We should not be using GitHub, absolutely, but primarily for decentralisation.

In terms of training AI, they will find Open Source wherever it is and use it. They will also no doubt train on paid for / private repos too. I would guess the most useful source of data is people using copilot.

I am in the process of setting up pycharm with some locally hosted AI. It will not be as good but gives me peace of mind that sensitive data is not sucked into some AI cloud with unpredictable consequences.