Self-Reflection / Self-Verification

Self-reflection enables models to critique their answers, detect inconsistencies, and improve solutions. The agent generates an initial response, evaluates it against rules or examples, then produces a refined answer. This reduces hallucinations and increases reliability. Techniques include critique-and-revise loops, verifier models, multi-pass reasoning, and self-evaluation prompts.

Initial Output → Critique Module → Revised Output → Validator

Verifier model, evaluator prompts, error-checking rules, reasoning paths