Constitutional AI
Constitutional AI is a training approach developed by Anthropic in which a model is guided by a written set of principles (a constitution) and uses AI feedback to critique and improve its own responses.
What is Constitutional AI?
It reduces reliance on large amounts of human labelling of harmful content and makes the values being trained more explicit and transparent.
How it works
The model critiques and revises its outputs against the principles, and an AI-generated preference signal based on them is used for reinforcement learning.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions