Tag
提出了 SMART-Med，一种完全本地化的多智能体医学推理方法，通过不确定性驱动的路由机制决定何时单智能体作答、何时召集专家协作，并在 MedVR-3K 数据集上微调，以 3.8 倍推理加速和 72% token 节省达到与稠密多智能体方法相当的性能。
A controlled ablation study of LLM self-reflection in conflict forecasting finds that typed action routing drives performance gains, while diagnostic scaffolding and taxonomy vocabulary add no measurable value, with replication on GPT-4o.