someone should plot the entropy of a large language model asked to give a random number and see if it's even close to pseudorandom