Constitutional AI
In one line: Anthropic's training method where Claude is trained against a written 'constitution' of values — rather than ad-hoc human feedback for every example.
Constitutional AI (CAI) is Anthropic's approach to training Claude. Instead of having humans rate every response (expensive, inconsistent), CAI uses a written list of principles — the 'constitution' — that the model uses to critique and revise its own responses.
Principles include things like: 'Choose the response that is most helpful, honest and harmless.' Claude uses these principles to self-improve during training, producing a model that's more consistent across edge cases and easier to audit.
You'll feel CAI in Claude's tendency to be more cautious about harmful requests and more thoughtful in nuanced situations than other models.
See it in action — ask any AI about constitutional ai on AskAI.free.
Try it free →