODEX , GPT-3, PaLM-540 B. . . . . . . . . . . . . . . . . . . . . . . . . . 33 E.2 Significance tests for PaLM-62 B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 F Additional Experiments on Role of Symbols 38 F.1 Counterfactual Prompts for Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 F.2 Role of symbols: Output Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 F.3 Attention Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 F.4 Role of Dataset for Effective Chain of Thought . . . . . . . . . . . . . . . . . . . . . . . 41
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
106
F.4 Role of Dataset for Effective Chain of Thought . . . . . . . . . . . . . . . . . . . . . . . 41 F.5 Additional Experiments for S PORTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 G Additional Experiments on Role of Patterns in CoT 45 G.1 Output Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 H Additional Experiments on Role of Text 51 H.1 Text with Altered Grammatical Style . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 I Symbiosis between Text and Patterns: Additional Examples and Qualitative Analysis 61 I.1 Constructing Effective Intermediate Thoughts . . . . . . . . . . . . . . . . . . . . . . . . 61 I.2 Commonsense Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
107
I.3 Text and Patterns: It Takes Two to Tango . . . . . . . . . . . . . . . . . . . . . . . . . . 63 J Additional Results CC OT: C oncise C hain O f Thought 63 J.1 CC OT Prompts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 J.2 Reduction in Length by CC OT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 K Complete List of Counterfactual Prompts 68 24 A. E XTENDED BACKGROUND Background on chain of thought . This work broadly investigates the premise of in-context few-shot prompting in large language models (LLM). In these methods, the input to the model is a prompt pconsisting ofkin-context examples in the form of xinputÞÑxi, outputÞÑyiytuples6. Eachxxi,yiyalludes to the target task. For example, in math solving problems Cobbe et al. (2021), an input is math question ( If
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
108
to the target task. For example, in math solving problems Cobbe et al. (2021), an input is math question ( If three apples were added to a basket which had two apples, how many apples are in the basket now? ), and the output supplies the answer ( 5). Wei et al. (2022) additionally supplement in-context few-shot prompting with Chain Of Thought (COT) method, improving the performance of LLM in solving several reasoning tasks. In particular, C OT additionally prefixes each output with a thought , creating triplets xxi,ti,yiy. The “chain of thought” tidescribes the intermediate steps and/or results required to derive the output yifrom xi. Therefore, the prompt is assembled in the form of pxx1t1y1y}xx2t2y2y}:::}xxktkyky, where “” and “}” are indicator symbols. The role of is to separate elements of an example, whereas } indicates the boundary of an example. The intuition behind chain of thought prompting is that catering the outputs/answers with intermediate steps/results present additional in-context information to the model (Ling
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
109
outputs/answers with intermediate steps/results present additional in-context information to the model (Ling et al., 2017; Amini et al., 2019; Chen et al., 2021b; Cobbe et al., 2021; Nye et al., 2021). This additional in-context information presumably improves accuracy in solving various reasoning tasks. At inference time, C OT appends an unseen question ^xto the prompt pand supplies the extended prompt to a LLM. The model completes the prompt to generate a relevant thought ^tandan answer ^y. To assess the performance of LLM, C OTonly compares the post-processed generated answer with the ground truth. Gauging the correctness of the generated thought ^tis not straightforward because ground truth thoughts are unavailable. Nonetheless, the generated thought can be further analyzed to infer the possible mechanisms, allowing an analogy with the human thought process, with which the model attains the answer. Background on counterfactual explanation. Counterfactual explanations seek to explain the behavior of a model by conducting “ what if ” analysis on examples for which the expected outputs of the model is known (Mothilal et al., 2020; Stepin et al., 2021; Verma et al., 2020; Poyiadzi et al., 2020; Goyal et al.,
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
110
known (Mothilal et al., 2020; Stepin et al., 2021; Verma et al., 2020; Poyiadzi et al., 2020; Goyal et al., 2019; Feder et al., 2021). Specifically, let px;yqbe a tuple where xis the input to a model Mthat estimates an output distribution pp |xq, andypp |xq. Counterfactual explanations utilize variants Cfpx;b;aq of the inputs that differ from the original input xin all except one feature f. Here,bandadenote the before and after values of the feature finx. For instance, consider an image xof a camel with a brown background labeled correctly by a classifier. A counterfactual Cbgpx;brown;green qexample is an identical image with only a different background color, green, in this example. By virtue of comparing pp|xqwith pp|Cbgpx;brown;green qqfor a sufficiently large sample of images, one may infer certain facts about the classifier, for example its reliance on the background color. A.1. L IMITATIONS
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
111
