Off topic, Slashdot has devolved further and added AI banner ads to the bottom of their site that persists through noscript/ublock origin and require manual and repeated intervention. I've reluctantly had to say that it was a good boy, and that it was time to go..

Slashdot is a news article aggregator. Can a better link can be found? ..like from the source?

Here you go: https://www.fsf.org/blogs/licensing/2026-anthropic-settlemen...

If the GPL actually holds, all LLMs need to be GPL'd. They all used GPL code in their training, they're all derivative works.

So far, all the relevant court cases that have actually been adjudicated have come down on the side of training LLMs to be a sufficiently transformative use of copyrighted material to fall under "fair use". If a case about the GPL were to break the same way, the GPL would still hold, but it would have no hold over this particular use as the GPL can't restrict "fair use" any more than any other set of copyright restrictions can.

Whether they are derivative works in the context of copyright law (which the GPL relies upon) has not yet been decided by the courts AFAICT. So your assertion may be your personal opinion but we don't know if the law agrees or not yet. From some quick searches it seems that the answer isn't a slam dunk one way or another and is still working its way through the courts.

Here's an article that seems relevant to the topic:

"District Court Finds That Using Copyrighted Works to Train Large Language Models Is Fair Use"

https://www.finnegan.com/en/insights/ip-updates/district-cou...