AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

A Stitch in Time Saves Nine: Proactive Self-Refinement for Language Models

Jinyi Han, Xinyi Wang, Haiquan Zhao, tingyun li, Zishang Jiang, Sihang Jiang, Jiaqing Liang, Xin Alex Lin, Weikang Zhou, Zeye Sun, Fei Yu, Yanghua Xiao

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Parallel and early-exit decoding · 4 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (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 changedWhereMeasuredEffect
add RL-trained proactive self-refinement (PASR) over vanilla decodingQwen2.5-7B · 10-task averageaverage score: 55.9 -> 61.7Improved
add RL-trained proactive self-refinement (PASR) over vanilla decodingQwen3-8B · 10-task averageaverage score: 60.9 -> 69.1Improved
add reinforcement learning rather than prompting for in-generation self-refinement (ablation)Qwen2.5-7B · 10-task averageaverage score: 40.2 -> 61.7Improved
add reinforcement learning rather than prompting for in-generation self-refinement (ablation)Qwen3-8B · 10-task averageaverage score: 51.4 -> 69.1Improved
add reinforcement learning rather than instruction fine-tuning (ablation)Qwen2.5-7B · 10-task averageaverage score: 57.0 -> 61.7Improved
add reinforcement learning rather than instruction fine-tuning (ablation)Qwen3-8B · 10-task averageaverage score: 52.6 -> 69.1Improved
add multi-answer comparison reward (ablation)PASR (Qwen2.5-7B) · 10-task averageaverage score: 57.2 -> 61.7Improved
add comparison-based refinement reward (ablation)PASR (Qwen2.5-7B) · 10-task averageaverage score: 56.2 -> 61.7Improved
add comparison-based refinement reward (ablation)PASR (Qwen2.5-7B) · Winoaccuracy: 57.0 -> 65.3Hurt
add multi-answer comparison reward (ablation)PASR (Qwen2.5-7B) · GSM8Kaccuracy: 75.7 -> 88.8Improved
add comparison-based reward (ablation)PASR (Qwen2.5-7B) · AIME24accuracy: 10.0 -> 10.0No clear effect
add oracle feedback to Self-RefineQwen2.5-7B · 10-task averageaverage score: 57.5 -> 62.3Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy