---
id: 20260619-T0-09
title: "研究提出通过Agent轨迹分析来诊断模型行为缺陷"
title_en: "Diagnosing Model Behavior via Agent Trajectories: Bridging the Model-Harness Gap"
url: https://ai.daily.yangsir.net/daily/20260619-T0-09
issue_date: 2026-06-19
publish_date: 2026-06-18T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2606.17454
---

# 研究提出通过Agent轨迹分析来诊断模型行为缺陷

arXiv上新发表论文指出，AI Agent的表现不仅是模型问题，更是系统问题。研究发现，模型假设与执行框架之间的差异往往导致性能瓶颈。该研究提出通过分析Agent的运行轨迹来解构模型行为，从而识别出单纯的模型评估无法发现的系统性缺陷。这种方法强调了在优化底层模型的同时，优化Agent架构的重要性。

## English Version

**Diagnosing Model Behavior via Agent Trajectories: Bridging the Model-Harness Gap**

A new arXiv paper posits that AI agent performance is fundamentally a systems problem, not just a modeling one. It highlights how gaps between model assumptions and harness behavior limit capabilities. The researchers propose using agent trajectories to dissect model behavior, identifying systemic blind spots that standard evaluations miss.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2606.17454)

**详情页**：https://ai.daily.yangsir.net/daily/20260619-T0-09

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*