English(EN) When LLMs Forget What They Believe: Characterizing State Integrity Collapse in Long-Horizon Dialogues

新研究定义了大型语言模型在长对话中的“状态完整性崩溃”

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-23 04:00

一篇新研究论文介绍了大型语言模型中“状态完整性崩溃”的概念，描述了一种现象：模型会记住指令，但在长对话中停止应用该指令。该研究提出了一个名为 StateTrack-Bench 的基准来评估各种约束类型下的这一问题，包括数值、格式、事实和逻辑规则。研究结果表明，相当一部分失败源于行为漂移而非简单的遗忘，而一种称为结构化状态锚定（Structured State Anchoring）的技术可以提高约束依从性。 AI

影响强调了大型语言模型在扩展交互中的可靠性方面的一个关键限制，需要超越单轮合规性的新评估方法。

排序理由该集群包含一篇详细介绍大型语言模型新现象和基准的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 4 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍大型语言模型新现象和基准的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Same-day; Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Yancheng Zhu, Yifan Yuan · 2026-10-23 04:00

When LLMs Forget What They Believe: Characterizing State Integrity Collapse in Long-Horizon Dialogues

arXiv:2610.31314v1 Announce Type: new Abstract: Large language models can repeat an instruction while failing to keep obeying it. We study this recall-compliance dissociation in long-horizon dialogue and define State Integrity Collapse: a model preserves a stated constraint in me…

本文如何被排名

报道来源 [1]

When LLMs Forget What They Believe: Characterizing State Integrity Collapse in Long-Horizon Dialogues

相关话题