MoHoBench:通过无法回答的视觉问题评估多模式大型语言模型的诚实性.pptx

MoHoBench:通过无法回答的视觉问题评估多模式大型语言模型的诚实性.pptx

Abstract

RecentlyMultimodalLargeLanguageModels(MLLMs)haveachievedconsiderableadvance-mentsinvision-languagetasks,yetproducepotentiallyharmfuloruntrustworthycontent.

Despitesubstantialworkinvestigatingthetrust-worthinessoflanguagemodels,MMLMs’ca-

文档评论(0)

1亿VIP精品文档

相关文档