#large language models
Jun 2026
Detecting Clinical Hallucinations in LVLMs via Counterfactual Visual Grounding Uncertainty
A vision-traceable hallucination detection framework that audits arbitrary LVLM responses via visual evidence grounding, requiring neither modification nor internal access to the hidden states of LVLMs, while providing interpretable localization evidence and strong cross-model transferability.
Xiao Song, Haonan Qin, Zhaoxu Zhang et al.
· arXiv.org · 0 citations