让智能体学会发现用户没说出口的需求，自动判断是否该打扰对方。
AURA: Intent-Directed Probing for Implicit-Need Surfacing in Situated LLM Agents
- 在感知场景后增加推断步骤，生成带评分的隐性需求结构
- 隐性需求覆盖提升7%，探针调用减少82%且无违规操作
- 适合需要隐私敏感决策的对话系统和智能助手
一个情境化查询如“林伟在哪？”往往隐含更多意图：用户可能还想知道林伟是否空闲、心情如何或是否值得现在打扰。传统工具使用型智能体仅回答字面问题即停止。AURA 在场景感知与工具调用之间插入推理步骤，生成意图框架（IntentFrame）——一种结构化的隐性需求估计，并通过标量间隙得分控制每查询的探测预算与工具选择。在包含100个查询、4个场景的隐性意图基准测试中，相较于ReAct式探测，AURA的隐性需求覆盖提升0.07（p < 10^-6），其中三个场景单独显著；该增益在另一骨干模型上复现，提示消融实验表明性能提升源于间隙校准而非答案记忆。在事实查询任务中，控制器以牺牲部分准确率为代价，实现82%更少的探测次数，并在隐私敏感片段中零违规工具调用。具体作用范围见局限性部分。代码、模拟器与基准数据集已开源至https://github.com/innovation64/AURA。
原文摘要 · Abstract (English)
A situated query like "where is Lin Wei?" often encodes more than its literal content: the user may also want to know whether Lin Wei is free, in a good mood, or worth interrupting now. Standard tool-use agents answer the literal question and stop. AURA inserts an inference step between scene perception and tool use that produces an IntentFrame: a structured estimate of the implicit need with a scalar gap score that controls per-query probe budget and tool selection. On a 100-query four-scene implicit-intent benchmark, AURA improves implicit-need coverage over ReAct-style probing (Delta = +0.07, p < 10^-6); three of four scenes are individually significant, the gain reproduces on a second backbone, and a prompt ablation attributes the lift to gap calibration rather than answer memorisation. On factual lookup the controller trades raw accuracy for 82% fewer probes and zero forbidden-tool violations on a privacy-sensitive slice; scope conditions are detailed in Limitations. Code, simulator, and benchmark are released at https://github.com/innovation64/AURA.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。