Aligned but Not Partner-Specific: Distinguishing How Multimodal LLM Agents Succeed in Reference Games Without Human-Like Conventions 事件

Aligned but Not Partner-Specific: Distinguishing How Multimodal LLM Agents Succeed in Reference Games Without Human-Like Conventions arXiv:2606.08081v1 Announce Type: cross Abstract: Repeated reference games test whether interlocutors replace their initially long descriptions with shorter, partner-specific conventions grounded in shared interaction history. Prior work shows that multimodal LLMs fail to become more efficient across rounds, although they align on the labels they use. How can we d