This exactly echoes our training methodology: language/image/video/audio are all deeply interconnected modalities of human experience. And multimodal intelligence should be grounded in language which is a generalizable and scalable computational system. Really, really appreciate you digging this up!🙌