Self-Correction

7 frameworks — Techniques that let AI critique, verify, and iteratively refine its own outputs.

Overview

Self-correction techniques enable AI to critique, verify, and iteratively refine its own outputs. Rather than accepting the first response, these methods build feedback loops that catch errors, verify facts, and improve quality through multiple rounds.

Use self-correction when accuracy matters more than speed — fact-checking, high-stakes content, or any task where errors have real consequences. These techniques are particularly powerful when combined with external tools for verification.

Technique Comparison

Side-by-side comparison of all 7 frameworks in this category.

Technique Year Best For Key Strength Complexity
Self-Refine 2023 Output quality Iterative improvement Medium
Self-Verification 2022 Answer accuracy Multi-solution check Medium
Chain-of-Verification 2023 Fact-checking Verification questions Medium
CRITIC 2023 Tool-augmented External verification High
Cumulative Reasoning 2023 Incremental building Proposition chain Medium
Self-Calibration 2022 Confidence assessment Reliability scoring Low
Reflexion 2023 Learning from failure Verbal self-feedback High
Recommended Starting Point

Start with Self-Refine

Self-Refine is the most intuitive self-correction technique — generate an output, critique it, then improve it. No external tools required, just iterative refinement until quality meets your standards.