From Atari to ChatGPT: How AI Learned to Follow Instructions
Linear Digressions
00:00
Human preferences as a training signal
Ben traces the 2017 insight to use human pairwise preferences to teach agents what humans like.
Play episode from 05:28
Transcript
