Video most closely represents our multimodal reality which LLMs are unable to parse. Understanding across modalities can't just be a bolted on as a feature to existing LLMs. Multimodal foundation models actually have to be so from inception. We are changing the paradigm of how video is accessed and consumed.