LLM智能体轨迹中的成本-效用对齐：剖析、归因、诊断、适应与评估
原标题：Cost-Utility Alignment in LLM Agent Trajectories:Profiling,Attribution,Diagnosis,Adaptation,and Evaluation
AI 摘要
本文提出一个面向LLM智能体轨迹的成本-效用对齐框架，包含成本剖析、效用归因、错位诊断、定向适应和评估五个阶段，将资源消耗与任务贡献视为同一执行过程的双重账本。效用归因通过过程代理、信息依赖和反事实重放等方法提供因果证据，以指导诊断和适应。该框架为资源感知的智能体设计与部署提供了结构化基础，并分析了近期系统、归因方法和评估协议。
正文节选
Cost–Utility Alignment in LLM Agent Trajectories: Profiling, Attribution, Diagnosis, Adaptation, and Evaluation Abstract. LLM agents execute tasks through multi-step trajectories that accumulate cost in tokens, latency, monetary fees, and environmental risk while producing utility only at the aggregate task level. Prior surveys address inference optimization, agent capabilities, or evaluation in isolation, leaving practitioners without principled tools to determine whether a trajectory’s resourc