In search of the next generation of multimodal datasets