---
id: 20260314-T0-03
title: "DIVE论文提出多代理任务合成方法提升工具泛化能力"
title_en: "DIVE Paper Boosts Tool Use via Task Diversity Scaling"
url: https://ai.daily.yangsir.net/daily/20260314-T0-03
issue_date: 2026-03-14
publish_date: 2026-03-13T04:00:00.000Z
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2603.11076
---

# DIVE论文提出多代理任务合成方法提升工具泛化能力

arXiv论文《DIVE: Scaling Diversity in Agentic Task Synthesis》提出新方法解决LLM工具使用泛化问题。研究发现当前合成任务多样性不足导致模型在任务和工具集变化时表现脆弱。通过扩展任务多样性，模型能更好适应不同工具集，提升通用性。该研究为强化代理系统鲁棒性提供了新思路。

## English Version

**DIVE Paper Boosts Tool Use via Task Diversity Scaling**

The arXiv paper 'DIVE: Scaling Diversity in Agentic Task Synthesis' addresses LLM tool use brittleness. It identifies insufficient task diversity as the cause of poor generalization across different toolsets. By scaling diversity, the method improves adaptability, offering new solutions for robust agent systems.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2603.11076)

**详情页**：https://ai.daily.yangsir.net/daily/20260314-T0-03

---

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