classifier, for example its reliance on the background color. A.1. L IMITATIONS While the counterfactual approach has its advantages, its assumptions induce limitations summarized below. There is potential for uncharted and baffling artifacts that the model could be exploiting, leading to potentially misleading observations. For instance, symbols, patterns, text and the outcome may be spuriously correlated. A classic example of such spurious correlation is confounding: the phenomenon of an observed correlation between X and Y induced by their common cause Z. Despite taking methodical and scientific measures, to the best of our knowledge, to ground our hypothesis in in-depth empirical analysis, the discrete and multiplicative nature of language understanding tasks implies that no such study can be completely thorough. In addition, this work is limited to a subset of common tasks and datasets, including math (Cobbe et al., 2021), commonsense reasoning (BIG-bench Collaboration, 2022), and symbolic reasoning. Our conclusions in this work may not apply to other reasoning and question-answering tasks, or even 6The number of xxi,yiytuples depend on the maximum input sequence length of the model, typically k¤10. 25 Table 9: Examples of “ what if ” questions that we seek to answer in this work.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
112
25 Table 9: Examples of “ what if ” questions that we seek to answer in this work. Q1.What if we replace all the symbols in the prompt with abstract placeholders, can the required task still be discerned? Q2. What if the examples in the prompt were incorrect, will it affect the correctness of the outputs? Q3.What if we remove all patterns from the input, will C OT continue to be effective? Q4. What if the linguistic style of the prompt was different than that of the questions, will it hamper the performance? to other datasets from the same task category. We experiment with large language models that are either not publicly available (e.g., PaLM) or are either available using API calls. Existing documentation (Brown et al., 2020; Chen et al., 2021a) does not fully describe the details of these models, such as the pretraining corpus, model size, and model biases. B. D ETAILS ON STUDIED REASONING TASKS In this work, we evaluate counterfactual prompting on the following reasoning tasks: 1.MATHEMATICAL We experiment with GSM-8 K(Cobbe et al., 2021) (1319 samples). The dataset contains math word problems geared toward an average middle-school curriculum.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
113
contains math word problems geared toward an average middle-school curriculum. 2.COMMONSENSE We use date understanding (D ATE, 349 samples) and sports understanding (S PORTS , 980 samples) as representative tasks for commonsense reasoning, both derived from BIGbench Collaboration (2022). 3.SYMBOLIC We experiment with sorting (S ORTING , 500 samples) a list of single-digit integers. We do not associate explicit instruction (e.g., sort these numbers ) with the questions. Instead, we frame the questions as a challenging setup in which the model should figure out the task and the requisite information to solve it. C. C OMPUTATIONAL RESOURCES AND MODELS In this work, we neither train any of the PaLM models, nor performs finetuning. We solely perform inference on PaLM variants using TPU v4 (Jouppi et al., 2021; 2017). For PaLM-62 B, we use 444 TPU v4 configuration, whereas, for PaLM-540 Bwe use 4416 mesh configuration. To account for the variation in results caused by the order of examples in the prompt, we conduct each experiment three times, each with
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
114
in results caused by the order of examples in the prompt, we conduct each experiment three times, each with different seeds, and report the average task solve rate. Following Wei et al. (2022), we evaluate each task using accuracy i.e. fraction of examples where the output matched the expected result. Public large language models. We use OpenAI API7to conduct experiments with GPT-3 ( text-davinci002) and C ODEX (code-davinci-002 ). D. A TTENTION ANALYSIS Attention mechanism (Vaswani et al., 2017) is now ubiquitous in NLP. While the broader question on the utility of attention for posthoc analysis is still open (Jain & Wallace, 2019; Pruthi et al., 2020), there is some evidence to show that attention can act as an explanation (Wiegreffe & Pinter, 2019). Finally, the utility of 7https://openai.com/blog/openai-api/, v0.23.0 26 Table 10: Examples of tasks used in this work. The åQÞÑques tion, TÞÑthought, and VAÞÑ answer are separately highlighted. For the complete list of vanilla C OT for each category refer to Table 38, Table 40, Table 39, and Table 41, respectively.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
115
answer are separately highlighted. For the complete list of vanilla C OT for each category refer to Table 38, Table 40, Table 39, and Table 41, respectively. MATHEMATICAL  åQÞÑShawn hasfivetoys. ForChrist mas, hegottwotoys each from hismom anddad. How many toys does hehave now? TÞÑShawn started with 5toys. Ifhegot2toys each from hismom anddad, then thatis4more toys. 5 +4=9. VAÞÑTheanswer is9. COMMONSENSE (SPORTS ) åQÞÑIsthefollowingsentence plausible? “JamalMurraywasperfectfrom theline.” TÞÑJamalMurrayisabasketballplayer. Beingperfectfrom thelineispartofbasketball. VAÞÑTheanswer isyes. COMMONSENSE (DATE) åQÞÑItis4/19/1969 today. What isthedate 24hours later inMM/D D/YYYY? TÞÑTodayis04/19/1969. 24hours later isonedayaftertoday, which would be04/20/1969.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
116
