text
stringlengths 4
378
| likes
stringlengths 0
4
| reply
stringlengths 0
309
|
|---|---|---|
Fun panel on open source AI that took place last week at the AI House in Davos. | 16 | Panel on Open Source in AI from the AI House at Davos with |
RIP Arno Penzias.
He was the head of Bell Labs' research organization, known as "Area 11," when I joined in 1988. | 81 | |
It makes sense to cryptographically watermark *authentic* photos and videos.
But camera manufacturers, image processing software vendors, and distribution platforms have to agree on a standard.
Watermarking AI-generated or heavily manipulated content may have some usefulness,… | 564 | Mandating watermarking of AI text is a magical thinking solution to a non-existent problem that will not work in reality.
OpenAI gave up on theirs because it had 26% accuracy.
How good do you think your watermarking tech will be if the best OpenAI could come up with was 26%?… |
It's a joke, right? | 1K | - Meta, by open sourcing competitive models (e.g., Llama 3) they reduce AI orgs' revenue/valuations/ability to buy more GPUs and scale AI models twitter.com/DanHendrycks/s… |
Yay, I'm a Marvel superhero!
Where's my Iron Man suit? | 391 | InstantID demo is now out on Spaces.
Thanks |
Very honored to have received the 2023 Global Swiss AI Award in Davos.
The ceremony took place in the Davos city hall in the presence of the mayor and involved me playing a kind of Swiss cowbell glockenspiel. | 763 | |
Global Swiss Award website:
https://mindfire.global/global-swiss-ai-award… | 27 | |
Cowbells | 45 | |
In 1995, Patrick John Hayes established the Simon Newcomb Award for the silliest argument against the possibility of artificial intelligence.
Simon Newcomb was the Canadian astronomer and mathematician who "proved" that manned heavier-than-air flight was impossible shortly… | 266 | I propose the creation of the DeTuring Award.
It will be granted to people who are consistently trying (and failing) to deter society from using computer technology by scaring everyone with imaginary risks.
As the Turing Award is the Nobel Prize of computing, the DeTuring Award… |
MIT Economist
@amcafee
explains why revolutions caused by new general purpose technologies (GPT), such as AI, do *not* cause mass unemployment. | 381 | @ylecun |
Welcome to FAIR, Taco!
We're thrilled to count you as a colleague. | 211 | Two weeks ago I joined Meta / FAIR, and I couldn't be more excited about this new chapter. Meta is indeed the only place left that supports highly ambitious long-term oriented & fundamental research projects and has a strong commitment to open science and open source. (and has… twitter.com/ylecun/status/… |
The
@wef
in Davos was intense and fruitful. | 471 | |
A great interview of Nick Clegg and me in El Païs about AI.
We discuss the limits of current technology, the necessity of building human-level AI assistants, and the regulation debates. | 158 | |
I propose the creation of the DeTuring Award.
It will be granted to people who are consistently trying (and failing) to deter society from using computer technology by scaring everyone with imaginary risks.
As the Turing Award is the Nobel Prize of computing, the DeTuring Award… | ||
Confused about the history of AI R&D at Meta?
Here is the TL;DR | 314 | @nathanbenaich |
Confused about the recent announcement about FAIR?
Here is the TL;DR: FAIR and GenAI are now sister-but-separate organizations under Chief Product Officer Chris Cox. | 65 | @giffmana |
Self-Rewarding LLMs.
From FAIR+NYU. | 630 | New paper! |
AI could help with carbon capture. | 336 | Meta and |
Removing a space can make a phrase lose all its meaning, as in
"Open AI" -> "OpenAI"
and also as in
"Le Cun" -> "LeCun" | 559 | Zuck and Yann LeCunn will go down as heroes in human history!
Fighting for Open AI when the incumbents sought to shut it down!
Unbelievable how much the vibes from Meta have changed over the last year. |
I have a simpler name for "group attention": pooling. | 432 | The impressive in-context learning abilities of LLMs has created the need for larger context windows. Recently, researchers discovered that we can easily extend the context window of a pretrained LLM with one simple trick (and no extra training)… |
Precisely.
The general public often isn't aware that ideas don't just pop up out of the vacuum.
In a domain where people publish, there is often a long trail of interacting contributions, each of which plays a role. | 500 | Lot of people forget that just a few months prior to the Transformer, FAIR had the state of the art non recurrent machine translation model: ConvS2S. The highest order bit on the impact Transformers had was parallel forward and backward passes for training sequential models,… twitter.com/ylecun/status/… |
A big bunch of nails in the coffin of the idea that open source AI models are more dangerous than closed ones.
