论文

多个 AI 智能体因共享同一模型而能互相理解非常规通信

精选理由

Ethan Mollick 提出一个有意思的观察:同一模型驱动的智能体之间容易"串通",隔离比想象中难。

Ethan Mollick 在 X 上分享的观察显示,由同一模型驱动的多个智能体之间容易互相识别对方生成的创造性通信方案,原因在于它们共享同一个底层"大脑"。这解释了为什么把不同智能体彼此隔离变得困难。Mollick 还提出疑问:这种互通能力是否适用于不同模型家族之间。

原文 · Ethan Mollick

This is both neat & a sign of why it is proving to be so hard to isolate agents from each other. They share a common "brain" which means they don't have a lot of trouble recognizing creative communication solutions generated by other AIs

Wonder if it works across model families?