视觉工具使用的幻觉：对图像思维模式的因果审计

AI 摘要

该研究对多模态大语言模型的视觉工具使用进行了因果审计，发现尽管聚合准确率有所提升，但返回的视觉证据往往对答案没有因果影响，或调用计划不连贯。研究提出了视觉证据增益等干预指标，并在六个模型和五个基准上揭示了两种失败模式，将这种不一致称为视觉工具使用的幻觉。

正文节选

The Illusion of Visual Tool-Use: A Causal Audit of Thinking with Images Abstract Visual tool-use in multimodal LLMs often lacks causal effectiveness, with returned observations frequently failing to influence answers or being used incoherently despite aggregate accuracy improvements. The "thinking-with-images" paradigm equips multimodal LLMs with active visual operations such as crop-and-zoom. However, models using these operations often achieve only marginal or negative gains over direct infere

发布时间：—

抓取时间：2026-08-13 12:01

来源机构：Hugging Face