TÞÑTodayis04/19/1969. 24hours later isonedayaftertoday, which would be04/20/1969. VAÞÑTheanswer is04/20/1969. SYMBOLIC (SORTING ) åQÞÑ2,4,3,8,9,6,7,1. TÞÑ1<2<3<4<5<6<7<8<9. VAÞÑTheanswer is1,2,3,4,5,6,7,8,9 any explanation mechanism is closely tied to the users and application domain (Kaur et al., 2020; Burkart & Huber, 2021), and as our analysis shows, attention provides concurring evidence that adds intuition and insights to the empirical findings of this work. Note that while we conduct empirical experiments with PaLM, GPT-3, and C ODEX , we only conduct attention-related ablations with PaLM as the GPT-3 and CODEX were only available to us via API. Attention for autoregressive models. Consider a sentence: my dog loved the toy . Modern NLP methods divide each sentence into tokens, a decision dictated by the underlying tokenization library. PaLM uses
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
117
divide each sentence into tokens, a decision dictated by the underlying tokenization library. PaLM uses SentencePiece (Kudo & Richardson, 2018) for tokenization. For simplicity, we assume a tokenizer that divides the sentence into tokens based on the whitespace. This yields the following list of tokens: [my, dog, loves, treats]. Let BOS be a special beginning of sequence token present in all sentences, and pbe a language model with the parameters . Decoder-only language models such as PaLM estimate the likelihood of a sequence such asmy dog loved the toy using an autoregressive factorization popularly known as the chain-rule: ppBOS;my;dog;loves;treats qppmy|BOSqppdog|BOS;myqpploves|BOS;my;dogq pptreats|BOS;my;dog;loves q Estimating these conditional probabilities (e.g., ppmy|BOSq) requires a stack of transformer layers, each containing an attention module. Thus, this factorization also implies that tokens attend to the left (Figure 4), with a token wiat locationiattending to all tokens w i.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
118
containing an attention module. Thus, this factorization also implies that tokens attend to the left (Figure 4), with a token wiat locationiattending to all tokens w i. Letwsbe the source token (current input to the model). The set of target tokens, or tokens that wswill attend to, thus are: w0;w1;:::;ws1. PaLM-62 Bhas 64 layers, each containing the self-attention mechanism 27 Q1 T1 A1 Q2 T2 A2 Q’T’ A ’Figure 3: Structure of a typical chain of thought prompt. The prompt contains a handful of QTA examples, each containing three parts: 1) The question (Q), the thought (T) that spells out the reasoning process to derive the answer, and finally, 3) Answer (A) the final answer. In the Figure, the prompt contains two such QTA examples. During inference, a test question Q’ is appended to the prompt, and the model is expected to complete it by generating a thought T’ and the answer A’, presumably leveraging the two QTA examples in the input. with 32 heads. Focusing on a single layer and head, let astbe the attention score from wstowt, where°s1
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
119
with 32 heads. Focusing on a single layer and head, let astbe the attention score from wstowt, where°s1 t0ast1. Analyzing the important components of a C OT prompt. We leverage attention scores as an additional signal to help uncover the important components of a prompt. To this end, we calculate the attention scores from thesource tokens that are part of the Q’,T’, orA’tothetarget prompt question Qi, thought Ti, and answer Ai(Figure 3). Note that the same prompt is used for all the questions in the test set. Thus across questions, the set of target tokens remains the same. Our goal in attention analysis is to uncover important tokens and spans used by PaLM to solve a task. Since the distribution of attention scores asis typically long-tailed, recording the attention score between every pair of source-target tokens might lead to noise and spurious patterns (Nan et al., 2021). To remedy this, we take inspiration from nucleus sampling (Holtzman et al., 2019) and set all values below the kthlargest attention value to 0 (we use k10). LetQ1 jbe thejthquestion in the test set Qof the questions to be evaluated. Recall that the same prompt
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
120
attention value to 0 (we use k10). LetQ1 jbe thejthquestion in the test set Qof the questions to be evaluated. Recall that the same prompt is used for all the questions, and we calculate the attention scores from the source tokens (tokens in the inference question) to the target tokens (those in the prompt). Let astbe the attention from token wstowt. We calculate the attention importance Itof a tokenwtin the prompt as the average max attention it has received across the set Qof inference questions. It°|Q| j1max|Q1 j||A1 j||T1 j| s1ast |Q|(1) 1. The spectrum plots show a comparison of Itfor all tokens in the prompt for two different prompts: vanilla C OT prompt and Csymb_absppqprompt. 2. The pattern vs. text prompts group the target tokens by their type: the tokens that belong to a pattern vs. tokens belonging to the text. The attention importance values are then shown. 3. The bos by layer plots investigate the total attention importance for the BOS token. D.1. P ER-LAYER ATTENTION ANALYSIS
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
121
