Reproduce “Decentralized LLM Workflows via Sequential Auctions” 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

Ian Gemp, Shubhendu Trivedi, Adam Fisch, Fantine Huot, Michael Kaisers, Mirella Lapata, William W. Cohen, Alekh Agarwal and 2 more

The authors present a market for multi-step workflows: agents bid for task states, transform and pass them downstream, with payments rewarding earlier contributors. Learned bidding networks shade bids; underlying language models remain frozen. In synthetic translation, outcomes depend on competition density. In News Market, Mirror Descent achieves 73.40% success versus 48.40% for same-model ReAct, at higher reported cost.

Summary by alphaXiv.

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