作者位次 #4/4ACL 2026Poster· Natural Language Processing
When Seeing Overrides Knowing: Disentangling Knowledge Conflicts in Vision-Language Models
Francesco Ortu · Zhijing Jin · Diego Doimo · Alberto Cazzaniga
摘要
Vision-language models (VLMs) increasingly combine visual and textual information to perform complex tasks. However, conflicts between their internal knowledge and external visual input can lead to hallucinations and unreliable predictions. In this work, we investigate the mechanisms that VLMs us...