作者位次 #3/4ACL 2026Poster· Natural Language Processing
M agic B ench: Diagnosing Visual Agency Loss and Semantic Dependency in Multimodal LLM s
Tang Da Huang · Weidong Tang · Wen Qi Xu · Xianpeng Guo
摘要
Multimodal Large Language Models typically assume linguistic context invariably enhances visual understanding. We study this assumption in semantic adversarial scenarios, specifically magic tricks, where narration deliberately diverges from physical reality. We introduce MagicBench, a diagnostic...