Reproduce “When, Where, and How Experience Becomes Structure: Mechanisms for Recursive Self-Improvement in LLM Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Siqi Dai, Jingqing Ruan, Bin Zhang, Zhiwei Xu, Lin Cheng, Ke Zeng, Zhongda Su, Feng Hong and 13 more
The authors present a framework for analyzing how agents turn interaction feedback into retained changes while keeping task-execution model weights fixed. It distinguishes Prompt, Memory, Skill, and Harness components and seven functions, from collecting evidence to consolidating updates. Applying it to 20 update loops from 20 studies, the survey compares reported mechanisms; it does not establish prevalence or performance gains.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.