Model Adaptation · Software component
Critique-Revision Generator
Software componentModel AdaptationModelsarc:CritiqueRevisionGenerator
A training-data generator that has a model critique its own response against a randomly sampled constitutional principle, then revise the response to remove identified violations.
Responsibility. Produces principle-aligned revised responses for supervised fine-tuning.
Also known as: Supervised self-critique (SL-CAI), Self-critique-and-revision cycle
Relationships
is configured by structural
- Constitution abstract Ch9.5
- Critique Rubric Ch9.5
invokes dependency
reads dependency
is orchestrated by control
produces lifecycle
Design guidance
- SHOULD sample a principle per critique so training covers the whole constitution.
- SHOULD iterate critique and revision so the model learns to generate aligned responses from the outset rather than via post-hoc correction.
Classification
- Patterns
- Critique-revise loopRandom principle samplingSelf-critique
- Quality attributes
- Performance efficiency (ISO/IEC 25010)Transparency and accountability (NIST AI RMF: accountable and transparent)
- Risks mitigated
- Human annotator exposure to disturbing contentPoor annotation scalability
Sources
- Ch9.5: T. Nguyen, "Constitutional AI," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 9.5. ISBN: 9798244538229.