You forgot to mention multimodality. GPT-4 will most likely be able to handle both text and images. Who knows, maybe also sequences of images.