Constitutional AI

Anthropic's technique for self-supervised alignment — the AI critiques its own outputs against a written "constitution".

Definition

Constitutional AI (Bai et al. 2022) trains models on principles ("be helpful, harmless, honest") by having the AI rewrite its own outputs to satisfy them. Reduces but doesn't eliminate the need for human raters.

Examples

  • Used in Claude's training pipeline

Related terms

Looking for actual gigs that use Constitutional AI? Browse all live listings or filter by $100+/hr.