实时时钟下的智能体评估与教学：RealtimeBench 与 TORSI
原标题：The World Won't Wait: Evaluating and Teaching Agents Under a Live Clock
AI 摘要
研究者提出 RealtimeBench 基准，包含六类游戏共 180 个任务，在模型推理期间游戏持续运行，以评估 LLM 智能体的实时决策能力。配对实验显示，GPT-6.1-sol 在游戏暂停时优于更快的 Jev，但游戏持续运行时反而落后（平均分 1.51 对 13.51）。为此提出 TORSI 方法：慢速教师模型在暂停环境中研究失败并生成可复用指令，快速学生模型在游戏中无需调用教师即可执行，两轮教学将 Jev 平均分从 13.51 提升至 24.19，并改善全部六类游戏。
正文节选
The World Won’t Wait: Evaluating and Teaching Agents Under a Live Clock Abstract Large language model (LLM) agents can solve complex interactive tasks, yet reasoning and tool calls take time. In continuously changing environments, an otherwise correct action may arrive too late. We introduce RealtimeBench, with 180 tasks across six game types, to evaluate agents while games continue running during inference. Paired evaluations show that GPT-6.1-sol outperforms the faster Jev when games are pause