How It Works
Constitutional AI
A training method in which a model is given a written list of principles — a “constitution” — and uses it to critique and revise its own answers, then to judge which answers better follow the principles. Those judgments replace most human labels for harmlessness. Critics point out that the company writing the constitution still decides whose values go into it.
Origin
Introduced by Anthropic in “Constitutional AI: Harmlessness from AI Feedback” by Yuntao Bai and colleagues (December 2022). Anthropic published the constitution used for its Claude models in May 2023.