---
id: 20260604-T0-05
title: "大模型推理过度有害？新研究评估长推理风险"
title_en: "Large Reasoning Models May Harm from Overthinking"
url: https://ai.daily.yangsir.net/daily/20260604-T0-05
issue_date: 2026-06-04
publish_date: 2026-06-03T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2606.02835
---

# 大模型推理过度有害？新研究评估长推理风险

arXiv论文发现，大型推理模型(LRMs)通过增加测试时计算来生成显式中间推理轨迹，但更长推理并不总是有益的。研究表明，当前模型可能因过度推理而产生有害结果，建议重新评估推理长度与性能的关系。该研究对LRM的部署策略有直接影响。

## English Version

**Large Reasoning Models May Harm from Overthinking**

An arXiv paper finds that while Large Reasoning Models (LRMs) generate explicit intermediate traces through increased compute, longer reasoning isn't always beneficial. Research suggests LRMs may produce harmful results from overthinking, directly impacting deployment strategies for these models.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2606.02835)

**详情页**：https://ai.daily.yangsir.net/daily/20260604-T0-05

---

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