And a huge warning to regulators who are on the way to throw the AI baby with the regulatory bath water.
What has been happening since the release of Llama-2 is an… | 1K | |
There is literally no other company doing this today:
- open research towards human-level AI
- open source AI platform enabling a huge AI ecosystem
- wearable device to interact with always-on AI assistants | 2K | Open Source AGI is an amazing vision. You are building a very powerful technology, and, actually aligning to what makes sense for the world: more people have a say in what makes sense and doesn't. Zuck and |
When something is viewed as a common software platform it has to be open source. | 884 | @wef |
A discussion between Karl Friston and me at the #WEF2024 in Davos. | 91 | @helloVERSES |
No open source AI ==> No AI startup ecosystem. | 842 | @ylecun |
FAIR's mission is to develop the science and technology for human-level AI assistants: machines that understand the world, can perceive, remember, reason, plan, and act.
In the not-to-distant future, all of our interactions with the digital world will be mediated by AI… | 1.5K | Zuckerberg says Meta wants to build AGI and open source it, brings Meta's AI group FAIR closer to generative AI team; Meta will own 340K+ H100 GPUs by 2024 end ( |
Better link. | 89 | Meta is all in on open source AGI, says Zuckerberg |
"So, first of all, let me assert my firm belief that the only thing we have to fear is...fear itself — nameless, unreasoning, unjustified terror which paralyzes needed efforts to convert retreat into advance."
-- Franklin D. Roosevelt. | 794 | Imagine if all the people creating the web browser and the early internet were constantly warning you that the internet would destroy the world instead of doing what it actually did, which is make the world more connected than ever before, while giving you access to all the… |
Avec
@EmmanuelMacron
à Davos. | 83 | |
Meta has always tried to do the Right Thing.
Meta has always practiced open research in AI.
Meta has been promoting open source AI platforms.
After numerous discussions over the last year (sometimes contentious) a consensus is emerging that open source AI platforms are… | 1.7K | @ylecun |
With
@EmmanuelMacron
in Davos. | 159 | |
A panel on AI at the
@wef
in Davos earlier today.
With
@AndrewYNg
,
@DaphneKoller
, @kaifulee,
@aidangomez
, and me, masterfully moderated by Nick Thompson from The Atlantic. | 463 | The Expanding Universe of Generative Models with |
To take Giant Steps you need a large learning rate. | 142 | What are the 20 most influential Atlantic Records jazz albums? A thread |
The Renaissance painting one at the bottom left is my favorite. | 614 | Create personalized stylish photo without LoRA training with |
Awesome text-to-{music,sound} system from FAIR.
And yes, it's open source. | 847 | Happy to share MAGNeT |
I love this.
This is why open source AI platforms will win: it's the only way for AI to cater to highly diverse languages, cultures, values, and centers of interest. | 1.3K | We release Kan-LLaMA [ಕನ್-LLama] — A 7B Llama-2 LoRA PreTrained and FineTuned on "Kannada" tokens |
Meta spends twice more on R&D than Amazon, Alphabet, and Microsoft, and 4 times more than Apple, when normalized by revenue.
The only other big tech that comes close is Nvidia.
That tells you a lot about who is building the technology of tomorrow. | 3.4K | It is startling to see how much of the world's R&D spending comes from (mostly American) tech giants.
The R&D spending of Amazon is greater than the R&D spending of all companies and government in France. Alphabet beats Italy.
Pepsi's R&D spending beats all sources in Nigeria. |
We need multiple AI assistants for the same reason we need multiple sources of news and information.
We need open source AI platforms for the same reason we need content-agnostic information distribution channels.
Diversity. | 983 | Just checking in on alignment of LLMs in China, it's going about how you'd expect. |
There are many good things and bad things about living in the US and living in France.
Depending on your situation, profession, aspirations, and taste, you might prefer one or the other.
Personally, I like both and enjoy the best each has to offer | 301 | @Appyg99 |
Encoders are useful. | 670 | Generative large language models (LLMs) are based upon the decoder-only transformer architecture. Currently, these types of generative LLMs are incredibly popular. However, I use encoder-only architectures for 90% of use cases as a practitioner. Here’s why… |
Passion Dance by McCoy Tyner (piano), with Joe Henderson (tenor sax), Ron Carter (bass), Elaine Jones (drums).
