Seriously. Can I copyright 34 * 712 * 9.2 * pi? I didn't think you could.

I'm in the "model weights aren't copyrightable" camp myself. I think the license exists largely to shield Meta from liability or the actions of third parties.

Humans make many choices that effect the trained weights: curation and choice of datasets, training schedules and hyperparameters. If these choices are made with an eye towards the generated results, rather than mechanically based upon test scores, why doesn't that rise up to the minimal level required to get a copyright on the weights?

It might. As I say, it’s up for debate. A judge might look at the 1kB of hyperparameters versus the 1TB of training data, and the 10 person-years of human effort versus 100,000 GPU-years of computer effort, and conclude differently.

Does Google have copyright of their search index? Never tested, as far as I know.

"sweat of the brow" was rejected in Feist, the effort involved makes no difference. The minimum creativity bar for obtaining copyright is very low but above an alphabetical ordered listing of all of your customers.

This seems like a reasonable position to take. Can you copyright the contents of a vacuum bag after pouring it down a gallon board as "art"?

Did you have any meaningful hand in constructing the contents?

> AFAIK it’s still an open question whether there is any copyright in model weights

There's definitely copyright when you ask the model to spit out Chapter 3 of a Harry Potter book and it literally gives it to you verbatim (Which I've gotten it to do with the right prompts). There's no world where the legal system gives Meta a right to license out content that never belonged to them in the first place.

In the US, it's not an open question. Feist v Rural holds that any work needs to possess a minimum spark of (human) creativity to be able to be copyrighted; data collected by the "sweat of the brow" is explicitly not allowed to be copyrighted. Thus things like maps and telephone books aren't really copyrightable (they do retain a "thin copyright" protection, but in the present context, you're going to say that the code has copyright but the model weights do not). Most European jurisdictions do recognize a "sweat of the brow" doctrine, and they could be copyrightable there.

What's not clear is whether or not the model weights infringe the copyright of the works they were trained on.

I wonder what Meta’s attitude towards copyright in general is. Last I heard their lawyers were trying to say that it’s all good to pirate copyrighted works to make machine learning models.