Where is the multimodal goal post? On the Ability of Foundation Models to Recognize Highlights in Videos.

arXiv:2601.16333v3 Announce Type: replace-cross Abstract: Foundation models are used for many real-world applications involving language understanding and generation from temporally-ordered multimodal events. In this work, we study the ability of models to ide