Sooo much energy in McCoy and Joe's solos in this 1967 recording !
https://youtu.be/0zdoOyeRTCA?si=2D8httj156CeZmDY…
https://music.youtube.com/watch?v=0zdoOyeRTCA&si=G4T2ITJrwcGjdrSo…
https://open.spotify.com/track/0lELi5BqmUO4hXTFfAUf60?si=3pOz1yr3QPe04twpO7Qu0A… | 86 | |
s/Elaine/Elvin/ | ||
SeamlessExpressive from Meta-FAIR on HuggingFace: speech-to-speech translation that preserves the voice and the expression. | 155 | SeamlessExpressive enables high-quality speech translation that maintains the original speaker's vocal style, tone and unique expressions in translated outputs.
More documentation on our |
You know Elon, properly-run social networks do their best to take down misinformation about elections. | 3.2K | 1. |
Seriously, if you are a legal resident in the US, would you attempt to register to vote and risk losing your visa or green card and be deported, just so you can vote?
If you are an undocumented immigrant, would you attempt to register to vote and risk being thrown in jail and… | ||
Nice.
This used to be really, really hard. | 1.3K | Announcing surya - a multilingual text line detection model for documents. It gives you accurate line-level bboxes and column breaks.
Find it here - |
Perceptual discrepancy. | 58 | A bit late to this, but YouGov polling matches a lot of what I've been saying. Most Americans say 2023 was great, good, or OK for them, but a majority say it was bad for the country — and the disparity is largely driven by Republicans |
Brains beat brawn. | 818 | it’s actually insane that a hacked together 8*30b model by like 5 cracked french guys is beating the billions poured into anthropic.
honestly very bearish on anthropic twitter.com/lmsysorg/statu… |
Good. | 60 | The latest Commission agenda (dated 7 January) confirms a Connectivity package on digital networks and infrastructure and the Initiative to open up European supercomputer capacity to AI start-ups on 21 February. |
Very cool study comparing ConvNext and ViT architectures of similar sizes, trained in either supervised mode or with a CLIP-style method, and compared on a variety of properties.
A collaboration between FAIR and MBZUAI.
Web page: https://kirill-vish.github.io/beyond-imagenet-accuracy/… | 472 | How to choose a vision model for your specific needs?
How do ConvNet / ViT, supervised / CLIP models compare with each other on metrics beyond ImageNet?
Our work comprehensively compares common vision models on "non-standard" metrics. (1/n) |
Audiobox demo: audio generation from voice and text prompts. | 192 | You can now try Audiobox, our new foundation research model for audio generation that can generate audio using a combination of voice inputs and natural language text prompts.
Try the demo |
Modems use an adaptive linear classifiers to turn analog signals into bit strings. It's called an "adaptive equalizer"
It is trained in two phases. The first one is supervised, the second one is self-supervised:
1. In the initial phase, when the modem first connects to another… | 1.3K | Did you know that some of the sounds old modems made when connecting (those 'bipppppp' tones) were actually due to them using a self-supervised learning method for adaptation? |
The speed at which a technology disseminates in the economy is limited by how fast people learn to use it. | 573 | Technology maturity isn’t sufficient to drive adoption
Insightful analysis of the rise of tractors in US farms, put in parallel with Generative AI, by |
This is inspired by
@erikbryn
's work on the effect of technological revolutions on the labor market. It takes 15 to 20 years for technology revolutions to have a measurable impact on productivity because that's how long it takes for people to learn the new skills. I'm not sure… | ||
- AI techies: oh look, we can translate hundreds of languages automatically. This can lower linguistic barriers and connect humanity.
- AI negativists: "This is the world we're creating. Removing the humanity from how we learn to connect with humanity." | ||
In case you are wondering, this is an actual quote from a tweet. I didn't retweeted that tweet because I don't want to turn this general comment into a personal attack. | ||
Monk. | 145 | You can only choose one. Who’s your go-to jazz composer?
