awesome-machine-learning-on-source-code

People have trained ML models on code thats on Github before co-pilot. (lots of examples here: https://github.com/src-d/awesome-machine-learning-on-source-...) There's nothing proprietary that other interested people or companies couldn't easily replicate here.

It seems tricky since ML models aren't really good at model building and code needs consistent variable names and structuring. The OpenGPT model is OK at doing sentences but can't really keep track of player state. The RNN in https://karpathy.github.io/2015/05/21/rnn-effectiveness/ can't use variable names consistently. The recent papers in https://github.com/src-d/awesome-machine-learning-on-source-... still can't use names consistently. Those kinds of issues don't prevent someone from whipping up something, but guarantee it'll have terrible accuracy. Anybody trying to use it would probably spend more time debugging the generated Prolog than they would just writing the Prolog themselves. PL research isn't really focused on translation anyway; the current trends I've seen are block-structured editors where you would snap things together t

Yes, this may be a good starting point — https://github.com/src-d/awesome-machine-learning-on-source-... Thanks for checking it out!