发表机构
KAIST（韩国科学技术院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究在 Hanabi 衍生环境中探究 LLM 与陌生伙伴的合作，发现模型能解码意图惯例但接收决策常不一致，动作建议比规则陈述更能提升合作，且激活转移有潜力但替代方案效果不稳定，凸显了从伙伴信息到行动转化的局限。
AI 中文摘要
与不熟悉的伙伴合作需要适应事先未知的沟通惯例。我们在一个受控的、源自 Hanabi 的环境中研究这一问题，该环境具有脚本化的提示生成、LLM 控制的接收决策和冻结的模型权重。在八个 LLM 中，线性探针恢复意图惯例的准确性显著高于恢复目标惯例，但接收选择并不始终与发送者的惯例一致。我们将探针预测的惯例和真实惯例分别以一般规则或外部计算的动作建议的形式呈现。规则陈述带来的合作变化适中且依赖于模型，而动作翻译平均带来更大的收益。在 Qwen3-8B 的案例研究中，匹配状态下的陈述反转显示，对动作建议的敏感性远高于对规则陈述的敏感性。来自 oracle 动作和非 oracle 提示复述捐赠者的激活转移提高了两类动作的意图准确性，但测试的替代方案未能可靠地重现这些收益。综合来看，这些结果区分了惯例可解码性、对惯例信息的敏感性和合作性能，并突出了将可用的伙伴信息转化为接收决策方面的局限性。
英文摘要
Cooperation with unfamiliar partners requires adapting to communication conventions that are not known in advance. We study this problem in a controlled Hanabi-derived environment with scripted hint generation, LLM-controlled receiving decisions, and frozen model weights. Across eight LLMs, linear probes recover intent conventions substantially more accurately than target conventions, yet receiving choices do not consistently agree with the sender's convention. We compare probe-predicted and ground-truth conventions presented either as general rules or as externally computed action recommendations. Rule statements yield modest and model-dependent changes in cooperation, whereas action translation produces larger gains on average. In a Qwen3-8B case study, matched-state statement reversals reveal much greater sensitivity to action recommendations than to rule statements. Activation transfers from oracle-action and non-oracle hint-restatement donors improve intent accuracy on both action classes, but the tested alternatives do not reliably reproduce these benefits. Together, these results distinguish convention decodability, sensitivity to convention information, and cooperative performance, and highlight limitations in turning available partner information into receiving decisions.
