elvis · @omarsar0 · X·2026-08-28 03:01·10小时前
AI 导读

新研究将智能体轨迹压缩为紧凑有限状态机,在12个公开数据集上仅用7-43个状态,以0.997的适应度重放留出数据,毫秒级构建。FSM状态上下文在下一步预测上全面优于Agent Workflow Memory,失败预测AUROC达0.94。作者认为行为拓扑更多由部署框架而非底层LLM塑造。

elvis@omarsar0
32AI 编辑部评分,满分 100
2026-08-28 03:01· 10小时前
AI 导读

新研究将智能体轨迹压缩为紧凑有限状态机,在12个公开数据集上仅用7-43个状态,以0.997的适应度重放留出数据,毫秒级构建。FSM状态上下文在下一步预测上全面优于Agent Workflow Memory,失败预测AUROC达0.94。作者认为行为拓扑更多由部署框架而非底层LLM塑造。

Another good paper. Interesting finding on the benefits of the agent harness.

DAIR.AI// 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...

来源:elvis· x.com