English(EN) How Language Models Choose Sides: Internal Representations of Instruction Hierarchy

新研究揭示LLM内部冲突解决信号

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-01 04:00

研究人员调查了经过指令微调的大型语言模型如何处理用户和系统之间的冲突指令。他们开发了一个包含41个配对约束的基准测试，发现模型表现出三种不同的行为：尊重层级、反层级和无影响。例如，Llama-3.1-8B 经常忽略系统指令。研究还表明，Llama-3.1-8B 残差流中的内部信号能够准确预测冲突结果，这表明偏向用户的仲裁不一定源于无法检测到冲突。 AI

影响为理解LLM的决策过程提供了见解，可能有助于未来模型实现更好的控制和对齐。

排序理由学术论文，详细介绍了关于LLM行为的新研究发现。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 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，详细介绍了关于LLM行为的新研究发现。[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.AI TIER_1 English(EN) · Enrique Balp-Straffon, Chih-Hao Hsu, Rushiraj Gadhvi, Sunishchal Dev, Callum Stuart McDougall, Anusha Mujumdar · 2026-09-01 04:00

语言模型如何选择立场：指令层级的内部表征

arXiv:2608.28648v1 Announce Type: new Abstract: We study how instruction-tuned LLMs arbitrate direct conflicts between system and user instructions. We introduce a benchmark of 41 paired constraints with deterministic verifiers and evaluate eight models under matched baseline, co…

本文如何被排名

报道来源 [1]

语言模型如何选择立场：指令层级的内部表征

相关实体

相关话题