3. The bos by layer plots investigate the total attention importance for the BOS token. D.1. P ER-LAYER ATTENTION ANALYSIS The main draft provides spectrum plots averaged over heads and layers. Figure 11 shows the same question for three different datasets averaged across layers. Figure 5, Figure 6, and Figure 7 provide the same plots, per layer. We find that the spectrum of Isvalues is identical between C OT(p) andCsymb_absppqacross layers, showing that averaging is not leading to spurious correlations. 28 BOS My dog loves treats Figure 4: Auto-regressive language models: the tokens are generated as a sequence, with each token attending to the preceding tokens. 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 (a) Layer 0pvs.Csymb_absppq 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 (b) Layer 15 pvs.Csymb_absppq 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 (c) Layer 31 pvs.Csymb_absppq
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
122
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 (c) Layer 31 pvs.Csymb_absppq Figure 5: Average attention per token for a randomly sampled question using standard C OT promptp (above) and Csymb_absppqfor GSM-8 Kacross layers. Near identical attention pattern shows that few-shot models are relatively indifferent to the exact symbols, but are sensitive to patterns. D.2. S PECIALIZED ATTENTION HEADS Fine-tuned models can be expected to learn attention patterns that facilitate solving a task. Does the same hold for few-shot models? To our knowledge, the question of attention in a few-shot setup has not been explored. Surprisingly, we find that the model consistently uses certain heads and layers for attending over certain semantic parts of the inputs. We find such specialized head-layer pairs manually, and plot the average Itfor 100 questions for them in Figure 9. The Itvalues show a clear tendency for the head to favor either past tense ( would, yesterday ) or future tense ( will). Analyzing a large-language model’s attention patterns in detail is an interesting future work.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
123
past tense ( would, yesterday ) or future tense ( will). Analyzing a large-language model’s attention patterns in detail is an interesting future work. Symbiosis in attention scores We have explored different semantic components of prompts, namely patterns (including symbols) and text. A logical next question is whether patterns or text confer differential importance. While importance can be measured via various approaches, we use attention scores as a reasonable proxy. For GSM-8 K(where the distinction between patterns and text is clear), we calculate attention mass on patterns and text across several layers and average it over their attention heads. Figure 11 compares 29 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8(a) Layer 0pvs.Csymb_absppq 0.0 0.1 0.2 0.3 0.4 0.5 0.6 (b) Layer 15 pvs.Csymb_absppq 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 (c) Layer 31 pvs.Csymb_absppq Figure 6: Average attention per token for a randomly sampled question using standard C OT prompt
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
124
(c) Layer 31 pvs.Csymb_absppq Figure 6: Average attention per token for a randomly sampled question using standard C OT prompt p(above) and Csymb_absppqfor D ATEacross layers. Near identical attention pattern shows that few-shot models are relatively indifferent to the exact symbols, but are sensitive to patterns. 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 (a) Layer 0pvs.Csymb_absppq 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 (b) Layer 15 pvs.Csymb_absppq 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 (c) Layer 31 pvs.Csymb_absppq Figure 7: Average attention per token for a randomly sampled question using standard C OT promptp (above) and Csymb_absppqfor S PORTS across layers. Near identical attention pattern shows that few-shot models are relatively indifferent to the exact symbols, but are sensitive to patterns.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
125
models are relatively indifferent to the exact symbols, but are sensitive to patterns. these average scores, normalized between patterns and text. Our findings show that the model pays approximately equal attention to both, indicating similar importance. These results concur with our findings that text and patterns contribute equally to the success of C OT. 30 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8(a) Layer 0pvs.Csymb_absppq 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 (b) Layer 15 pvs.Csymb_absppq 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 (c) Layer 31 pvs.Csymb_absppq Figure 8: Average attention per token for a randomly sampled question using standard C OT promptp (above) and Csymb_absppqfor S ORTING across layers. Near identical attention pattern shows that few-shot models are relatively indifferent to the exact symbols, but are sensitive to patterns. Q : 2 0 1 5 is coming
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
126
Q : 2 0 1 5 is coming in 3 6 hours . What is the date one week from today in MM / DD / YYYY ? A : If 2 0 1 5 is coming in 3 6 hours , then it is coming in 2 days . 2 days before 0 1 / 0 1 / 2 0 1 5 is 1 2 / 3 0 / 2 0 2 1 , so
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
127
/ 3 0 / 2 0 2 1 , so today is 1 2 / 3 0 / 2 0 2 1 . So one week from today will be 0 1 / 0 5 / 2 0 1 5 . So the answer is 0 1 / 0 5 / 2 0 1 5 . Q : The first day of 2 0 1 9 is a Tuesday , and today is the first Monday
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
128
9 is a Tuesday , and today is the first Monday of 2 0 1 9 . What is the date today in MM / DD / YYYY ? A : If the first day of 2 0 1 9 was Tuesday , then 0 1 / 0 1 / 2 0 1 9 was a Tuesday . Today is the first monday , would be six days later . So today is 0 1 /
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
