“Generative AI approaches that work well for language do not work for video or other sensory inputs because the real world is high-dimensional, continuous, and complex.”