> with a focus only on books with an ISBN number, the identification system introduced in 1966
So it's not likely to be rare and precious books. It's things which are already in the Library of Congress or its many equivalents.
If they're forced by laws to destroy the results of the scans after training on them, as some have implied, that's bad, on the chance there is some actual lost media in there. But if they keep the scans, or even the transcripts, that's probably an improvement on the status quo, to be honest.
I've had family members in the used books business, and trust me the fate of the vast majority of these books was always to be pulped.
Yeah, we’re in a funny position. By all accounts it is fair use (at least in the US) to train models (and build search indexes, e.g. Google Books), but sharing the books dataset itself is absolutely forbidden (clear non-transformative copying).
Anyone that wants to train a model needs to procure and destroy their own physical copy of each book!
How is this in any way, shape, or form a promotion of the arts and sciences anymore?
This could very easily be turned into a preservation and archiving operation with just a tweak of the laws, or a carve-out.
And make the bank once, and make it legal to train on? How people in the bank get compensated is a different question, but -while almost impossible to settle on an individual basis- could be settled reasonably in bulk by some form of mandatory implied statutory contract?
How is this in any way, shape, or form a promotion of the arts and sciences anymore?
Well, if the book is still in production the author will get the royalties. Though if it were still an active release they should just sell digital copies.
letting companies train on books for only the price of a paperback and then forcing them to destroy their scans would truly be the worst of all worlds. (ps: the scanning process typically destroys the book too; they chop off the spines.)
Rare, certainly: there are plenty of vanity publications that were chucked into the bin by almost everyone who was unlucky enough to be given a copy. Precious? Well, with the help of an electronic friend I found ISBN 978-3-8365-7349-8. Apparently a copy of that book is worth about £40k. Can anyone beat that?
(EDIT: I'm assuming "precious" means the same as valuable here to make the question easier to answer. In fact, of course, people usually say "precious" when they mean a personal or emotional attachment or cultural importance rather than economic value.)
> Anthropic said: "None of our data acquisition programmes buy and destroy rare or antiquarian books."
This statement was clearly carefully worded, using the words "rare or antiquarian" to mean "pre-1966, pre-ISBN". Their statement is answering a question that wasn't asked, so that (if they need to) they can claim in the future that they never said they weren't destroying books with ISBNs.
> on the chance there is some actual lost media in there. But if they keep the scans, or even the transcripts, that's probably an improvement on the status quo
Where can I see the scans and the transcripts in this "better status quo"?
This is a good thing to a point, right? They sell heaps of stock, some old / outdated stock (like "Pass Your Driving Test, 2018 Edition" as per the article), in bulk, at full price, no questions asked. Retailer's dreams come true.
(I don't believe them being digitized for consumption by AI will impact book sales much, as ebooks, google books, project gutenberg, etc didn't either)
As long as the AI devouring a book isn't making it harder for a person to get access to a copy of the book. There might not be many copies of some out-of-print books.
Perhaps Anna's Archive will be the last bastion of some books, especially if they manage to acquire the internal collections of some of the AI companies, à la Spotify. (ie. the parts of the collections that didn't come from Anna in the first place.)
It's great for the retailer, although I think that example was picked out because it's an extreme example of a book that'll never sell. But the volume of these orders shows another way the AI companies are pissing away their investors' money.
German copyright law has no Fair Use. There are copyright exemptions for private or academic use, but commercial use has no such exemption. Books have special protection even according to § 53 (4) b).
And yes, it's the duplication which is illegal, not the publication of them.
"Individual reproductions of a work by a natural person for private use on any medium are permitted, provided they are not used directly or indirectly for commercial purposes …"
Two American self published nonfiction authors I know reported large purchases of new books earlier this year, more than 50 units at a time, via Amazon. This is unusual because the books are not well known. Because the books do not sell well otherwise, they were very few used copies available for sale on Amazon.
Here's my AI theory: someone is purchasing lots of books en masse to be packaged and sold for LLM training to multiple clients (not just a single company like Anthropic), but only scanning once and then reselling the scans while preserving a "chain of custody" proof that individual copies were purchased. Claude shot that idea down for reasons related to the intricacies of US first sale doctrine and copyright law, but had an ambiguous response when I proposed that it might be Chinese companies doing something similar for training AI models that are intended for possible resale to overseas markets.
Publishing any out-of-copyright books would be awesome thing to do. But it seems none of the books in this order would fall under that, and it's unclear how many old books they are actually ingesting. Current reports are mostly talking about relatively recent books
Preservation rather than ingesting them into the gaping maw of an Intelligence Engine that preserves the soul of the book in crystalized neural network form? That would be a copyright violation.
