Training AI Without Writing A Reward Function, with Reward Modelling

259,419 views

About this video

Hand-picked by a BrighterTube curator · found via HackerTube (Karpathy's dataset of videos Hacker News recommends

Explains reward modelling, where an agent learns what humans want from pairwise preference comparisons, walking through the 2017 backflipping-robot paper that RLHF grew from; built straight from the research.

Robert Miles explains reward modelling: training AI from human feedback when you can't write the reward function.

Video by Robert Miles AI Safety , played through YouTube's embedded player.

Part of Best Of, Technical