Skip to content

Source paper

Self-Improvement Memory Enables Continual Recursive Self-Improvement

Xin Yu, Fei Xia, Jiaqi Pan and 7 more

Xin Yu, Fei Xia, Jiaqi Pan, Youbao Tang, Chengzhi Liu, Adams Yu, Sicong Jiang, Yanfei Chen, Lingzhou Xue, Ting Yu

arXiv 2611.18856 · 202631 upvotes on Hugging Face

Research ideas from this paper (3)

Generated by AI. The paper’s authors did not write these.

AI summary

This paper introduces Continual Recursive Self-Improvement (CRSI), a framework for continuing LLM agent harness optimization across changes in task benchmarks or underlying policy models. To prevent loss of context and redundant edits across transitions, the authors propose Self-Improvement Memory (SIM), which records harness diagnoses, attempted code modifications, rationales, and verified outcomes. Across coding, text classification, agentic workspace, and engineering benchmarks, maintaining SIM enables higher downstream performance and more effective reuse and revision of inherited harness components compared to harness-only transfer.

How it works (AI summary)

The system embeds a structured memory mechanism (SIM) into an agentic outer loop comprising an LLM proposer, evaluator, and selector optimizing an executable harness codebase. Rather than discarding development history across domain or model shifts, SIM organizes past edits hierarchically by root-cause hypotheses, tracking predicted versus observed test outcomes and execution provenance. The proposer and selector ingest this persistent log to decide whether to retain, revise, replace, or prune inherited harness components without modifying frozen policy model weights.