Glossary
What is RLAIF? Definition and meaning
Uses an AI model, rather than humans, to produce the preference labels used for alignment.
Definition
RLAIF: Uses an AI model, rather than humans, to produce the preference labels used for alignment.
In short: Reinforcement learning from AI feedback.
When it is used
Scaling alignment data when human labelling is too slow or costly.
Learn more
The full guide to RLAIF explains how it works, when to use it and how it compares with the alternatives, with examples. The glossary lists every term on the site from A to Z.