So when’s the ruling against OpenAI and the like using the same copyrighted material to train their models
this post was submitted on 04 Sep 2024
950 points (98.6% liked)
Technology
60052 readers
2928 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 2 years ago
MODERATORS
But OpenAI not being allowed to use the content for free means they are being prevented from making a profit, whereas the Internet Archive is giving away the stuff for free and taking away the right of the authors to profit. /s
Disclaimer: this is the argument that OpenAI is using currently, not my opinion.
Ah, I see you got that all wrong.
Open ~~IA~~ AI uses that content to generate billions in profit on the backs of The People. The Internet Archive just does it for the good of The People.
We can't have that. "Good for The People" is not how the economy works, pal. We need profit and exploitation for the world to work...
load more comments
(3 replies)
Hot on the heels of this one, I'd imagine.
Fat chance. Line must go up.
So, let's say we create an llm that will be fed will all the copyrighted data and we design it, so that it recalls the originals when asked?! Does that count as piracy or as the kind of legal shananigans openai is doing?
load more comments
(1 replies)
load more comments
(18 replies)
If OpenAI can get away with going through copy-righted material, then the answer to piracy is simple: round up a bunch of talented Devs from the internet who are writing and training AI models, and let's make a fantastic model trained on what the internet archive has. Tell you what, let Mistral's engineers lead that charge, and put an AGPL license on the project so that companies can't fuck us over.
I refuse to believe that nobody has thought of this yet
An AI trained on old Internet material would be like a synthetic Grandpa Simpson:
"In my day we said 'all your base' and laughed all day long, because it took all day to download the video."
This stupid thing just keeps saying “I can Haz Cheeseburger”. What the hell does that even mean?
What do you think Mistral trains its models on? Public domain stuff?
load more comments
(7 replies)
Fuck Copyright.
A system for distributing information and rewarding it's creators should not be one based on scarcity, given that it costs nothing to copy and distribute information.
It was fine when the limited duration was a reasonable number of years. Anything over 30 years max before being in the public domain is too long.
Yeah. In a better world where the US court system doesn't get weaponized and rulings aren't delayed for years or decades, I would argue 8 to 15 years is the reasonable number, depending on the type of information being copyrighted.
That was fine then, but it makes zero sense today.
If a book is on sale widely to the public, and it costs nothing to copy and distribute that book to everyone, why shouldn't we?
The fundamental problem with copyright is it is a system that rewards creators by imposing artificial scarcity where there is no need for one. Capitalism is a system designed around things having value when they're scarce, but information in a world of computers and the internet is inherently unscarce the instant it's digitized. Copyright just means that we build all these giant DRM systems to impose scarcity on something that doesn't need it so that we can still get creators paid a living.
But a better system would for paying creators would be one of attribution and reward, where everyone can read whatever they want or stream whatever they want, and artists would be paid based on their number of views.
But a better system would for paying creators would be one of attribution and reward, where everyone can read whatever they want or stream whatever they want, and artists would be paid based on their number of views.
Which would be enforced through copyright...
load more comments
(19 replies)
I personally like the idea that Copyright should be on par with design patent law. An initial filing 10-15yrs plus two additional opportunities to renew and extend it for 10 years if the creator can make supplementary creations that were dependent on and based off of the original works. -In the case of novels, that would equate to new sequels or prequels.
load more comments
(1 replies)
Artificial scarcity at its finest. Imagine recording a song digitally, then pretending there are a limited amount of copies of that song in existence. Then you sell an agreement to another person that says they have to pretend there is only a certain made up number of copies that they bought, and if they allow more than that number of people to listen to those copies at rhe same time, they will get sued for "stealing" additional pretend copies?
I hope everybody can see how this is the insane and pathetic result of Capitalism's unrelenting drive to commodify everything it possibly can in the pursuit of profit.
As always, the solution is sailing the high seas. Throughout history, those who created or saved illegal copies/translations of literature and art were important to preserving and furthering human knowledge.
Many incredibly powerful people, empires, and countries have tried very hard to suppress that, but they keep failing. You cannot suppress the human drive for curiosity and knowledge.
True, and the fleet is big and strong. There are many people seeding hundreds of terabytes of books/research papers/etc. The knowledge will not be lost. Yarr, can't catch me in the high seas...