Remember kids, if you say "I was doing it for AI", it's legal. If you say "I was doing it for the good of humanity," well, we still remember Aaron Swartz.
But of course, unlike humanity, AI is really worth it!
I wish we had competent regulators. Force the scans to be purchased through an official repository. Save the scans and sell them to other companies who also want to train models. Force the companies to make all book requests through the official repository. Give some portion of money paid back to the publishers. This is a solvable problem.
I didn't think about this happening. It reminds me of the cross-border challenges with crypto. Each country has laws to control its author rights or money supply, but those are hard to enforce in the international setup.
no, because that's cost and everything needs to be the cheapest possible. only valid spend is money going to the shareholders.
and chopping off the spine and putting the stack of paper into an automated scanner saves sooo much money over a $5/hr employee flipping pages all day
i think that they might actually want to do the ram gambit all over again - buy up these rare books so that competitors can't get them. destroying them after scanning makes doubly sure that the competition won't get their hands on it
When I hear these stories of AI companies buying all the books, I think back to Kevin Kelly in 2008 or 2009. He talked about how books are less expensive than at any other time in history and easier to buy and that could change so it makes sense to buy a lot of books.
> [...] I was near to the point of actually digitizing and getting rid of all my paper books.
> I was that close about five years ago, but then I had an
epiphany. I went to private library, and I realized that books
were never as cheap as they are today. They never will be as
cheap, and that there's some power about having these things in
paper always available, no batteries, never obsolete, and that if
you made a library now, you would never be able to make some
of these libraries in 50 years, so I decided to keep and to
cultivate this paper library as something that was going to be
powerful in the future.
I remember seeing photos of his library but can't find them anymore. This is the only thing I could find:
Is this a second form of AI psychosis, as in AI training corpus psychosis?
The need for all the content, Moar!! Feed me. Is this the Paperclip Maximiser in the form of the Training Material Maximiser?
Will it be that, in the end, the lack of "Pass Your Driving Test, 2018 Edition" was the cause of driving rule hallucinations in all previous models? Will this finally get OpenAI back in front of Anthropic? Hurrah! We found it!
I was thinking the inverse the psychosis is that someone who sells books is alerted that they are selling more books, as if someone buying books needs to give an explanation. If I want to seed software with books why is that newsworthy? Print more books, maintain profitable margins on books, enjoy success as a book seller?
If a book is rare and you wanted to preserve it don't sell it so easily, increase the price, or put it into a contract system like Ferrari does with cars they sell.
Till it hits books that are out of print, we already had news about AI scourge companies destroying rare books coz scanning them in destructive ways was slightly cheaper.
And as is happens in history, any time some entity mass destroys the books, they turn out to be bad guys
There's a very simple solution to this, but the people who rabidly argue for a copyright length of 70 years + the lifetime of the author won't want to hear it.
> with a focus only on books with an ISBN number, the identification system introduced in 1966
So it's not likely to be rare and precious books. It's things which are already in the Library of Congress or its many equivalents.
If they're forced by laws to destroy the results of the scans after training on them, as some have implied, that's bad, on the chance there is some actual lost media in there. But if they keep the scans, or even the transcripts, that's probably an improvement on the status quo, to be honest.
I've had family members in the used books business, and trust me the fate of the vast majority of these books was always to be pulped.
> But if they keep the scans, or even the transcripts, that's probably an improvement on the status quo, to be honest.
only if they share the scans/transcripts and don’t hide it away.
Yeah, we’re in a funny position. By all accounts it is fair use (at least in the US) to train models (and build search indexes, e.g. Google Books), but sharing the books dataset itself is absolutely forbidden (clear non-transformative copying).
Anyone that wants to train a model needs to procure and destroy their own physical copy of each book!
How is this in any way, shape, or form a promotion of the arts and sciences anymore?
This could very easily be turned into a preservation and archiving operation with just a tweak of the laws, or a carve-out.
And make the bank once, and make it legal to train on? How people in the bank get compensated is a different question, but -while almost impossible to settle on an individual basis- could be settled reasonably in bulk by some form of mandatory implied statutory contract?
How is this in any way, shape, or form a promotion of the arts and sciences anymore?
Well, if the book is still in production the author will get the royalties. Though if it were still an active release they should just sell digital copies.
> Anyone that wants to train a model needs to procure and destroy their own physical copy of each book!
Why? Couldn't they resell or give away the books after scanning them?
Per the courts, it actually helps their "fair use" case if they destroy the book instead of re-selling it.
