English(EN) When Context Misleads: Surprisal, Energy and Attention Entropy as Metrics of Coherence Illusions in LLMs

研究发现：大型语言模型会表现出与人类读者相似的“连贯性幻觉”

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-14 04:00

一篇新的研究论文探讨了大型语言模型（LLMs）中的“连贯性幻觉”，并将其与人类阅读行为进行类比。研究发现，大型语言模型和人类一样，可能会被误导，认为不连贯的文本是连贯的，尤其是在前面的上下文中存在与预期延续相匹配的干扰项时。研究人员使用惊奇度、注意力熵和一个新颖的能量指标来识别和量化这些幻觉，表明其与人类认知存在共享的潜在机制。 AI

影响揭示了大型语言模型理解能力中潜在的脆弱性，表明需要更鲁棒的连贯性评估方法。

排序理由学术论文发表在arXiv上，详细介绍了关于大型语言模型行为的新发现。[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: 36 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 学术论文发表在arXiv上，详细介绍了关于大型语言模型行为的新发现。[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: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Ece Takmaz, Nitin Kumar, Li Kloostra, Jakub Dotlacil · 2026-09-14 04:00

当语境误导时：惊奇度、能量和注意力熵作为LLM连贯性幻觉的度量

arXiv:2606.21203v2 Announce Type: replace Abstract: Psycholinguistics studies show that human readers fall for coherence illusions: an incoherent discourse can seem coherent simply because a distractor matches what comes next. We investigate whether Dutch language models (6 monol…

本文如何被排名

报道来源 [1]

当语境误导时：惊奇度、能量和注意力熵作为LLM连贯性幻觉的度量

相关实体

相关话题