// ∞ Reasoning & Test-Time Compute
Self-Refine
One model drafts, critiques its own output, then revises – repeated a few times.
Self-Refine 自我反馈迭代 2023 Foundational core
IN PLAIN WORDS
No extra training or external reward: the model answers, is prompted to critique the answer, then rewrites using the feedback, until it is satisfied or hits a round limit. It beat one-shot generation on seven task types including code, writing and maths, and is the template for many agents’ “reflect” step.
Used in: Multi-round self-revision of code and writing; agents’ reflection and self-check steps
01Original paper
Self-Refine: Iterative Refinement with Self-Feedback
Aman Madaan, Niket Tandon, Prakhar Gupta, et al. · 2023
arXiv ↗ ✓ Verified · arXiv API
02Where it sits
Learn the left column first; the right column builds on this one.
Learn first
Then you can learn
- Nothing in the atlas builds on it yet
Related:ReflexionLLM-as-a-Judge
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
04Where the course teaches it
For AI / data
- Verification: verified, checked 2026-10-22.
- JSON: /api/learn/algorithms.json (id:
self-refine)