ModelRefs / Reinforcement Learning from Human Feedback (RLHF) — AI Glossary
Reinforcement Learning from Human Feedback (RLHF) — AI Glossary
A training paradigm using human preference signals and RL to align model outputs with human values and instructions.
Overview
RLHF (Stiennon et al. 2020; Ouyang et al. 2022) has three stages: supervised fine-tuning on demonstrations, reward model training on preference pairs, and PPO optimization against the reward model. Used to train InstructGPT, ChatGPT, Claude, and Gemini. Largely complemented by DPO, which skips the explicit RL stage.
Reference details
| Topic | training |
|---|---|
| Also known as | RLHF, human preference learning |
| Last reviewed | 2026-06-24 |
Related terms
Primary source
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Reinforcement Learning from Human Feedback (RLHF) — AI Glossary.
Frequently asked questions
What is Reinforcement Learning from Human Feedback (RLHF)?
A training paradigm using human preference signals and RL to align model outputs with human values and instructions.
Is Reinforcement Learning from Human Feedback (RLHF) the same as RLHF?
Yes — RLHF, human preference learning are common aliases for Reinforcement Learning from Human Feedback (RLHF).
What concepts are related to Reinforcement Learning from Human Feedback (RLHF)?
Closely related concepts include rlhf, dpo, reward model, ppo.