3

Mitigating Agreement Bias in LLMs via Contradictory-Knowledge Prompt Triads and Reinforcement Training
Measures and mitigates sycophancy in LLMs using contradictory-knowledge prompt triads and reinforcement training, building on a post-training-stage audit of open-recipe models. In preparation for ACM FAccT.
Chao Péter Yang
FinCom: A Financial Multi-Agent Demo with Disagree-or-Commit Deliberation
A governed multi-agent framework that embeds structured dissent into financial AI committees via the Disagree-or-Commit protocol, improving reasoning accuracy and risk awareness over consensus-seeking baselines. Submitted to TMLR.
Chao Péter Yang, Zixiao Tan, Kaisen Yao, Ziyu Zhou, Eleanor Jiang, Michael Wu
PDF Source Document
SchenkerLink: Human-in-the-Loop Hierarchical Music Analysis with Uncertainty-Aware Graph Link Prediction
SchenkerLink introduces an uncertainty-aware, human-in-the-loop framework for hierarchical music analysis with graph link prediction, supporting interpretable, collaborative analysis. Under review for KDD 2027.
Stephen Ni-Hahn, Jerry Zhu, Chao Péter Yang, Cynthia Rudin, Simon Mak, Yue Jiang