News for hackers. Stuff that matters.
Re:Facebook is aggressively going after LLaMA repos with DMCA's by bottled_poe on Friday March 24, 2023 @08:42AM (#35288273)
Doesn’t matter. The weights wouldn’t exist without many tens of millions (at least) in investment. The outcome seems very clear.
Re:Facebook is aggressively going after LLaMA repos with DMCA's by papercrane on Friday March 24, 2023 @08:47AM (#35288328)
US copyright law cares about originality and expressiveness, not labour and cost. It can be very expensive to collect and print a list of every business and their phone number in a book, but the result is not copyrightable in the US.
Re:Facebook is aggressively going after LLaMA repos with DMCA's by evandale on Friday March 24, 2023 @12:05PM (#35291054)
This is a pretty compelling argument to me that the weights shouldn't be able to be copyrighted. In essence, aren't they a list of numbers you can search through to get a result?

Would it makes sense to say a "give me a rowboat on the water" prompt (search) is the same as a "give me the phone number of company XYZ" search in a phonebook?

What exactly is a prompt anyway? Can you copyright the assembly that a compiler spits out? Is an AI prompt the same as source code and its output is the assembly a compiler would generate? Does that mean the model is a compiler? I assume a compiler can be copyrighted, so then maybe the weights can be copyrighted? Or would it make more sense that the combination of (weights + prompt + seed + output) is copyrightable?

I don't have answers to any of these or know if they're reasonable questions but I'm starting to find this all very fascinating.

Re:Facebook is aggressively going after LLaMA repos with DMCA's by pygy_ on Friday March 24, 2023 @08:52AM (#35288373)
IANAL, but I don't think the DMCA applies. Copyright applies to creative work, and model weights aren't that.

It is however my understanding that downloading them can be considered a misappropriation of a trade secret.

Folks that would contest the bogus DMCA takedown requests would be liable to a trade secret suit.

Re:Facebook is aggressively going after LLaMA repos with DMCA's by rvnx on Friday March 24, 2023 @09:07AM (#35288524)
archive.org may be the legal solution ?
Re:Facebook is aggressively going after LLaMA repos with DMCA's by UncleEntity on Friday March 24, 2023 @10:01AM (#35289179)
You can’t just assert “this is a secret” and force people you have no contractual relationship with to comply with your assertion.

IMHO, not a lawyer.

The person who leaked the weights on BitTorrent is definitely liable for violating whatever restrictions they agreed to to get access to them though.

Re:Facebook is aggressively going after LLaMA repos with DMCA's by postalrat on Friday March 24, 2023 @08:50AM (#35288351)
If you are mining Bitcoin can you copyright a block you discover? You probably payed a lot of money for it.
Re:Facebook is aggressively going after LLaMA repos with DMCA's by blackbear_ on Friday March 24, 2023 @09:08AM (#35288533)
Don't forget that transformers, the basic architecture of these large language models, were introduced by Google, and many other basic building blocks were introduced by universities and research centers all over the world. So no, the outcome is not very clear, at all.
Re:Facebook is aggressively going after LLaMA repos with DMCA's by nl on Friday March 24, 2023 @09:40AM (#35288921)
This is irrelevant.