Comment by jakey_bakey

At the risk of sounding relentlessly skeptical - surely by training the code on GitHub data you're not actually creating an AI to solve problems, but creating an extremely obfuscated database of coding puzzle solutions?