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.
Self-Correction 7
Self-Refine
2023Iteratively generate, critique, and refine outputs without external input.
Still active techniqueSelf-Verification
2022Generate multiple solutions and verify each against the problem.
Still active techniqueChain-of-Verification
2023Plan, execute, and verify a series of fact-checking questions.
Still active techniqueCRITIC
2023Use external tools to verify and correct AI-generated content.
Still active techniqueCumulative Reasoning
2023Build answers incrementally through iterative propositions.
Still active techniqueSelf-Calibration
2022Assess confidence in answers to determine reliability.
Still active techniqueReflexion
2023Reflect on past failures through verbal self-feedback for improvement.
Still active techniqueSTaR
2022Bootstrap reasoning ability by iteratively generating and fine-tuning on successful reasoning chains.
Still active techniqueQuiet-STaR
2024Train models to generate internal reasoning at every token for improved prediction without explicit prompts.
Still active techniqueVerify-and-Edit
2023Verify each reasoning step for errors, then edit the chain to fix mistakes before the final answer.
Still active techniqueProgressive-Hint
2023Use the model’s previous answers as progressive hints to iteratively refine toward the correct solution.
Still active techniqueTechnique 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 |
Related Categories
Explore other framework categories that complement Self-Correction.