Training
Constitutional AI
Anthropic's method of training models to follow a written set of principles, using AI feedback guided by those principles.
Instead of relying only on human ratings, the model critiques and revises its own answers against the principles. The approach makes the intended values explicit and easier to inspect.
Frequently asked questions
What does Constitutional AI mean?
Anthropic's method of training models to follow a written set of principles, using AI feedback guided by those principles. Instead of relying only on human ratings, the model critiques and revises its own answers against the principles. The approach makes the intended values explicit and easier to inspect.