DAIR.AI推文
从智能体轨迹中提取自动机
该研究提出从智能体轨迹中提取紧凑的有限状态机（FSM），以分析智能体行为中模型与外部框架的贡献。在12个公开数据集上，诱导出的FSM仅含7至43个状态，对保留数据的重放适应度达0.997，且跨分割拓扑几乎一致，构建时间仅需毫秒。FSM状态上下文在下一步预测上优于Agent Workflow Memory，基于状态的故障预测AUROC最高达0.94，在线监控可提前停止失败运行。作者认为行为拓扑更多由部署框架而非底层LLM塑造。
译文
// 从智能体轨迹中提取自动机 // 你的智能体行为有多少来自模型本身，又有多少来自你包裹它的外部框架？这项新工作将整个智能体轨迹语料库压缩为单个紧凑的有限状态机。在十二个公开数据集上，所归纳出的状态机仅有7到43个状态，以0.997的适应度重放留出数据，且在不同数据划分下拓扑几乎一致，构建过程仅需毫秒级时间。在每一个与真实标签匹配的数据集上，FSM状态上下文在下一步预测任务中都优于Agent Workflow Memory。每个状态的行为特征在故障预测上达到了最高0.94的留出AUROC，同时一个在线监控器能够仅凭部分轨迹将失败运行排在通过运行之前，从而在完成之前提前触发早停。作者认为，行为拓扑更多是由部署框架塑造的，而非底层的LLM。论文：https://arxiv.org/abs/2608.23670 论文讨论：https://academy.dair.ai/papers/automata-from-agent-traces-2608.23670
// Automata from agent traces // How much of your agent's behavior comes from the model, and how much from the harness you wrapped around it? New work collapses an entire corpus of agent traces into a single compact finite-state machine. Across twelve public datasets the induced machines run 7 to 43 states, replay held-out data at 0.997 fitness with near-identical topology across splits, and build in milliseconds. FSM-state context beats Agent Workflow Memory on every ground-truth-matched dataset for next-step prediction. Per-state behavioral features reach held-out AUROC up to 0.94 for failure prediction, and an online monitor ranks failing runs above passing ones from a partial trace, triggering early stopping well before completion. The authors suggest that behavioral topology gets shaped more by the deployment harness than by the LLM underneath it. Paper: https://arxiv.org/abs/2608.23670 Chat with Paper: https://academy.dair.ai/papers/automata-from-agent-traces-2608.23670
