RLHF (Reinforcement Learning from Human Feedback)

Definition

RLHF is a technique that builds a reward model from human preference judgments and then uses reinforcement learning to align a model's behavior. It is widely used to improve helpfulness and safety.