Reproduce “Multi-Agent Memory Inconsistency: Where Factual Conflicts Disappear in LLM Agents” 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

Hejia Zhang, Zhongming Yu, Yimin Ding, Wei-Ting Chen, Hengjia Yu, Wentao Ni, Yijun Liu, Cong Xu and 4 more

The authors evaluate how multi-agent document review handles planted numerical conflicts across 73 conflicts in 11 Harvey LAB tasks. They track whether both conflicting claims remain available from source reads through worker reports and lead evidence to the final report. Mandatory delegation raises source-pair recall from 50.2% to 68.7%, but lowers final detection from 16.2% to 13.2% in this evaluation.

Summary by alphaXiv.

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