Neural nets even have their own "universality theorem", meaning any function at all, any sort of process out there in the world, anything, could in theory be ap

Neural nets even have their own "universality theorem", meaning any function at all, any sort of process out there in the world, anything, could in theory be approximated by a neural net. Yet everyone just does MNIST :)