廉价谈话可稳定 LLM 智能体的策略互动
原标题：Cheap Talk Stabilizes Strategic Interaction in LLM Agents
AI 摘要
东北大学 MAGICS 实验室研究 LLM 智能体在重复博弈中的行为稳定性，测试四个 7–9B 开源模型在囚徒困境、雪堆、猎鹿和和谐博弈中的表现。结果显示，智能体生成的「廉价谈话」（非约束性预沟通）在 96 个模型-博弈-情境组合中有 71 个显著提升了动作持续性，仅 5 个在社会或团队框架下出现反转。研究进一步在 Qwen 中识别出降低动作不确定性和减少轮间漂移两个输出层通道，并在晚期 Transformer 层发现一个历史平衡的政策-内容方向，投影剔除该方向会增加闭环博弈中的策略切换。
正文节选
Cheap Talk Stabilizes Strategic Interaction in LLM Agents Abstract Large language models are increasingly deployed as interacting agents, making the persistence of their action policies across repeated interaction critical for reliable multi-agent operation. We investigate whether and how agent-generated, non-binding pre-play communication (“cheap talk”) increases such persistence in four open-weight 7–9B-parameter LLMs. Our experiments span four repeated two-player games—Prisoner’s Dilemma, Sno