What happens if you feed a bunch of source code (like say, github's database) to a neural net and have it try to generate some novel programs?
>>53586342
It'll probably make better programs than the poo-in-loos
Neural net will also give you a lot of fail mutation.
someone did something kind of similar - a parser that identified common ways of doing various things in Ruby (based on a bunch of public repositories) that worked as an interactive IDE sort of thing
>>53586434
Ahh nice. Wonder if you could build a natural language->code translator with a library of exquisitely commented code
Read that, he fed the linux source to a recursive neural network: http://karpathy.github.io/2015/05/21/rnn-effectiveness/