I'm not pirating. I'm building my model.
this post was submitted on 25 Jun 2025
109 points (97.4% liked)
Technology
71866 readers
4354 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- 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, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
To anyone who is reading this comment without reading through the article. This ruling doesn't mean that it's okay to pirate for building a model. Anthropic will still need to go through trial for that:
But he rejected Anthropic's request to dismiss the case, ruling the firm would have to stand trial over its use of pirated copies to build its library of material.
I also read through the judgement, and I think it's better for anthropic than you describe. He distinguishes three issues:
A) Use any written material they get their hands on to train the model (and the resulting model doesn't just reproduce the works).
B) Buy a single copy of a print book, scan it, and retain the digital copy for a company library (for all sorts of future purposes).
C) Pirate a book and retain that copy for a company library (for all sorts of future purposes).
A and B were fair use by summary judgement. Meaning this judge thinks it's clear cut in anthropics favor. C will go to trial.
An 80 year old judge on their best day couldn't be trusted to make an informed decision. This guy was either bought or confused into his decision. Old people gotta go.
Did you read the actual order? The detailed conclusions begin on page 9. What specific bits did he get wrong?
I'm on page 12 and I already saw a false equivalence between human learning and AI training.
Is it this?
First, Authors argue that using works to train Claude’s underlying LLMs was like using works to train any person to read and write, so Authors should be able to exclude Anthropic from this use (Opp. 16).
That's the judge addressing an argument that the Authors made. If anyone made a "false equivalence" here it's the plaintiffs, the judge is simply saying "okay, let's assume their claim is true." As is the usual case for a preliminary judgment like this.
Page 6 the judge writes the LLM “memorized” the content and could “recite” it.
Neither is true in training or use of LLMs
The judge writes that the Authors told him that LLMs memorized the content and could recite it. He then said "for purposes of argument I'll assume that's true," and even despite that he went ahead and ruled that LLM training does not violate copyright.
It was perhaps a bit daring of Anthropic not to contest what the Authors claimed in that case, but as it turns out the result is an even stronger ruling. The judge gave the Authors every benefit of the doubt and still found that they had no case when it came to training.
Depends on the content and the method. There are tons of ways to encrypt data, and under relevant law they may still count as copies. There are certainly weaker NN models where we can extract a lot of the training data, even if it's not easy, from the model parameters (even if we can't find a prompt that gets the model to regurgitate).
IMO the focus should have always been on the potential for AI to produce copyright-violating output, not on the method of training.
If you try to sell "the new adventures of Doctor Strange, Steven Strange and Magic Man." existing copyright laws are sufficient and will stop it. Really, training should be regulated by the same laws as reading. If they can get the material through legitimate means it should be fine, but pulling data that is not freely accessible should be theft, as it is already.
That "freely" there really does a lot of hard work.
It means what it means, "freely" pulls its own weight. I didn't say "readily" accessible. Torrents could be viewed as "readily" accessible but it couldn't be viewed as "freely" accessible because at the very least you bear the guilt of theft. Library books are "freely" accessible, and if somehow the training involved checking out books and returning them digitally, it should be fine. If it is free to read into neurons it is free to read into neural systems. If payment for reading is expected then it isn't free.
Plantifs made that argument and the judge shoots it down pretty hard. That competition isn't what copyright protects from. He makes an analogy with teachers teaching children to write fiction: they are using existing fantasy to create MANY more competitors on the fiction market. Could an author use copyright to challenge that use?
Would love to hear your thoughts on the ruling itself (it's linked by reuters).
I hate AI with a fire that keeps we warm at night. That is all.
80% of the book market is owned by 5 publishing houses.
They want to create a monopoly around AI and kill open source. The copyright industry is not our friend. This is a win, not a loss.
What, how is this a win? Three authors lost a lawsuit to an AI firm using their works.
The lawsuit would not have benefitted their fellow authors but their publishing houses and the big ai companies.
How exactly does this benefit "us" ?
Because books are used to train both commercial and open source language models?
Keep in mind this isn't about open-weight vs other AI models at all. This is about how training data can be collected and used.
If you aren't allowed to freely use data for training without a license, then the fear is that only large companies will own enough works or be able to afford licenses to train models.
If they can just steal a creator's work, how do they suppose creators will be able to afford continuing to be creators?
Right. They think we have enough original works that the machines can just make any new creations.
😠
It is entirely possible that the entire construct of copyright just isn't fit to regulate this and the "right to train" or to avoid training needs to be formulated separately.
The maximalist, knee-jerk assumption that all AI training is copying is feeding into the interests of, ironically, a bunch of AI companies. That doesn't mean that actual authors and artists don't have an interest in regulating this space.
The big takeaway, in my book, is copyright is finally broken beyond all usability. Let's scrap it and start over with the media landscape we actually have, not the eighteenth century version of it.
I'm fairly certain this is the correct answer here. Also there is a seperation between judicative and legislative. It's the former which is involved, but we really need to bother the latter. It's the only way, unless we want to use 18th century tools on the current situation.
Yeah, I guess the debate is which is the lesser evil. I didn't make the original comment but I think this is what they were getting at.
Absolutely. The current copyright system is terrible but an AI replacement of creators is worse.
Yes precisely.
I don't see a situation where the actual content creators get paid.
We either get open source ai, or we get closed ai where the big ai companies and copyright companies make bank.
I think people are having huge knee jerk reactions and end up supporting companies like Disney, Universal Music and Google.
The companies like record studio who already own all the copyrights aren't going to pay creators for something they already owned.
All the data has already been signed away. People are really optimistic about an industry that has consistently fucked everyone they interact with for money.
Yes. But then do something about it. Regulate the market. Or pass laws which address this. I don't really see why we should do something like this then, it still kind of contributes to the problem as free reign still advantages big companies.
(And we can write in law whatever we like. It doesn't need to be a stupid and simplistic solution. If you're concerned with big companies, just write they have to pay a lot and small companies don't. Or force everyone to open their models. That's all options which can be formulated as a new rule. And those would address the issue at hand.)
Because of the vast amount of data needed, there will be no competitive viable open source solution if half the data is kept in a walled garden.
This is about open weights vs closed weights.
Cool than, try to do some torrenting out there and don't hide that. Tell us how it goes.
The rules don't change. This just means AI overlords can do it, not that you can do it too
I've been pirating since Napster, never have hidden shit. It's usually not a crime, except in America it seems, to download content, or even share it freely. What is a crime is to make a business distributing pirated content.
I know but you see what they're doing with ai, a small server used for piracy and sharing is punished, in some cases, worse than a theft. AI business are making bank (or are they? There is still no clear path to profitability) on troves pirated content. This (for small guys like us) is not going to change the situation. For instance, if we used the same dataset to train some AI in a garage and with no business or investor behind things would be different. We're at a stage where AI is quite literally to important to fail for somebody out there. I'd argue that AI is, in fact going to be shielded for this reason regardless of previous legal outcomes.
Agreed. And even if it were, it's always like this. Anthropic is a big company. They likely have millions available for good lawyers. While the small guy hasn't. So they're more able to just do stuff and do away with some legal restrictions. Or just pay a fine and that's pocket change for them. So big companies always have more options than the small guy.