Skip to content

Search across models, tools, services, companies, providers, APIs, plans, articles and news.

Training & tuning

RLHF

RLHF

RLHF

A technique where humans rank the model’s outputs and it learns to give more helpful, accurate and safe answers — the technical core behind ChatGPT’s and Claude’s behavior.

RLHF (Reinforcement Learning from Human Feedback) is the model’s upbringing stage: humans rank several model answers, a reward model learns those preferences, and the main model is trained to maximize the reward. The result: more helpful, accurate and polite.

This is the trick that separated ChatGPT from a raw text model; lighter variants (DPO) are standard today. The known critique: the model can drift toward pleasing humans rather than being correct — so high-quality human data stays the bottleneck.

In short: pre-training gives understanding, RLHF gives manners and honesty. Modern model training is the balance of these stages.

Back to glossary