[orthoganol]

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 :)