It's less work if you use a destructive method. They rip the spine off to get individual flatter sheets instead of scanning a bound book.
chop off the spine and you can use an automated scanner. to keep the book intact you'll need a flatbed scanner and an operator.
letting companies train on books for only the price of a paperback and then forcing them to destroy their scans would truly be the worst of all worlds. (ps: the scanning process typically destroys the book too; they chop off the spines.)
SBN started in 1967 in UK (W.H.Smith), official international book numbering (ISBN) is after 1970.
Source: https://www.isbn.org/ISBN_history
Strange to think of WH Smith as being at the forefront of technology.
Also weird that in the space of 5 years, they'd gone from an idea, got national buy in, and then got an international standard sorted.
There are plenty of rare and precious books with ISBNs.
Rare, certainly: there are plenty of vanity publications that were chucked into the bin by almost everyone who was unlucky enough to be given a copy. Precious? Well, with the help of an electronic friend I found ISBN 978-3-8365-7349-8. Apparently a copy of that book is worth about £40k. Can anyone beat that?
(EDIT: I'm assuming "precious" means the same as valuable here to make the question easier to answer. In fact, of course, people usually say "precious" when they mean a personal or emotional attachment or cultural importance rather than economic value.)
There are, but, quoting from the article:
> Anthropic said: "None of our data acquisition programmes buy and destroy rare or antiquarian books."
This statement was clearly carefully worded, using the words "rare or antiquarian" to mean "pre-1966, pre-ISBN". Their statement is answering a question that wasn't asked, so that (if they need to) they can claim in the future that they never said they weren't destroying books with ISBNs.
In the lost media sense? I'm not sure.
Of course there are always post 1960 books which are rare and valuable for being very limited run first editions, signed copies etc.
> on the chance there is some actual lost media in there. But if they keep the scans, or even the transcripts, that's probably an improvement on the status quo
Where can I see the scans and the transcripts in this "better status quo"?
It's not enough of an improvement that you might see them. It's enough of an improvement that they will likely still exist somewhere.
This is a good thing to a point, right? They sell heaps of stock, some old / outdated stock (like "Pass Your Driving Test, 2018 Edition" as per the article), in bulk, at full price, no questions asked. Retailer's dreams come true.
(I don't believe them being digitized for consumption by AI will impact book sales much, as ebooks, google books, project gutenberg, etc didn't either)
As long as the AI devouring a book isn't making it harder for a person to get access to a copy of the book. There might not be many copies of some out-of-print books.
Would be nice to at least force them to open their scanned books data, after all they stole everything else, they can contribute some back
"Open" it how, exactly? Make it available for free? As in violate copyright?
Perhaps Anna's Archive will be the last bastion of some books, especially if they manage to acquire the internal collections of some of the AI companies, à la Spotify. (ie. the parts of the collections that didn't come from Anna in the first place.)
And what, potentially make things easier for a competitor? Not in capitalism sir
It's great for the retailer, although I think that example was picked out because it's an extreme example of a book that'll never sell. But the volume of these orders shows another way the AI companies are pissing away their investors' money.
Download content to train? Immoral and illegal.
Buy it? You’re wasting investor money.
> “Under German law scanning books – regardless of the purpose – would not be permissible and constitute a clear violation of copyright law,”
Excuse me, what?
German copyright law has no Fair Use. There are copyright exemptions for private or academic use, but commercial use has no such exemption. Books have special protection even according to § 53 (4) b).
And yes, it's the duplication which is illegal, not the publication of them.
[1]: https://www.gesetze-im-internet.de/urhg/__53.html
https://law.stackexchange.com/a/93882
"Individual reproductions of a work by a natural person for private use on any medium are permitted, provided they are not used directly or indirectly for commercial purposes …"
Two American self published nonfiction authors I know reported large purchases of new books earlier this year, more than 50 units at a time, via Amazon. This is unusual because the books are not well known. Because the books do not sell well otherwise, they were very few used copies available for sale on Amazon.
Here's my AI theory: someone is purchasing lots of books en masse to be packaged and sold for LLM training to multiple clients (not just a single company like Anthropic), but only scanning once and then reselling the scans while preserving a "chain of custody" proof that individual copies were purchased. Claude shot that idea down for reasons related to the intricacies of US first sale doctrine and copyright law, but had an ambiguous response when I proposed that it might be Chinese companies doing something similar for training AI models that are intended for possible resale to overseas markets.
In the case of English-language books, they’re just working to legitimize the stuff they already obtained via, um, dubious methods.
