This raises an issue of trolling. What prevents developers to generate "inappropriate" code to feed it to this algorithm the same way they did with the Microsoft Chat bot for example? That will surely reflect on the quality of code generated by this AI system and therefore the stability and security of applications built.