Alignment and Safety

Reward Modeling

Approach where a reward model learns to predict human preferences, serving as a guide for reinforcement learning of main language models.

← Indietro