✨ Nora Tindall  🪐

Maybe I'm not understanding the situation properly but isn't using the code as training set read only? Is there a difference between reading the public code and using it as training data? As far as I know all licenses allow me to read the code without doing anything else with it.

Sure - reading is fine. If you download the repo and change even one letter, then send it to your friend, you now are covered by the GPL. They downloaded the repos, and they changed more than one letter.

But if I understand this correctly, they didn't change anything. The code copilot outputs should be unique.

But if you removed all the GPL code from the training set, surely CoPilot would return different results; therefore is CoPilot not depending upon the GPL code and as a derivative work would have to be releases as GPL itself?

If you ever read GPL code, learned from it, you're still able to write code with another license, right?

correct. this is because you are a human person and not a machine.

And whats the difference? Their software isn't going to reproduce any of the code its read unless it's trivial code, which can't be copyrighted anyway