Constitutional AI
An alignment approach where models self-critique and revise outputs against explicit policy principles
#Constitutional AI#CAI#policy-based alignment#self-critique
What is Constitutional AI?
Constitutional AI defines an explicit set of principles first, then trains models to critique and revise their own outputs against those principles.
Why does it matter?
Pure human review is hard to scale. Constitutional AI helps maintain consistent safety and policy behavior in larger training pipelines.
Practical point
If policy documents are vague, behavior becomes unstable. Teams should make prohibitions, priorities, and exception rules explicit.