Research map / Large language models / Agents, code and math
Self-Refine: Iterative Refinement with Self-Feedback
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Self-Refine iterative self-feedback | GPT-4 · Sentiment Reversal | task score: 3.8 -> 36.2 | Helped |
| add Self-Refine iterative self-feedback | GPT-4 · Dialogue Response | task score: 25.4 -> 74.6 | Helped |
| add Self-Refine iterative self-feedback | GPT-4 · Code Optimization | task score: 27.3 -> 36.0 | Helped |
| add Self-Refine iterative self-feedback | GPT-4 · Code Readability | task score: 27.4 -> 56.2 | Helped |
| add Self-Refine iterative self-feedback | GPT-4 · Math Reasoning | task score: 92.9 -> 93.1 | No clear effect |
| add Self-Refine iterative self-feedback | GPT-4 · Acronym Generation | task score: 30.4 -> 56.0 | Helped |
| add Self-Refine iterative self-feedback | GPT-4 · Constrained Generation | task score: 15.0 -> 45.0 | Helped |
| use specific actionable feedback instead of generic feedback (ablation) | Self-Refine · Code Optimization | task score: 26.0 -> 27.5 | Helped |
| use specific actionable feedback instead of generic feedback (ablation) | Self-Refine · Sentiment Reversal | task score: 31.2 -> 43.2 | Helped |
| use specific actionable feedback instead of generic feedback (ablation) | Self-Refine · Acronym Generation | task score: 54.0 -> 56.4 | Helped |
| increase feedback-refine iterations from 0 to 3 (ablation) | Self-Refine · Constrained Generation | task score: 29.0 -> 49.7 | Helped |
| use Vicuna-13B as base model (ablation) | Self-Refine · various tasks | refinement quality | Hurt |
Closest papers by meaning
- Toward Self-Improvement of LLMs via Imagination, Searching, and Criticizing (NeurIPS 2024 · similarity 0.91)
- ReVISE: Learning to Refine at Test-Time via Intrinsic Self-Verification (ICML 2025 · similarity 0.90)
- Large Language Models Cannot Self-Correct Reasoning Yet (ICLR 2024 · similarity 0.89)
- Self-Improvement in Language Models: The Sharpening Mechanism (ICLR 2025 · similarity 0.89)
- Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models (ICLR 2025 · similarity 0.89)
- ReGenesis: LLMs can Grow into Reasoning Generalists via Self-Improvement (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.