text
stringlengths
62
4.5k
arxiv-translator
stringlengths
51
2.86k
A more conservative API provider may remove access to the combination of logit bias and logprobs entirely. Indeed, after disclosing our attack to OpenAI, they removed the ability for logit bias to impact the top logprobs--thus preventing the attacks from the prior sections. To exploit situations such at this, we further develop several logprob-free attacks that recover the complete logit vector by performing binary search on the logit bias vector, albeit at increased cost.
보닀 보수적인 API κ³΅κΈ‰μžλŠ” λ‘œμ§“ 편ν–₯κ³Ό λ‘œκ·Έν”„λ‘œλΈŒμ˜ 쑰합에 λŒ€ν•œ μ•‘μ„ΈμŠ€λ₯Ό μ™„μ „νžˆ μ œκ±°ν•  수 μžˆλ‹€. μ‹€μ œλ‘œ μ˜€ν”ˆAI에 우리의 곡격을 κ³΅κ°œν•œ ν›„, 그듀은 λ‘œμ§“ 편ν–₯이 μƒμœ„ λ‘œκ·Έν”„λ‘œλΈŒμ— 영ν–₯을 λ―ΈμΉ˜λŠ” λŠ₯λ ₯을 μ œκ±°ν•˜μ—¬ 이전 μ„Ήμ…˜μ—μ„œ 곡격을 λ°©μ§€ν–ˆλ‹€. μ΄λŸ¬ν•œ 상황을 ν™œμš©ν•˜κΈ° μœ„ν•΄ λΉ„μš© 증가에도 λΆˆκ΅¬ν•˜κ³  λ‘œμ§“ λ°”μ΄μ–΄μŠ€ 벑터에 λŒ€ν•΄ 이진 검색을 μˆ˜ν–‰ν•˜μ—¬ μ™„μ „ν•œ λ‘œμ§“ 벑터λ₯Ό λ³΅κ΅¬ν•˜λŠ” μ—¬λŸ¬ λ‘œκ·Έν”„λ‘œλΈŒ μ—†λŠ” 곡격을 μΆ”κ°€λ‘œ κ°œλ°œν•œλ‹€.
3. 30/30 workout: 30 seconds of intense exercise followed by 30 seconds of rest for several rounds.
3. 30/30 μš΄λ™: 30초의 κ²©λ ¬ν•œ μš΄λ™ ν›„ 30초의 νœ΄μ‹μ„ μ—¬λŸ¬ λΌμš΄λ“œ λ™μ•ˆ μˆ˜ν–‰ν•œλ‹€.
Footnote 6: There is evidence from Grosse et al. (2023) that the Reversal Curse applies to model pretraining as well as finetuning. For cost reasons, we tested finetuning rather than pretraining.
각주 6: Grosse et al.(2023)의 증거가 μžˆλŠ”λ°, μ—­μ €μ£ΌλŠ” λͺ¨λΈ 사전 ν›ˆλ ¨λΏλ§Œ μ•„λ‹ˆλΌ λ―Έμ„Έ 쑰정에도 μ μš©λœλ‹€λŠ” 것이닀. λΉ„μš©μ μΈ 이유둜 사전 ν›ˆλ ¨λ³΄λ‹€λŠ” λ―Έμ„Έ 쑰정을 ν…ŒμŠ€νŠΈν–ˆμŠ΅λ‹ˆλ‹€.