Bayes-optimal learning of an extensive-width neural network from quadratically many samples | Read Paper on Bytez