Steven Byrnes comments on orthonormal’s Shortform

[Steven Byrnes]

Update: I wrote a post which expands on this comment: LLMs are (still) mostly powered by imitative learning, not RL.