Constitutional AI
Constitutional AI: An alignment approach that uses an explicit set of principles to guide response revision and learning. The model critiques and revises outputs according to a supplied constitution, reducing reliance on direct human labels for every case. Choosing and interpreting principles remains a significant human decision.
In italiano: Constitutional AILegge III · Come una rete impara dai datiThe model critiques and revises outputs according to a supplied constitution, reducing reliance on direct human labels for every case. Choosing and interpreting principles remains a significant human decision.
Examples
- Non-discrimination principles
- Self-revision of risky responses
- Feedback generated from written rules