AI training sets keep expanding, and I was thinking about the rate of progress and what natural limits would there be. I took a low-effort stab at estimating the size of a Panopticon text corpus, and was wondering are my assumptions/calculations reasonable?