Reproduce “Learning Multi-Agent Orchestration:Higher-Order Interactions in Policy 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

Kaixuan Liu, Guojun Xiong

IAPO improves orchestration of workers on long-input tasks by crediting individual decisions and pairwise interactions, then checking updates on fresh trajectories before accepting them. In an ablation on the authors’ constructed long-input MBPP task, it reached 81.1% accuracy at 5.1k inference tokens per problem, versus 63.8% at 15.7k for the first-order-only variant; this result is specific to that setup.

Summary by alphaXiv.

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