As for case #1 I personally believe this to be the same as learning from preexisting code as individuals do. But it is of course still an open questions and as I said I accept the FSF's arguments on not accepting LLM contributions to GNU Emacs. If the scenario is #2, then all LLM outputs are by definition free software; software in the public domain is fully free. The difference is that it is not /copyleft/, and so derivative works do not also have to be free software. But that is still the case for many FLOSS licenses.