See2Think: Do Multimodal Models Really Use Intermediate Visual States? Paper • 2607.26769 • Published about 1 month ago • 25