Reproduce “Agora: Enhancing LLM Agent Reasoning Via Auction-Based Task Allocation” 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
Kaiji Zhou, Aleš Leonardis, Yue Feng
Agora allocates reasoning units among language-model agents using confidence calibrated against past performance and a cost-aware auction. A planner decomposes requests, and the winning agents execute each unit before a composer combines outputs. On SPIQA’s strict L3≥0.8 measure, calibrated Agora scores 56.9%, versus 48.2% for Qwen and 46.9% without calibration; SPIQA provides the stage structure.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.