Glossary

Policy model — Glossary

In RL, the component that selects actions; in RLHF, the language model being optimized.

Reviewed:

Full definition

The policy maps states (or prompts) to actions (or responses). RLHF updates the policy to score well under a reward model while limiting drift from a reference policy.

Related encyclopedia topic

RLHF — Reinforcement Learning from Human Feedback (RLHF) shapes model behavior using preference data and reward models rather than only static supervised targets.

← Back to glossary · Learning hub