https://en.m.wikipedia.org/wiki/Manifold_hypothesis
i learned about this yesterday... last year, at one point, i had a bit of a mind blinder over how you can encode massive pieces of data with tiny pieces of data, i forget the scheme... completely forget it, but i'm sure it will come back to me...
it really flipped my wig at the time tho... i wish i could remember how i was envisioning utilising it
machine learning uses these things, apparently, and i can totally believe that you could devise a scheme to radically compress big long values like npubs and events, with zero chance of collision, would be an epic optimization for nostr event stores