@masek@infosec.exchange I see lots of copyright problems originating in the training data, which especially in the EU jurisdiction doesn't allow training of GenAI as fair use. There is no fair use doctrine, AND there is the rule that any new use needs a new license. Even if you restrict yourself to CC work during training, you can't know that this use is legal (maybe except CC0, which allows all possible uses). But those copyright problems appear in coding, too. The training data is most likely FOSS, but you still have this new use needs new license doctrine. A GPL'd program is not free for use as training data. The four freedoms are for humans.