---
id: 20260630-T0-15
title: "Supersede：LLM Agent在多轮对话中难以及时“更新记忆”"
title_en: "Supersede: LLM Agents Struggle to Update Memory in Multi-Turn Interactions"
url: https://ai.daily.yangsir.net/daily/20260630-T0-15
issue_date: 2026-06-30
publish_date: 2026-06-29T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2606.27472
---

# Supersede：LLM Agent在多轮对话中难以及时“更新记忆”

arXiv:2606.27472v1 发布。研究揭示了LLM智能体在长周期、多会话交互中存在的“记忆更新断层”问题：当事实发生变化（如用户搬家、价格更新）时，智能体往往无法正确使用新值并丢弃旧值。论文提出了名为Supersede的框架来诊断和训练这一能力，旨在提升Agent处理动态信息更新的可靠性。

## English Version

**Supersede: LLM Agents Struggle to Update Memory in Multi-Turn Interactions**

arXiv:2606.27472v1 released. Research reveals a 'memory-update gap' in LLM agents during long, multi-session interactions where facts change (e.g., user moves, price updates). Agents often fail to use current values and discard obsolete ones. The paper proposes a framework called Supersede to diagnose and train this capability, aiming to improve the reliability of agents in handling dynamic information updates.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2606.27472)

**详情页**：https://ai.daily.yangsir.net/daily/20260630-T0-15

---

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