This could possibly be an outgrowth of LLMs and could use drawings and images instead of numbers and words. It would have the images be really really small so it could store loads of images. How would this work?