Thelonious Monk | Miles Davis
Charles Mingus | Duke Ellington |
Miles is a close second. Wins on diversity. | ||
Venture capital firm Andreessen-Horowitz
@a16z
wrote a letter to the UK House of Lords in defense of open source AI platforms. | 1K | |
A few reasons why Auto-Regressive LLMs are very helpful for coding, even though they really can't plan. | 316 | Doesn't the fact that LLMs can "write code" prove that they can also reason and plan? (Hint: Nope..) |
One reason AI research has been progressing so fast is not just because of frequent and early publication of preprint on ArXiv and the exchange of open source code, but also because the ML/AI community has largely freed itself from the stranglehold of for-profit & paywalled… | 2.6K | @pmarca |
Hehe | 486 | Which big tech will be leading open-source AI in 2024? |
Paris AI is on fire. | 1.2K | The ML ecosystem in France is on fire |
Here is a list of inexcusable mistakes in slide design:
- not using all the space: showing a diagram or figure that does not fill as much space as possible on your slide. It's very frustrating for the audience to have to squint when there is ample unused space around a figure.
-… | 577 | @shimon8282 |
Reward densification. | 91 | @sainingxie |
Congrats
@lxbrun
,
@braizh
,
@delphinegroll
and the whole
@nabla_ai
team. | 253 | Nabla raises another $24 million for its AI assistant for doctors |
Nabla's shift from ChatGPT to open source LLMs (Llama and/or Mistral) is a perfect example of why I think open source AI platforms will constitute the foundation of the AI ecosystem of vertical applications. | ||
Hey,
@HachetteUS
,
@HachetteLivre
,
@GroupeLagardere
, and
@arnaudlagardere
, drop the hatchet against
@internetarchive
.
You are on the wrong side of history. | 609 | Friend of the Court Briefs Filed in Internet Archive’s Appeal — Hundreds of librarians joined prominent library organizations & non-profits in filing amicus briefs in Hachette v. Internet Archive. Read why they believe our appeal should succeed: |
Congrats
@denisyarats
and
@AravSrinivas
! | 449 | We are happy to announce that we've raised $73.6 million in Series B funding led by IVP with participation from NVIDIA, NEA, Bessemer, Elad Gil, Jeff Bezos, Nat Friedman, Databricks, Tobi Lutke, Guillermo Rauch, Naval Ravikant, Balaji Srinivasan. |
Why I turned down an offer for Director of Research at Google in January 2002. | 1.3K | @giffmana |
In which we learn that:
1. Academic books are at best "books" or merely "written material" just because their authors expect no remuneration.
2. I distributed "books" for free after doing what I'm being accused of refusing to do: asking for permission from the owners. Still not… | 210 | This is not a story about “free books.” It’s a story about a narrow & specific modality for distribution of written materials for which the authors expected no remuneration. The suggestion it’s about “books” writ large is a disingenuous stop on the non-apology tour. twitter.com/ylecun/status/… |
Let me tell a story about free books.
In the mid 1990s, I started a project called DjVu at AT&T Labs.
The purpose was to devise a new image compression format so that printed documents could be scanned at high resolution and distributed efficiently over the newly expanding… | ||
Interesting side story: Volume zero of NIPS (1987) was published by the American Institute of Physics, which is a non-profit publisher.
AIP had sold the rights to Springer, which is very much for profit. We contacted Springer USA and requested permission to scan and post the… | ||
It took less than 100 years from 1924. | 228 | @paulisci |
Only a small number of book authors make significant money from book sales.
This seems to suggest that most books should be freely available for download.
The lost revenue for authors would be small, and the benefits to society large by comparison. | 2.5K | A survey of 5,699 published authors, found that in 2022, their median gross pre-tax income from their books was $2,000.
You can generate more revenue than that in less time by mowing lawns in your neighborhood. |
A lot of misunderstanding in the comments.
I'm certainly not suggesting there should be a rule forcing authors to release their books for free or anything.
I'm just making the point that many authors who are more motivated by intellectual impact than by a thousand bucks of income… | ||
A more detailed explanation for those who still don't get the point. | 52 | @Rahll |
Soooo much misunderstanding of my tweet on book publishing!
I had to write a long explanation (below).
I'm guessing that a lot of negative reactions are caused by some assumption of ill intent on my part within the current debate about copyright and AI.
But this has absolutely… | 435 | @Rahll |
Happy Doom-Free New Year | ||
[sorry for the hash code collision,
@ID_AA_Carmack
] | ||
AI is not some sort of natural phenomenon that we have no control over.
AI is being built by us, humans.
Hence, if you're scared of AI, what you are actually scared of are your fellow humans.
You probably have doubts about the ability of institutions to do the right things with… | ||
Better access to knowledge is ....
