Reinforcement learning from human feedback (RLHF)
Reinforcement learning from human feedback (RLHF) is a method for training a language model that has already undergone pretraining, so that it behaves in accordance with human preferences.
On this page
Source content checked . English translation published 2026-10-27.
In business automation
Models used as assistants go through a stage that guides them to behave as people expect. Reinforcement learning from human feedback (RLHF) is a method for training a language model that has already undergone pretraining, so that it behaves in accordance with human preferences. People building automation do not run RLHF themselves, but understanding it helps explain why different models behave differently.
Examples
These examples illustrate a possible workflow. Adapt the connections and rules to your own systems.
People rating responses
In the RLHF process, people compare several model responses and mark which one is better. These ratings are used to train the model to prefer responses similar to those selected.
Relevant tools
- Anthropic
- This definition is from Anthropic's glossary.
What to check
- Assuming that RLHF makes a model immune to errors. It guides behavior according to preferences; it does not add knowledge or guarantee accuracy.