Not a surprise, but still somehow crushing. It's a loss for us all.
Ah, I see we're burning the Library of Alexandria again... Just as with last time, the survival of texts will rely upon copies.
Oh sure I want to read copyright books it's an issue, but OpenAI does it and it's vital to their business so they can keep going.
load more comments
(2 replies)
My understanding is that the IA had implemented a digital library, where they had (whether paid or not) some number of licenses for a selection of books. This implementation had DRM of some variety that meant you could only read the book while it was checked out. In theory, this means if the IA has 10 licenses of a book, only 10 people have a usable copy they borrowed from the IA at a time.
And then the IA disabled the DRM system, somehow, and started limitlessly lending the books they had copies of to anyone that asked.
I definitely don't like the obnoxious copyright system in the USA, but what the IA did seems obviously ~~wrong~~ against the agreement they entered into. Like if your local library got a copy of Book X and then when someone wanted to borrow it they just copied it right there and let you keep the copy.
ETA: updated my wording. I don't believe what the IA did was morally wrong, per se, but rather against the agreement I presume they entered into with the owners of the books they lent.
They disabled drm during lockdown so people had something to do
Which was nice of them, but that doesn’t mean they should’ve done that, especially in the eyes of the law. (Also, if you’re after free ebooks, why are you pirating them on archive.org instead of libgen?)
load more comments
(4 replies)
I definitely don’t like the obnoxious copyright system in the USA, but what the IA did seems obviously wrong.
The publisher-plaintiffs did not prove the "obvious wrong" in this case, however US-based courts have a curious standard when it comes to the application of Fair Use doctrine. This case ultimately rested on the fourth, most significantly-weighted Fair Use standard in US-based courts: whether IA's digital lending harmed publisher sales during the 3-month period of unlimited digital lending.
Unfortunately, when it comes to this standard, the publisher-plaintiffs are not required to prove harm, rather only assert that harm has occurred. If they were required to prove harm they'd have to reveal sales figures for the 27 works under consideration--publishers will do anything to conceal this information and US-based courts defer to them. Therefore, IA was required to prove a negative claim--that digital lending did not hurt sales--without access to the empirical data (which in other legal contexts is shared during the discovery phase) required to prove this claim. IA offered the next best argument (see pp. 44-62 of the case document to check for yourself), but the data was deemed insufficient by the court.
In other words, on the most important test of Fair Use doctrine, which this entire case ultimately pivoted upon, IA was expected to defend itself with one arm tied behind its back. That's not 'fair' and the publishers did not prove 'obvious' harm, but the US-based courts are increasingly uninterested in these things.
edited: page numbers on linked court document.
load more comments
(9 replies)
Easy solution. Update the web-scraper they use to include an LLM. Then its for "training"
As long as they have a tech billionaire in charge they should be fine.
They could also rename the project to: "The AI Archive" and add lots of buttons with multicolor gradients.
load more comments
(1 replies)
Libgen.rs
Direct link to the court document: https://storage.courtlistener.com/recap/gov.uscourts.ca2.60988/gov.uscourts.ca2.60988.306.1.pdf
Side note: court listener's RECAP is often quite disliked by the legal system. They do not like it when people put stuff from PACER fee waved sources on there like Aaron Schwartz did. https://en.m.wikipedia.org/wiki/Free_Law_Project
load more comments
(1 replies)
They need to rename themselves "Intelligent Archive" then claim they're an AI service that can just happen to regenerate whole books.
Really unfortunate. I wonder why nobody foresaw this when they started the stupid NEL thing.
Edit: NEL is the thing where the Archive removed all borrowing restrictions except 10 books per account and some sort of basic verification that you were in the US
load more comments
(1 replies)
But I'm training my organic LLM, can't I?
what does warrior do? The git readme seems to just be setup instructitons
I had the same question. Here's the answer:
The Archive Team Warrior is a virtual archiving appliance. You can run it to help with the Archive Team archiving efforts. It will download sites and upload them to our archive—and it’s really easy to do!
The warrior is a container running inside a virtual machine, so there is almost no security risk to your computer. ("Almost", because in practice nothing is 100% secure.) The warrior will only use your bandwidth and some of your disk space, as well as some of your CPU and memory. It will get tasks from and report progress to the Tracker.
load more comments
(5 replies)
view more: next ›