Reproduce “MetaExp: Learning to Self-Improve through Experience via Verifiable Cross-Task Environment Scaling” 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
Renhao Li, Xinjie Shen, Yang Su, Jianhong Tu, Derek F. Wong, Dayiheng Liu, Min Yang
MetaExp trains agents to turn feedback from related, varied tasks into reusable skills, using verified task families and trajectories to supervise task execution and skill updates. In cold-start tests, model parameters remain fixed while an empty skill bank accumulates experience. On SpreadsheetBench, success rises from 65.0% to 73.5% after 40 tasks; evidence supports transfer only across tested held-out benchmarks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.