@april I argue that the latter weighs even more.

When my copyrighted works are published under BSD/MIT/… licences, they are made available to the average folk with attribution being basically the only “remuneration”. That’s not a lot and easily given by the recipients, so failing to do so weighs worse than failing a part of one of the more complex licences, both copyleft and EULA sides.

ML/LLM/so-called “AI” naturally drops that very attribution by compressing lossily, mixing, and decompressing the inputs. So therefore I’m very much not okay with them using my works under current circumstances.