RLHF (Reinforcement Learning from Human Feedback)

A training method that uses human ratings to teach a model to produce more helpful, safe, and polite responses.