alt Hacker NewsAnthropic trained Claude on illegally obtained Stanford professors' books
Comments
> “To put me to work that way, without remittance, is a form of slave labor,” said Wolff, professor emeritus of English and winner of the National Medal of Arts. “They’re using my work without paying for it. As time goes on, they’re going to make the lives of writers very difficult if there’s no fruit to their labor.”
This is obviously hyperbolic, but it was a very poor choice of words.
It also relies on an assumption - that AI will become a substitute for buying books - that I think is likely untrue. You’ve been able to easily gain a surface-level understanding of anything for free online for decades, and for that matter, you’ve been able to easily pirate any book for decades! Hasn’t stopped people from buying books, or going to college, or doing any number of other expensive things that are technically unnecessary if the goal is just to gain information.
That professor saying that the use of work he already produced amounts to "slave labor"... copyright infringement is substantially different than slavery.
The title is (I believe intentionally) inaccurate. Better would be:
Anthropic trained on illegally acquired Stanford professors' books
The article does not say the training was illegal (although does discuss the professors push for a licensing model).
In a just world they would have to pull all models, for which this is the case, have auditors walking in left and right, supervising their training of the next version, to make sure they only train on material, that they are allowed to train on, and meticulously have them document every little corpus of data they use and its sources.
In reality we won't see this behavior stop, until Anthropic is hit with a huge penalty sledgehammer that threatens its existence. This is a gold rush and without real consequences, nothing will change.
> “The essential problem … is what economists call ‘transaction costs,’” Goldstein wrote in an email to The Daily. “Millions of individual copyrighted works may be copied in the course of any training incident, and it’s simply impracticable for the platform to obtain licenses from the copyright owner of every — or, indeed, any — single work.”
Seems strange. The question should be whether you legally have to do so or not. If you do whether it is impractical or not is not a reason to break the law. Perhaps it makes your business not feasible but so be it. If you ran a clothing store that required you steal from other brands and remix it for your customers, you couldn't say "it's impractical for me to get permission", no?
Training on the book being fair use for something whose whole deal is to take those ideas and exercise them is the steal of our time...
So you come with this amazing idea regarding anything... how to build a chip, how to write software... and you write a book on how to...
Then a model eat it... spit it for free (they could get an advantage from being able to exercise that logic) and it's fair use...
Then... why torrent it, can't they sent someone to buy the book and scan it or buy it in pdf?
Humanity is suffering from large scale distillation attack from Anthropic.
> “To put me to work that way, without remittance, is a form of slave labor,” said Wolff, professor emeritus of English and winner of the National Medal of Arts.
Are the late authors of Project Gutenberg also enslaved?
Title is misleading. The ruling said training on books is fair use. What got them in trouble was grabbing the books from LibGen instead of just buying them, which is a pretty dumb way to lose $1.5B. If I torrented half a million books I'd be getting a lot more than a $3k-per-book settlement.
I'm willing to bet the vast majority of students and professors at Stanford have consulted an "illegally obtained" textbook at some point. Either directly from Anna or second hand.
I hope the Stanford profs got paid... this alum is still owed $60k for ANT pirating my books.
Here's my understanding, someone correct me if I'm wrong:
- Anthropic et al basically grabbed whatever content they could to scan and train the model.
- The vast majority of what they scanned were books they bought at wholesale in big lots. Stuff being weeded out from Goodwills or libraries and (mostly) bound for destruction anyway.
- They didn't copy the texts per-se, they studied the relationships between words to build a statistical/predictive model of natural language.
- Whether this a fair or illegal use of copyrighted text already seems like an adorably obsolete question. Regardless, the damage is basically done - they can neither prove one way or another how much any single text influenced their model and they cannot unscramble the egg without starting over.
- Rather than get stuck in court for decades splitting the hairs about copyright law, they wrote massive checks to the publishing industry and agreed to a bunch of restrictions around copyrights.
- For the foreseeable future, every independent copyright holder now has grounds to take grievances to Anthropic and get some form of compensation.
But it's only the Chinese openweight model companies that are stealing anything, aren't they?
Of course they did, same with almost every other bit of training material that isn't synthetic (and still indirectly created with illegally obtained training data)
I’m fine with it as long as Anthropic is fine we can distill their models to create own without burning all that energy (vide DeepSeek case)
Of course they did. AI companies have pirated and stolen all sorts of content. What we know of is probably only the tip of the iceberg.
As for the training being fair use, this is still disputed:
https://www.courthousenews.com/ai-training-of-copyrighted-ma...
> the title
sounds bad...
> the article: "pirated the work of over 10 Stanford professors to train its Claude models"
that just... doesn't seem like much compared to so much other scraping though, right?
A single act of copyright infringement is a crime... a billion is a statistic!
The title change from "illegally trained" to "illegally obtained" implies that training on illegally obtained books is legal. I wouldn't be surprised if that's the case. Surely Anthropic has a legal team that signed off on it.
They did the same with literally everyone's data. Is it supposed to matter more because a "Stanford professor's" work was included?
If a permanent injunction is granted in the Wolff case, does that mean Anthropic would have to re-train all of their available models, but explicitly exclude those books from the dataset?