Public domain books - no need to bother, they’ve scraped PG and any other source and can do with it what they like.
Used booksellers should raise the price of used books that are of no interest to any real reader.
An instant and obvious good these companies could do is to digitize these books ala google books.
And get sued even harder by publishers?
Publishing any out-of-copyright books would be awesome thing to do. But it seems none of the books in this order would fall under that, and it's unclear how many old books they are actually ingesting. Current reports are mostly talking about relatively recent books
Preservation rather than ingesting them into the gaping maw of an Intelligence Engine that preserves the soul of the book in crystalized neural network form? That would be a copyright violation.
Then they would have to deal with copyright. AI is the washing machine that leaves everything spotless.
Remember kids, if you say "I was doing it for AI", it's legal. If you say "I was doing it for the good of humanity," well, we still remember Aaron Swartz.
But of course, unlike humanity, AI is really worth it!
There will be an Open AI / Anthropic leak eventually. One can only hope.
I wish we had competent regulators. Force the scans to be purchased through an official repository. Save the scans and sell them to other companies who also want to train models. Force the companies to make all book requests through the official repository. Give some portion of money paid back to the publishers. This is a solvable problem.
I didn't think about this happening. It reminds me of the cross-border challenges with crypto. Each country has laws to control its author rights or money supply, but those are hard to enforce in the international setup.
Can't these genius AI companies spend a few pennies to develop a non-destructive book scanner?
What's the use of all this AI if they can't figure out this relatively simple task.
My understanding was that they had to legally destroy the books after they processed them for whatever copyright strangeness was involved.
no, because that's cost and everything needs to be the cheapest possible. only valid spend is money going to the shareholders. and chopping off the spine and putting the stack of paper into an automated scanner saves sooo much money over a $5/hr employee flipping pages all day
i think that they might actually want to do the ram gambit all over again - buy up these rare books so that competitors can't get them. destroying them after scanning makes doubly sure that the competition won't get their hands on it
When I hear these stories of AI companies buying all the books, I think back to Kevin Kelly in 2008 or 2009. He talked about how books are less expensive than at any other time in history and easier to buy and that could change so it makes sense to buy a lot of books.
> [...] I was near to the point of actually digitizing and getting rid of all my paper books.
> I was that close about five years ago, but then I had an epiphany. I went to private library, and I realized that books were never as cheap as they are today. They never will be as cheap, and that there's some power about having these things in paper always available, no batteries, never obsolete, and that if you made a library now, you would never be able to make some of these libraries in 50 years, so I decided to keep and to cultivate this paper library as something that was going to be powerful in the future.
I remember seeing photos of his library but can't find them anymore. This is the only thing I could find:
https://colossus.com/article/flounder-mode/
Am I supposed to know who that is?
The second paragraph in that story lays out the main bullets.
He's worth paying attention to. For the past thirty years or so that I've been aware of him he's been more prescient than most.
Is this a second form of AI psychosis, as in AI training corpus psychosis?
The need for all the content, Moar!! Feed me. Is this the Paperclip Maximiser in the form of the Training Material Maximiser?
Will it be that, in the end, the lack of "Pass Your Driving Test, 2018 Edition" was the cause of driving rule hallucinations in all previous models? Will this finally get OpenAI back in front of Anthropic? Hurrah! We found it!
> The need for all the content, Moar!! Feed me
"MOAR input!!!"
- Johnny 5, Short Circuit. That was a great film. Seems it was unexpectedly prescient too.
https://www.youtube.com/watch?v=WnTKllDbu5o
As a teenager I remember I was amazed at the speed at which Johnny 5 would consume that book.
I was thinking the inverse the psychosis is that someone who sells books is alerted that they are selling more books, as if someone buying books needs to give an explanation. If I want to seed software with books why is that newsworthy? Print more books, maintain profitable margins on books, enjoy success as a book seller?
If a book is rare and you wanted to preserve it don't sell it so easily, increase the price, or put it into a contract system like Ferrari does with cars they sell.
Till it hits books that are out of print, we already had news about AI scourge companies destroying rare books coz scanning them in destructive ways was slightly cheaper.
And as is happens in history, any time some entity mass destroys the books, they turn out to be bad guys
There's a very simple solution to this, but the people who rabidly argue for a copyright length of 70 years + the lifetime of the author won't want to hear it.
Your local school throws away books by the dumpster load once a year or every other year.
But anyone can mistreat or fail to care for a book, if it's rare and you don't want to risk it don't sell it.
if they are going to just violate copyright blatantly and price is no object
why not just every textbook for every subject in high school and college?
Not exactly top journalism...