Mitigating reasoning hallucination through Multi-agent Collaborative Filtering

Jinxin Shi, Jiabao Zhao, Xingjiao Wu, Ruyi Xu, Yuan-Hao Jiang, Liang He

Abstract: Highlights•Specifically solved the hallucination problem in model inference.•Systematically described the phenomenon of unstable response quality.•Proposed methods to improve the stable output of high-quality responses.•Proposed an automated approach for constructing experience repositories.•Extensive experiments demonstrate the effectiveness of this research.