Groc-PO: Grounded Context Preference Optimization for Truthful Multimodal LLMs
多模态大模型真实性新方法:Groc-PO通过偏好优化约束上下文,有效减少幻觉,已被顶会ACM-MM接收。
arXiv:2607.13712v1 Announce Type: cross Abstract: Despite the rapid progress of Multimodal Large Language Models (MLLMs), they still suffer from untru…