I just thought about how you could probably "store" a pretty good approximation of a picture in a LLM and some input parameters that are orders of magnitude sma

I just thought about how you could probably "store" a pretty good approximation of a picture in a LLM and some input parameters that are orders of magnitude smaller in size than the original. I think that's what this paper is about right?