研究：智能体规则层与LLM评审门控并非独立失效

1. 12:00
   arXiv · Artificial Intelligence

   评测整体堆栈而非单层：智能体动作的确定性门与 LLM 门会独立失效吗？

   研究在三个语料库的 1,119 个已标注智能体动作上，检验运行时门控堆栈“错误相乘”的假设，堆栈含一层确定性规则与四个 LLM 评审。STRICT 定义下任意两个评审组合约为 1.2 至 1.4 层，规则层加一个评审为 1.86 至 2.09 层；评审耦合的难度占比在 31.8% 至 61.8% 间不可识别。

Current heat 9·Comparable peak 10（Oct 7）·Comparable change over 24 hours –

The trend compares only the same participants observed continuously; its range may be smaller than the current heat count. Move or click on the chart to inspect hourly heat; use the left and right arrow keys to switch.