Reproduce “Recursa: Toward Recursively Self-Improving Foundation Model Systems” 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
Erkang Zhu, Yin Lin, Xiaorong Shi, Xuanrui Lin, Jingren Zhou
Recursa adapts a text-based repository of instructions, skills, and memories to address knowledge and behavior gaps in domains. It bootstraps from domain materials, then uses task attempts, traces, and feedback to revise and validate reusable guidance through single or multiple learners, without changing model weights. On Harvey LAB’s held-out tasks, DeepSeek-V4.1-Flash all-pass rises from 30.0% to 55.0% with multi-learner learning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.