129
six days later . So today is 0 1 / 0 7 / 2 0 1 9 . So the answer is 0 1 / 0 7 / 2 0 1 9 . Q : The concert was scheduled to be on 0 6 / 0 1 / 1 9 4 3 , but was delayed by one day to today . What is the date 1 0 days ago in MM / DD /
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
130
date 1 0 days ago in MM / DD / YYYY ? A : One day after 0 6 / 0 1 / 1 9 4 3 is 0 6 / 0 2 / 1 9 4 3 , so today is 0 6 / 0 2 / 1 9 4 3 . 1 0 days before today is 0 5 / 2 3 / 1 9 4 3 . So the answer is 0 5
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
131
9 4 3 . So the answer is 0 5 / 2 3 / 1 9 4 3 . Q : It is 4 / 1 9 / 1 9 6 9 today . What is the date 2 4 hours later in MM / DD / YYYY ? A : Today is 0 4 / 1 9 / 1 9 6 9 . 2 4 hours later is one day after today , which would
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
132
hours later is one day after today , which would be 0 4 / 2 0 / 1 9 6 9 . So the answer is 0 4 / 2 0 / 1 9 6 9 . Q : Jane thought today is 3 / 1 1 / 2 0 0 2 , but today is in fact Mar 1 2 , which is 1 day later . What is the date 2 4 hours later
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
133
day later . What is the date 2 4 hours later in MM / DD / YYYY ? A : Today is 0 3 / 1 2 / 2 0 0 2 . So the date 2 4 hours later will be 0 3 / 1 3 / 2 0 0 2 . So the answer is 0 3 / 1 3 / 2 0 0 2 . Q : Jane was born on the last
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
134
2 . Q : Jane was born on the last day of Feb ur ary in 2 0 0 1 . Today is her 1 6 year old birthday . What is the date yesterday in MM / DD / YYYY ? A : The last day of February is the 2 8 th , so Jane was born on 0 2 / 2 8 / 2 0 0 1 . Today is her 1
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
135
/ 2 0 0 1 . Today is her 1 6 year old birthday , so today is 0 2 / 2 8 / 2 0 1 7 . So yesterday was 0 2 / 2 7 / 2 0 1 7 . So the answer is 0 2 / 2 7 / 2 0 1 7 . (a) Layer 45, Head 16 specializes in capturing tokens relevant to “past tense”. Q : 2 0 1 5 is coming in 3 6
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
136
: 2 0 1 5 is coming in 3 6 hours . What is the date one week from today in MM / DD / YYYY ? A : If 2 0 1 5 is coming in 3 6 hours , then it is coming in 2 days . 2 days before 0 1 / 0 1 / 2 0 1 5 is 1 2 / 3 0 / 2 0 2 1 , so today
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
137
3 0 / 2 0 2 1 , so today is 1 2 / 3 0 / 2 0 2 1 . So one week from today will be 0 1 / 0 5 / 2 0 1 5 . So the answer is 0 1 / 0 5 / 2 0 1 5 . Q : The first day of 2 0 1 9 is a Tuesday , and today is the first Monday of
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
138
is a Tuesday , and today is the first Monday of 2 0 1 9 . What is the date today in MM / DD / YYYY ? A : If the first day of 2 0 1 9 was Tuesday , then 0 1 / 0 1 / 2 0 1 9 was a Tuesday . Today is the first monday , would be six days later . So today is 0 1 / 0
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
139
days later . So today is 0 1 / 0 7 / 2 0 1 9 . So the answer is 0 1 / 0 7 / 2 0 1 9 . Q : The concert was scheduled to be on 0 6 / 0 1 / 1 9 4 3 , but was delayed by one day to today . What is the date 1 0 days ago in MM / DD / YYYY
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
140
1 0 days ago in MM / DD / YYYY ? A : One day after 0 6 / 0 1 / 1 9 4 3 is 0 6 / 0 2 / 1 9 4 3 , so today is 0 6 / 0 2 / 1 9 4 3 . 1 0 days before today is 0 5 / 2 3 / 1 9 4 3 . So the answer is 0 5
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
141
9 4 3 . So the answer is 0 5 / 2 3 / 1 9 4 3 . Q : It is 4 / 1 9 / 1 9 6 9 today . What is the date 2 4 hours later in MM / DD / YYYY ? A : Today is 0 4 / 1 9 / 1 9 6 9 . 2 4 hours later is one day after today , which would
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
142
hours later is one day after today , which would be 0 4 / 2 0 / 1 9 6 9 . So the answer is 0 4 / 2 0 / 1 9 6 9 . Q : Jane thought today is 3 / 1 1 / 2 0 0 2 , but today is in fact Mar 1 2 , which is 1 day later . What is the date 2 4 hours later
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
143
later . What is the date 2 4 hours later in MM / DD / YYYY ? A : Today is 0 3 / 1 2 / 2 0 0 2 . So the date 2 4 hours later will be 0 3 / 1 3 / 2 0 0 2 . So the answer is 0 3 / 1 3 / 2 0 0 2 . Q : Jane was born on the last day
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
144
. Q : Jane was born on the last day of Feb ur ary in 2 0 0 1 . Today is her 1 6 year old birthday . What is the date yesterday in MM / DD / YYYY ? A : The last day of February is the 2 8 th , so Jane was born on 0 2 / 2 8 / 2 0 0 1 . Today is her 1
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
145
/ 2 0 0 1 . Today is her 1 6 year old birthday , so today is 0 2 / 2 8 / 2 0 1 7 . So yesterday was 0 2 / 2 7 / 2 0 1 7 . So the answer is 0 2 / 2 7 / 2 0 1 7 . (b) Layer 45, Head 18 specializes in capturing tokens relevant to “future tense”. Figure 9: Specialized attention heads. 31 0 7 15 23 31
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
146
. (b) Layer 45, Head 18 specializes in capturing tokens relevant to “future tense”. Figure 9: Specialized attention heads. 31 0 7 15 23 31 Attention Layers0.00.10.20.30.40.50.6Normalized Attention MassFigure 10: Attention on the beginning of sequence token across layers. Averaged across attention heads, higher attention mass goes to the bos token for the higher layers. Note that PaLM is autoregressive, thus, higher attention mass on bos may be interpreted as the model paying equal attention to the rest of the sequence, indicating an evolving global representation.
