I've always wondered if we could leverage the vast amount of GitHub code - that assumably all compiles without error or undefined behaviour on their master branches - train some sort of neural net to better catch syntax errors.
Has anyone done something like this, or am I riding the 2016 neural net hype train still?