AI RESEARCH
Steering the Verifiability of Multimodal AI Hallucinations
arXiv CS.LG
•
ArXi:2604.06714v1 Announce Type: cross AI applications driven by multimodal large language models (MLLMs) are prone to hallucinations and pose considerable risks to human users. Crucially, such hallucinations are not equally problematic: some hallucination contents could be detected by human users(i.e., obvious hallucinations), while others are often missed or require verification effort(i.e., elusive hallucinations). This indicates that multimodal AI hallucinations vary significantly in their verifiability.