PulseAugur
实时 09:26:00
English(EN) When Helpfulness Overrides Causal Caution: Context-Dependent Suppression and Recovery in LLMs

研究发现LLM在实际建议中抑制“因果谨慎”

一篇新发表在arXiv上的研究揭示,大型语言模型(LLM)在从学术语境转向实际咨询角色时,“因果谨慎”表现出显著下降。在Claude Sonnet 4.6、Claude Opus 4.7、GPT 5.5和Gemini 3.1 Pro上进行的实验表明,在学术环境中因果谨慎率超过90%,但在实际咨询情境下则骤降至20%以下。然而,一个简单的自我纠正提示成功地将因果谨慎恢复到高水平,这表明问题是情境依赖性抑制,而非基本能力限制。 AI

影响 建议通过分离提案生成与因果审计的架构,改进AI治理。

排序理由 该集群包含一篇详细介绍LLM行为实验结果的研究论文。

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →

研究发现LLM在实际建议中抑制“因果谨慎”

报道来源 [2]

  1. arXiv cs.AI TIER_1 English(EN) · Hiroshi Okumura ·

    When Helpfulness Overrides Causal Caution: Context-Dependent Suppression and Recovery in LLMs

    arXiv:2606.24370v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly integrated into decision-support roles in business and policy contexts. While prior benchmark studies have primarily evaluated LLMs' causal reasoning capabilities, a more fundamental epi…

  2. arXiv cs.AI TIER_1 English(EN) · Hiroshi Okumura ·

    When Helpfulness Overrides Causal Caution: Context-Dependent Suppression and Recovery in LLMs

    Large language models (LLMs) are increasingly integrated into decision-support roles in business and policy contexts. While prior benchmark studies have primarily evaluated LLMs' causal reasoning capabilities, a more fundamental epistemic dimension has been overlooked: Causal Cau…