AI lab building multimodal foundation models with native text, image, audio, and video understanding.