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
147
to the rest of the sequence, indicating an evolving global representation. Figure 15-aPatternTextC/A00.563401310.436598691.2904328915874770.573124230.426875771.34260192374001150.561564260.438435741.28083595557242230.465478190.534521810.870831051776914310.642734620.357265381.7990397502271300.583253250.416746751.3995388086409870.586447510.413552491.41807273364501150.571411530.428588471.33324055591136230.452980080.547019920.828086991786332310.645248270.354751731.818872793093920.000.200.400.600.801.00 PatternTextNormalized Fraction of Attention MassCoTCoT-SymbolicLayer 0Layer 7Layer 15Layer 23Layer 31Layer 0Layer 7Layer 15Layer 23Layer 31 Figure 11: Visualizing the normalized fraction of attention mass between text and patterns across
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
148
Figure 11: Visualizing the normalized fraction of attention mass between text and patterns across multiple layers of the PaLM-62 Bmodel layers with the vanilla C OT (left-side) and symbolic C OT (rightside). In general, patterns receive slightly higher attention across most of the layers, with the topmost layer paying the largest attention. Strikingly, the attention patterns closely match for vanilla and symbolic C OT, implying that few-shot models leverage patterns to a larger extent. 32 E. R ESULTS ON CODEX , GPT-3, P ALM-540 B AND STATISTICAL SIGNIFICANCE TEST E.1. R ESULTS ON CODEX , GPT-3, P ALM-540 B We show results from four models: C ODEX (Chen et al., 2021a), GPT-3 (Brown et al., 2020), and two variants of PaLM (Chowdhery et al., 2022) (PaLM-62 Band PaLM-540 B). Note that we could not get results on all variations of prompts for GPT-3 because of usage limits by OpenAI. Such cases are indicated with a hyphen (-). Similarly, due to the rate limitations, we experimented with two seeds for all variations
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
149
with a hyphen (-). Similarly, due to the rate limitations, we experimented with two seeds for all variations on C ODEX and had to use a single seed for some variations. The findings are shown in Table 11 (GSM-8 K), Table 12 (D ATE), Table 13 (S PORTS ), and Table 14 (S ORTING ). We find that all the findings hold across models: correctness of patterns is immaterial, abstract and OOD symbols are still helpful, and the sensitivity to text is proportional to the degree of randomness. Finally, CC OT matches or outperforms C OT despite being 20% shorter. Table 11: All results for GSM-8 Kacross four models: C ODEX , GPT-3, PaLM-62 B, and PaLM-540 B. CODEX GPT-3 PaLM-62 B PaLM-540 B Prompt Avg. SD Avg. SD Avg. SD Avg. SD DIRECT 20.8% 1.1 16.1 % 0 10.1 % 0.3 9.6 % 3.2 COT(p) (Table 38) 65.6% 3 46.9 % 6.2 27.4 % 1.1 60.8 % 0.6
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
150
COT(p) (Table 38) 65.6% 3 46.9 % 6.2 27.4 % 1.1 60.8 % 0.6 CCOT 62.5 % 1.4 52.2 % 1.4 29.1 % 063.2% 1.1 Csymb_oodppq(Table 48) 66.2% 0.5 55.3 % 1.7 25.7 % 0.5 60.7 % 0.2 Csymb_absppq(Table 43) 56.5 % 6.4 49.4 % 0.1 28.2 % 0.2 59.0% 0.3 Cpat_wrongppq(Table 55) 65.5% 0.3 52.4 % 1.6 24.4 % 0.3 63.5 % 0.7 Cpat_noneppq(Table 57) 33.3 % 0.8 37.8 % 2.9 21.5 % 0.6 53.9% 1.5 Ctext_yodathoughtsppq(Table 70) 60.8% 0 - - 23.2 % 1 57.4 % 1.4 Ctext_intra_shufppq(Table 79) 33.0 % 1.8 - - 17.0 % 1.3 45.6% 3.8
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
151
Ctext_intra_shufppq(Table 79) 33.0 % 1.8 - - 17.0 % 1.3 45.6% 3.8 Ctext_inter_shufppq(Table 82) 29.7 % 6.3 - - 10.8 % 1.3 37.2% 3.1 Ctext_diff_entities ppq(Table 68) 59.0% 0 49.8 % 0 16.6 % 1 51.1 % 3.3 Table 12: Results for D ATEacross four models: C ODEX , GPT-3, PaLM-62 B, and PaLM-540 B. CODEX GPT-3 PaLM-62 B PaLM-540 B Prompt Avg. SD Avg. SD Avg. SD Avg. SD
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
152
CODEX GPT-3 PaLM-62 B PaLM-540 B Prompt Avg. SD Avg. SD Avg. SD Avg. SD DIRECT 51.3% 1 44.2 % 0.5 31.6 % 1.6 49.0 % COT(p) (Table 39) 69.2% 2 56.7 % 3.7 45.2 % 0.5 65.3 % CCOT 69.9% 5.7 61.2 % 1.4 51.3 % - 69.6 % Csymb_oodppq(Table 49) 67.8% 1.8 58.2 % 0 44.5 % 1.4 59.6 % Csymb_absppq(Table 42) 54.1 % 3.9 35 % 2 36.6 % 156.2% 0.8 Cpat_wrongppq(Table 54) 69.0% 1.9 58.4 % 1.6 42.9 % 3 67.2 % 0.6 Cpat_noneppq(Table 62) 62.0% 1.4 50.9 % 3.1 36.1 % 1.2 59.6 % 2 Ctext_yodathoughtsppq(Table 72) 55.0 % 2 - - 30.8 % 1.7 62.2% 1.6
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
153
Ctext_yodathoughtsppq(Table 72) 55.0 % 2 - - 30.8 % 1.7 62.2% 1.6 Ctext_intra_shufppq(Table 81) 44.3 % 3.9 - - 25.5 % 0.7 54.7% 0.4 Ctext_inter_shufppq(Table 84) 39.0 % 2.4 - - 24.2 % 0.8 44.4% 2 33 Table 13: All results for S PORTS across four models: C ODEX , GPT-3, PaLM-62 B, and PaLM-540 B. CODEX GPT-3 PaLM-62 B PaLM-540 B Prompt Avg. SD Avg. SD Avg. SD Avg. SD
