Reproduce “Self-Improvement Memory Enables Continual Recursive Self-Improvement” 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

Xin Yu, Fei Xia, Jiaqi Pan, Youbao Tang, Chengzhi Liu, Adams Yu, Sicong Jiang, Yanfei Chen and 2 more

The authors introduce SIM, a memory for harness optimizers that records edits, rationales, predicted and observed outcomes, and provenance while the task model stays fixed. The optimizer consults and updates it during continued optimization after task or model changes. In workspace transfer, Full SIM reaches 84.75 versus 73.42 without memory; these main comparisons are single runs per condition.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.