Reproduce “When Debate Helps: Proposal Supply and Verification-Aware Readout in Multi-Agent Reasoning” 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
Zihao Zhao, Tunyu Zhang, Haizhou Shi, Yusong Zhao, Xinxi Zhang, Hao Wang
The authors study when multi-agent debate can recover correct answers missed by majority voting. They define recoverable headroom as cases where an agent proposed the correct answer but the initial vote was wrong, and test complementary agent selection and evidence-aware readout. On OLMo3-7B-Instruct, AC-Greedy improves mean accuracy by 2.01 percentage points over Vanilla Debate across eight columns, in these settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.