Mikita Balesni
@balesni
Former AI alignment researcher @openai. Past: @apolloaievals, Reversal curse, Out-of-context reasoning
参加 June 2013
704 フォロー中 6.9K ファン
maybe we could just pause on making models bigger or otherwise smart without CoT until we figure out interpretability
@eccentric1ty Chain of thought becames less interpretable over time because models become bigger anyway right
Also optimized chain of thought is already used by models in kind of weird uninterpretable ways
もっと見る
