Glossary · Model Training

RLAIF (Reinforcement Learning from AI Feedback)

A variant of RLHF where an AI model, rather than a human, provides the preference judgments used to train the reward signal, reducing the human labeling cost of alignment.