Reproduce “Consilience: Conformally Calibrated Communication Control for Hidden-Profile 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
Abhijith Babu, Ramneet Kaur, Vishal Pramanik, Olivera Kotevska, Nathaniel D. Bastian, Susmit Jha, Sunny Raj, Yanzhao Wu and 2 more
Consilience adaptively controls multi-agent discussion by reading agents’ beliefs and the transcript, selecting and screening interventions, speakers, and stopping decisions with round-specific conformal calibration of one-step regret. In HiddenBench, rule-based Consilience averaged 0.83 task accuracy across 12 open-weight models, versus 0.26 for fixed round-robin. Its best policy matched or exceeded Full Info on 11 of 12 models, except Qwen3-0.6B.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.