@ednewtonrex.bsky.social
CEO of Fairly Trained / Composer. Working towards fairer training data practices in generative AI.
People had speculated on what went into ‘books2’ for years. Now we know; pirated books powered the hugely successful launch of ChatGPT.
IMO this should be mentioned every time ChatGPT is discussed.
storage.courtlistener.com/recap/gov.us...
/end
GPT-3.5 was the model used when ChatGPT was launched in November 2022. It was the default free-tier model for 1 1/2 years, until July 2024.
4/5
The judge said that a lawyer for OpenAI admitted that the pirate library LibGen was the source of a dataset they called ‘books2’, and that this dataset was used to train GPT-3 and GPT-3.5.
3/5
It came out in one sentence, buried halfway down an opinion from the judge in Authors Guild et al v. OpenAI.
2/5
When OpenAI released ChatGPT, it was trained on pirated books.
This finally came to light in a lawsuit three months ago, but it has gone essentially unreported.
🧵 1/5
MPT-7B-StoryWriter-65k+ (Databricks)
Cerebras-GPT (Cerebras)
CodeGen (Salesforce)
And this is just those where this info has come out in court / is traceable through academic papers. It’s safe to assume there are many more.
The AI industry was built on piracy.
2/2
Models we know were trained on pirated books (a non-exhaustive list):
GPT-3 (OpenAI)
GPT-3.5 (OpenAI)
Claude 1 (Anthropic)
Claude 2 (Anthropic)
Llama 1 (Meta)
Llama 2 (Meta)
Llama 3 (Meta)
OpenELM (Apple)
BloombergGPT (Bloomberg)
Megatron-Turing NLG 530B (Microsoft & Nvidia)
🧵 1/2
Yes! Thrilled to hear this.
13.02.2026 11:17 — 👍 1 🔁 0 💬 0 📌 0Agreed. Some authors are not able to sue again, as they have settled - but the settlement only covers ~500,000 books. There should still be lots of people who can sue.
13.02.2026 10:37 — 👍 2 🔁 0 💬 1 📌 0Whatever your view of its legality, it’s pretty clear that it sucks for authors, letting Anthropic make money at their expense. Authors should get paid when their books are used to train AI, and should have the chance to say no to that training. Anthropic’s used books strategy gives them neither.
13.02.2026 09:51 — 👍 22 🔁 4 💬 2 📌 0As Judge Chhabria said in Meta v Kadrey, LLMs will likely compete with the books they are trained on by flooding the market. And as Dario Amodei himself said in 2021, big AI companies centralizing profits by training on books without the authors getting paid is a real concern.
7/n
You can’t scan it and sell it as an ebook, for instance. There are limits on what you can do with books you’ve bought, where what you would be doing would compete with the book’s rights holders.
6/n
Anthropic uses their huge war chest to get all the books in the world (that’s their aim) - authors get nothing.
IMO there are serious questions over whether this should be legal. Yes, they are buying the books. But you can’t just do anything you like with a book once you’ve bought it.
5/n
Before alighting on this plan, they were discussing licensing from book publishers, which would have meant money going to authors. But then they came up with the used books plan, and stopped all licensing discussions.
4/n
They called this Project Panama, using a codename because they didn’t want people to know they were doing it. (It ultimately came out through court documents.)
3/n
They spent tens of millions of dollars buying used books from wholesalers, in batches of tens of thousands at a time. These were shipped to Illinois, scanned, and pulped.
2/n
When Anthropic stopped training on books that were literally pirated, they managed to hit on the one way of buying books that means no money goes to authors: buying used books.
🧵 1/n
He has never faced criminal charges, and, as a co-founder of Anthropic, must now be extremely wealthy (it is known that many of his co-founders are billionaires).
/end
In 2019 & 2021, Ben Mann downloaded at least 5 million books from pirate libraries. He did so while working at OpenAI and Anthropic; the books were downloaded for the purposes of training AI models at those companies, two of the most successful commercial companies in recent history.
3/4
He was indicted on 13 felony counts, facing a maximum of 95 years in prison and $3 million in fines. Two years after his arrest, awaiting trial, he hung himself.
2/4
In 2010, Aaron Swartz downloaded 4.8 million articles from JSTOR. We don't know why - it may have been to make them freely available online (if so, he never did, as he was caught). He made no money from this, and there is no suggestion he intended to.
🧵 1/4
Whatever today’s models were trained on, billions of dollars were raised off the back of models trained on pirated books, and those books were downloaded by some of the most important researchers at the companies in question.
The AI industry is built on piracy.
/end
And he went further - he said he understood that LibGen had also been downloaded by Alec Radford at OpenAI. Radford led the development of GPT-1 and GPT-2.
3/4
He also downloaded another pirate library, Books3, the very month Anthropic was founded. It was used to train Claude 1 & Claude 2.
He *also* said he had previously downloaded LibGen when he was at OpenAI. OpenAI used books from LibGen to train GPT-3 and GPT-3.5.
2/4
It is crazy that book piracy was so critical to the early days of both OpenAI and Anthropic.
In 2025, Anthropic co-founder Ben Mann testified that he himself had downloaded the massive pirate library LibGen at Anthropic. He did so only a few months after the company was founded in 2021.
🧵 1/4
Generative AI competes with the work it is trained on, and the people behind that work. Every single new piece of evidence backs this up.
The theft must be stopped - government must step in.
ism.org/brave-new-world
/end
This new report shows the havoc AI is already wreaking in the creative sector.
Evidence from 10,000 British creatives shows:
- 58% of photographers have lost work to AI
- 32% of illustrators have lost commissions to AI
- 86% of authors say AI has reduced their earnings
🧵 1/2
I am genuinely astonished by how terrible this 'AI Skills Hub' is.
For the record, this cost *£4.1 million*.
There are no words.
/end
... and the similarly huge number of courses from big tech companies that are only useful if *checks notes* you have paid for their products.
29.01.2026 14:57 — 👍 7 🔁 0 💬 1 📌 0
All of this is before you get to the absolutely huge number of courses that cost hundreds of pounds to enroll in...
29.01.2026 14:57 — 👍 5 🔁 0 💬 1 📌 0