It's all built on stolen content.
There's a certain element of irony to all of this, what, with it being Stanford, the home of "forgiveness is easier to get than permission" and anti-regulation sentiment in the tech industry.
So as I’ve been told here numerous times, “AI companies do not steal”. Ah ok.
I think pushing ahead to a jury trial is a smart move. In the current climate you have a good chance of a very favorable jury pool.
They should get Opus or one of the latest models to be the morality judge of what goes into training.
I had an issue recently where I asked claude code to log into a developer user on Chrome on my local host for a system which is fully local and part of the code base the agent was developing and it refused to do it, like flat out refused, I did it on multiple sessions and I could not get Claude to budge on entering the credentials - it required me to manually sign in.
Would be amazing if they used that same level of training for their intake exercises.
They probably did it on purpose so that the professors would sue Anthropic later and make a lot of money.
How about all the copyright MIT licensed open source software? When does it become discovered that they were scraped for training without authorization and without giving attribution?
Did someone illegally press control-c and control-v again?
Well. At least I can ask an LLM about the contents of papers I can't access!
It would be funny to ask the profs if they are allowed, by law, to read 3 textbooks on a subject and then write a 4th themselves.
After a 1.5 billion settlement, the professor/authors are saying it's not enough.
The judge had an interesting summary of AI:
"He compared AI training to how humans can read, internalize and remember a book’s themes and style. “Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them — but to turn a hard corner and create something different,” he wrote."
Good!
Does copyright over knowledge-based works really have any legitimate position in today's society? We're entering a new age of how information is created, used, and distributed.
It’s so sad to watch humanity decide to go out with a whimper, yelling hopelessly at the machine.
Ability to pay might influence this outcome.
You don't say!
Very much worth noting here is the ruling due not say the training is the issue; it says the acquisition of the training material is the issue.
... Which, if enforced vigorously, means new raw AI can only be trained by those who can afford the raw material.
The Court just handed Anthropic and other VC-backed AI projects a significant moat.
Good! Information wants to be free, not locked up in some ivory tower. This is a great time upon us, when academics and ivy league elites can no longer gate keep knowledge and hoard it for their ego. Knowledge should benefit humanity as broadly as possible, not just the rich and their academic ilk.
“To put me to work that way, without remittance, is a form of slave labor,” said Wolff, professor emeritus of English and winner of the National Medal of Arts.
Yet again, Anthropic is the good guys. Though I think this is just the same story from a while ago but told through a Stanford lens, AFAICT?
This violates the RICO statute; the entire organization was aware of and relied on the widespread and organized violation of copyright laws as the basis of their entire business. It is a criminal organization, should be seized, and everyone in the chain of command that was aware of or executed on this plan should be tried criminally.
I do not think training an AI model is Fair Use, because LLM "training" is being used as a metaphor for building a matrix with a computer. Only people and animals train and learn; computers run statistical problems over input, which in this case was illegal input acquired by humans.
But it does not matter whether training LLMs is Fair Use, "Fair Use" is a strange and arbitrary legal term that can be expanded to include AI if we wish. This was organized crime whether or not LLM training on copyrighted materials is fair use. It was organized crime whether or not they planned to use them for AI at all.
AI is irrelevant. They could have trained it on GPL'd or MIT code (as long as they then distributed according to the terms of the licenses), and works in the public domain. They could have gotten what they were doing explicitly declared to be fair use by the Library of Congress, and then purchased the materials, like they are belatedly doing now. It wasn't necessary to violate copyright law to prove the technology. They could have proved the technology, then asked for copyright law to be clarified, and licensed huge volumes of works from the very few publishers we have left. Becoming a criminal organization was an explicit choice they made, and they discussed its illegality, and even the risk that it would be discussed on HN, which might inspire prosecution.
The reason they are not being prosecuted is because the rich are beyond the law; prosecutors will not prosecute them. The RICO statute is clear, and explicitly mentions copyright violation.
We can sentence the executives harshly, but go easy on the lower level people, because they are intelligent and valuable to advance the technology. The executives aren't necessary for anything. They're currently running finance scams that are extremely dangerous for the economy. They belong in prison to protect the public.
This is an option that people (real people, with names) have, and are not taking. When they say that they care about law and order, and are worried about the behavior of these companies, they are lying. Trump could literally nationalize them for what they've done. They're vulnerable to any state's attorney who has any author in their state whose book was downloaded. Libgen/Annas-archive would have been the most important honeypot in history.
[dead]
well duh
The primary value of a LLM is automated intellectual property laundering.
Didn't it train Claude on illegally obtained everything?
Why is this surprising? It's all built on theft.
And I presume the Anthropic CEO will be indicted on 13 felony counts under the Computer Fraud and Abuse Act and for wire fraud. Like Aaron Swartz, right?
Right?
N.B. JSTOR explicitly settled with Swartz, suffered no financial harm, and begged the federal government to drop the case. The government refused and prosecuted Swartz anyway. Anthropic settled for $1.5 billion primarily because of the theft and storage of the files, but this was somehow a Civil issue...