Hey all. I've been wondering this for a while but why is it infeasible to get a neural net to generate algorithms to solve hard math problems? Take large semi-prime factorization as an example...why couldn't we give a neural net a bunch of operations it can perform on any pass and just let it run until it finds a combination or pattern that works and then just continue training it up to make sure the pattern holds? I get that obviously it isn't that easy but if someone put some serious thought into its design why wouldn't it work?