A position paper arguing that consistency across views, modalities, and prompts should be the priority research target for unified multimodal models.
Feb 3, 2026