... As bad as opium | 243 | Free libraries as bad as free opium??!
"that makes the wholesale establishment of free public libraries almost as much a menace to the well-being of the country as would be the opening up of free saloons or opium joints" - The Marion Daily Mirror, 1910 (full page scan below) |
Who knew that coffee destroyed the fabric of society | 336 | Here's the story of another technology that faced massive backlash in its time that will sound very familiar to today's battles over |
Never mind the typewriter, what about P(doom | PowerPoint) ?
"We had 12.9 GB of slides on our network and I thought, 'What a huge waste of corporate productivity.' So we banned it. And we've had three unbelievable record-breaking fiscal quarters since. Now, I would argue that… | 680 | What is your typewriter P(doom) number?
"If the typewriter were abolished it would improve our thinking and lessen the threat of war." (1955) |
The one true American art form was once thought to destroy the fabric of society and corrupt the youth. | 198 | @PessimistsArc |
Lots of great advances in AI from Meta in 2023. | 707 | To close out 2023, here are 10 of the most interesting AI research advancements we shared on our feed this year — and where you can find more details on the work. |
I agree with these predictions about AI from
@MartinSignoux
. | 307 | What a year it's been for AI.
Anticipating what's next sounds very perilous, but I'll give it a try.
Here are eight AI predictions for 2024 |
Genuine question about image generation:
If someone uses a generative AI tool to produce an image that is substantially similar to a copyrighted piece (a drawing, painting, movie screenshot, etc), who should be liable for copyright infringement?
Should it be:
A. the company… | ||
Interesting take. | 214 | If you want to understand why the Times case has a near zero probability of winning, then read this thread.
This fellow does a nice write up and he seems sincere in his belief that what he is saying about the suit is accurate and correct when in fact it's basically just a lot… twitter.com/jason_kint/sta… |
Awww thanks,
@Dan_Jeffries1
.
I don't have much in common with Voltaire (sadly), but I like this citation of his:
"Le bonheur est souvent la seule chose qu'on puisse donner sans l'avoir et c'est en le donnant qu'on l'acquiert."
"Happiness is often the only thing one can give… | 534 | Yann has quickly become the Voltaire of AI.
It's amazing that we have someone with a big profile out there passionately talking real sense and clear thinking pragmatism, while chicken littles scream the sky is falling and advocate for massive centralized control regimes. |
René Barjavel foresaw our addiction to smartphones! | 646 | In 1947, French science fiction writer René Barjavel was asked for a prediction of the future. He imagined a future with cell phones-like portable TVs and electronic communication in the film La télévision, œil de demain |
How to do real-time speech-to-speech translation while minimizing the latency | 632 | SeamlessStreaming uses a learned read/write policy to enable delivery of fast & accurate translations even when faced with differences in language structure in many language pairs.
Details on Seamless |
A lot of things are getting better in the world. | 383 | A nice way to end the year with some data: 66 Good News Stories You Didn't Hear About in 2023 |
Happy Isaac Newton Birthday
Also, happy 76th Birthday to the transistor (actually, 76 years and 2 days). | ||
Punchline needs an update:
"Sure, easy image recognition with a ConvNet. Gimme PyTorch and a few hours." | 922 | What was jokingly called impossible just a few years ago, is now completely possible and available at the tips of our fingers now |
Good move, Poland.
You need nuclear badly. | 1.1K | |
Answering by approximate retrieval or by understanding+reasoning are two ends of a spectrum.
Humans are at various places on this spectrum, depending on the task, experience, and depth of understanding.
We see this in physics or math students: some will study very hard, do lots… | 1.4K | Unfortunately , too few people understand the distinction between memorization and understanding. It's not some lofty question like "does the system have an internal world model?", it's a very pragmatic behavior distinction: "is the system capable of broad generalization, or is… twitter.com/RichardMCNgo/s… |
Open source AI foundation models will wipe out closed and proprietary AI models for the same reason Wikipedia wiped out generalist commercial encyclopedia: crowd-sourced human contributions to open platforms can cater to a high diversity of interests, cultures, and languages. | ||
Dans les semaines précédents les négociations de L'AI Act à Bruxelles, une mystérieuse organisation a financé une campagne de propagande en faveur de régulations restreignant la R&D en IA.
Leur but était de faire interdire l'IA open source au niveau européen.
La campagne était… | 282 |
Les médias: |
End of preview. Expand
in Dataset Viewer.
- Downloads last month
- 48