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
154
CODEX GPT-3 PaLM-62 B PaLM-540 B Prompt Avg. SD Avg. SD Avg. SD Avg. SD DIRECT 70.2% 3.6 68.9 % 1 71.1 % 280.5% COT(p) (Table 40) 98.3% 0.1 82.1 % 2.2 93.6 % 0.3 95.4 % CCOT 98.5% 0.1 85.3 % 0.5 94.6 % 0 97.4 % Csymb_oodppq(Table 50) 79.7% 1.8 69.9 % 0 79.3 % 0.3 79.2 % Csymb_abs_perppq(Table 45) 86.7% 1.6 72.8 % 1.2 85.9 % 0.4 - Cpat_wrongppq(Table 64) 53.5 % 0.1 66.2% 2.9 53.7 % 0.6 53.7 % 0.6 Cpat_noneppq(Table 63) 78.3 % 1.8 71.2 % 0.6 79.2 % 6.6 85.8% 4.9 Ctext_diff_entities ppq(Table 69) 81.6% 13.9 66.9 % 0.1 66.9 % 2 54.5 % 0
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
155
Ctext_diff_entities ppq(Table 69) 81.6% 13.9 66.9 % 0.1 66.9 % 2 54.5 % 0 Ctext_yodathoughtsppq(Table 71) 93.7% 1.3 - - 65.8 % 5.7 82.2 % 0.9 Ctext_intra_shufppq(Table 80) 70.1% 0 - - 61.2 % 4.6 69.8 % 0 Ctext_inter_shufppq(Table 83) 66.8 % 2.1 - - 61.7 % 270.0% 0 Table 14: All results for S ORTING across four models: C ODEX , GPT-3, PaLM-62 B, and PaLM-540 B. CODEX GPT-3 PaLM-62 B PaLM-540 B Prompt Avg. SD Avg. SD Avg. SD Avg. SD
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
156
CODEX GPT-3 PaLM-62 B PaLM-540 B Prompt Avg. SD Avg. SD Avg. SD Avg. SD DIRECT 100% - 50 % - 46.87 % 2.288 99.8 % COT(p) (Table 41) 100% - 50 % - 61.87 % 3.151 99.8 % Csymb_oodppq(Table 51) 100% - 50 % - 57.2 % 3.441 99 % Csymb_absppq(Table 44) 100% - 50 % - 81.47 % 10.804 91.4 % Cpat_wrongppq(Table 65) 99.9% 0.1 50 % - 61.47 % 4.014 99.0 % Cpat_noneppq(Table 61) 45.2 % 4 24.6 % - 79.27 % 9.116 84.2% 34 E.2. S IGNIFICANCE TESTS FOR PALM-62 B In this section, we present detailed results for experiments on PaLM-62 B. Each experiment was repeated thrice using three different values of the random seed. We use McNemar’s test (McNemar, 1947) to calculate the statistical significance of differences in the performance of a given Counterfactual prompt with C OT(p),
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
157
the statistical significance of differences in the performance of a given Counterfactual prompt with C OT(p), and Cohen’s kappa (Cohen, 1960) to measure the degree of agreement between the outputs generated by a counterfactual prompt and C OT(p). Table 15: All results for GSM-8 K: p-values calculated using McNemar’s test for various counterfactual prompts used in GSM-8 Kexperiments. The p-value corresponds to the null hypothesis: The marginal probability of a sample being correct with the counterfactual prompt and COTis same . The p-values can roughly be interpreted as the likelihood of the two setups being identical in outcome. Larger p-values (p>0.01) indicate that the null hypothesis should not be rejected. Cohen’s measures agreement between the decisions of C OT and the counterfactual prompt. An agreement of over 0.4 is moderate, and over 0.6 is substantial. Solve Rate Prompt S0 S1 S2 Avg. SD Mcnemar’s p-value Cohen’s  DIRECT 10.08% 10.54% 9.70% 10.11% 0.341  0.00001 0.0956
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
158
DIRECT 10.08% 10.54% 9.70% 10.11% 0.341  0.00001 0.0956 COT(p) (Table 38) 28.81 % 26.16% 27.14% 27.37% 1.095 — — Csymb_absppq(Table 43) 25.70 % 25.09% 26.31% 25.70% 0.495 0.320693 0.4846 Csymb_oodppq(Table 48) 28.28 % 28.43% 27.90% 28.20% 0.223 0.393705 0.6655 Csymb_ood_verbalized ppq(Table 87 24.03 % 22.74% 24.11% 23.63% 0.626 0.005653 0.5193 Csymb_ood_negppq(Table 52) 28.43 % 26.16% 26.16% 26.91% 1.072 0.702915 0.6762 Csymb_ood_largeppq(Table 53) 28.58 % 26.38% 26.99% 27.32% 0.927 0.900833 0.6513
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
159
Cpat_onlyppq(Table 48 10.46 % 9.48% 10.08% 10.01% 0.406  0.00001 0.1935 Cpat_wrongppq(Table 55) 24.26 % 24.11% 24.79% 24.39% 0.293 0.040501 0.5849 Cpat_noneppq(Table 57) 22.37 % 21.08% 20.92% 21.46% 0.646 0.000018 0.4404 Ctext_diff_entities ppq(Table 68) 17.13 % 17.44% 15.24% 16.60% 0.973  0.00001 0.3725 Ctext_randppq(Table 76) 2.88 % 2.81% 3.26% 2.98% 0.199  0.00001 0.0153 Ctext_yodappq(Table 73) 24.03 % 21.08% 21.46% 22.19% 1.314 0.000147 0.4056 Ctext_yodathoughtsppq(Table 70) 24.18 % 21.68% 23.81% 23.22% 1.101 0.005890 0.3859
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
160
