I don't think it's a violation of copyright law because the GPL, as far as I understand it, gives the humans training the model the right to do so, and as long

I don't think it's a violation of copyright law because the GPL, as far as I understand it, gives the humans training the model the right to do so, and as long as they don't redistribute it, I don't have a problem with it. Once that code starts getting distributed, as happens with Copilot, then I think it's a problem. I'm just pointing out (repetitiously, because arguments of the type "but humans do X" are made all the time with LLMs) that the fact that a human brain doesn't violate copyright when a human reads and studies copyrighted code isn't relevant either way.