> We're still at 'AI generated works are public domain' aren't we?

As far as the US Copyright Office is concerned. I do not believe other countries have weighed in yet.

Also, when your training data is based on copyleft source code, and in some cases is a direct ripoff of it, will that license be applied to it?