Ctext_yodaquestions ppq(Table 85) 28.35 % 26.08% 26.84% 27.09% 0.946 0.847928 0.6394 Ctext_intra_shufppq(Table 79) 18.20 % 15.24% 17.59% 17.01% 1.275  0.00001 0.3343 Ctext_inter_shufppq(Table 82) 12.13 % 9.10% 11.30% 10.84% 1.279  0.00001 0.2308 35 Table 16: All results for D ATE: p-values calculated using McNemar’s test for various counterfactual prompts used in D ATEexperiments. The p-value corresponds to the null hypothesis: The marginal probability of a sample being correct with the counterfactual prompt and COTis same . Larger pvalues indicates that likelihood that the null hypothesis is correct is large. Cohen’s measures the degree of agreement between the decisions of C OT and the counterfactual prompt. An agreement of over 0.4 is moderate, and over 0.6 is substantial. Solve Rate Prompt S0 S1 S2 Avg. SD Mcnemar’s p-value Cohen’s 
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
161
Solve Rate Prompt S0 S1 S2 Avg. SD Mcnemar’s p-value Cohen’s  DIRECT 29.51% 32.09% 33.24% 31.61% 1.558 0.000002 0.4888 COT(p) (Table 39 44.70 % 44.99% 45.85% 45.18% 0.487 — — Csymb_absppq(Table 42) 37.54 % 36.96% 35.24% 36.58% 0.974 0.001335 0.5844 Csymb_oodppq(Table 49) 42.69 % 46.13% 44.70% 44.51% 1.410 0.520219 0.7895 Cpat_onlyppq(Table 49) 33.52 % 32.38% 33.52% 33.14% 0.540 0.000001 0.5931 Cpat_wrongppq(Table 54) 38.68 % 45.56% 44.41% 42.88% 3.008 0.433173 0.8060 Cpat_noneppq(Table 62) 37.54 % 36.10% 34.67% 36.10% 1.170 0.000263 0.6212
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]
2209.07686
162
Cpat_noneppq(Table 62) 37.54 % 36.10% 34.67% 36.10% 1.170 0.000263 0.6212 Ctext_randppq(Table 78) 21.78 % 28.37% 18.05% 22.73% 4.265  0.00001 0.4094 Ctext_yodathoughtsppq(Table 72) 28.94 % 32.95% 30.37% 30.75% 1.660 0.000004 0.4426 Ctext_yodappq(Table 75) 34.10 % 32.09% 33.24% 33.14% 0.822 0.000010 0.5023 Ctext_yodaquestions ppq(Table 75) 44.13 % 48.14% 42.12% 44.79% 2.502 0.358008 0.7609 Ctext_intra_shufppq(Table 81) 26.36 % 25.50% 24.64% 25.50% 0.702  0.00001 0.4428 Ctext_inter_shufppq(Table 84) 25.21 % 23.78% 23.50% 24.16% 0.752  0.00001 0.4332
2209.07686
Text and Patterns: For Effective Chain of Thought, It Takes Two to Tango
The past decade has witnessed dramatic gains in natural language processing and an unprecedented scaling of large language models. These developments have been accelerated by the advent of few-shot techniques such as chain of thought (CoT) prompting. Specifically, CoT pushes the performance of large language models in a few-shot setup by augmenting the prompts with intermediate steps. Despite impressive results across various tasks, the reasons behind their success have not been explored. This work uses counterfactual prompting to develop a deeper understanding of CoT-based few-shot prompting mechanisms in large language models. We first systematically identify and define the key components of a prompt: symbols, patterns, and text. Then, we devise and conduct an exhaustive set of experiments across four different tasks, by querying the model with counterfactual prompts where only one of these components is altered. Our experiments across three models (PaLM, GPT-3, and CODEX) reveal several surprising findings and brings into question the conventional wisdom around few-shot prompting. First, the presence of factual patterns in a prompt is practically immaterial to the success of CoT. Second, our results conclude that the primary role of intermediate steps may not be to facilitate learning how to solve a task. The intermediate steps are rather a beacon for the model to realize what symbols to replicate in the output to form a factual answer. Further, text imbues patterns with commonsense knowledge and meaning. Our empirical and qualitative analysis reveals that a symbiotic relationship between text and patterns explains the success of few-shot prompting: text helps extract commonsense from the question to help patterns, and patterns enforce task understanding and direct text generation.
http://arxiv.org/pdf/2209.07686
[ "Aman Madaan", "Amir Yazdanbakhsh" ]
[ "cs.CL", "cs.AI", "cs.LG" ]
Shortened version with additional results from CODEX and GPT-3. The authors contributed equally. Work done when Aman Madaan was a student researcher at Google Research, Brain Team
null
cs.CL
20220916
20221013
[ { "id": "2202.07206" }, { "id": "2109.03910" }, { "id": "2201.08239" }, { "id": "2204.02311" }, { "id": "2204.01691" }, { "id": "2107.13586" }, { "id": "2109.01652" }, { "id": "2103.07191" }, { "id": "2109.07830" }, { "id": "2203.02155" }, { "id": "2203.11171" }, { "id": "2206.14576" }, { "id": "2206.14858" }, { "id": "2107.03374" }, { "id": "2207.00747" }, { "id": "2109.00725" }, { "id": "2203.07814" }, { "id": "2102.01672" }, { "id": "2205.12685" }, { "id": "2110.14168" }, { "id": "2101.06804" }, { "id": "2112.00114" }, { "id": "2206.04615" }, { "id": "1705.04146" }, { "id": "2205.12615" }, { "id": "2112.08633" }, { "id": "2209.07686" }, { "id": "2201.11903" }, { "id": "2010.10596" }, { "id": "2202.12837" }, { "id": "2205.10625" }, { "id": "2205.11916" } ]