返回全部动态
LLM能否超越纳什均衡？自博弈多智能体游戏中的去中心化协调测试
原标题：Do LLMs Beat Nash? Testing Decentralized Coordination in Self-Play Multi-Agent Games
AI 摘要
麦吉尔大学的研究人员提出一个基准测试，评估大语言模型在无通信、单轮博弈中的协调能力，并与纳什均衡对比。结果显示，两个前沿托管模型在双人矩阵游戏中持续超过纳什基准，接近最优联合结果，而多数开源模型仅获得部分收益，且在四人及以上团队游戏中性能显著下降。该研究为多智能体系统的去中心化协调提供了新见解。
以上摘要由 AI 生成，可能存在误差。事实请以原文为准。
正文节选
Do LLMs Beat Nash? Testing Decentralized Coordination in Self-Play Multi-Agent GamesD. Sinishaw is with the Department of Electrical and Computer Engineering, McGill University (e-mail: deborah.sinishaw@mail.mcgill.ca). Q. Zhu and E. Meriaux are with the School of Computer Science, McGill University (e-mail: qile.zhu@mail.mcgill.ca; edwin.meriaux@mail.mcgill.ca). All authors are affiliated with the Centre for Intelligent Machines (CIM), McGill University, and are supervised by G. Dudek (e-mail:
发布时间：2026-08-14 12:00
抓取时间：2026-08-14 12:38
来源机构：arXiv