Loading…
Glossary term
A method in which a model is trained to follow a written set of principles.
Introduced by Anthropic, Constitutional AI trains a model to critique and revise its own answers against a written list of principles, the 'constitution'. AI-generated feedback based on those principles then supplements human feedback.
It makes the values being trained in explicit and easier to inspect.
It is one approach to reducing the need for large amounts of human labelling of harmful content.