Comment by faidit
7 days ago
How many AI companies are doing this? Say there are only 5 copies of a book left out there but 100 different companies want to shred it. Rare items could effectively vanish from the market. "Rare" meaning inclusive of high-quality items, disregard the bot opinion that mass book destruction is fine because most books are worthless anyway. And are any safeguards in place to make sure scanned books get saved in their original form for posterity before being recycled into trainingslop? Aside from Google's quasi-legal/ethical mass book scanning operation of course.
> "Rare" meaning inclusive of high-quality items,
This is the citation needed that is missing from every report so far, including this one which deliberately refuses to reveal anything about these books.
You’d think if there were examples of actually valuable, rare books being shredded that the journalists would at least be able to name one such example. Instead it’s always vague posting about the destruction without ever naming any examples.
I think it’s because if they named some example titles, everyone would see that they don’t care about these books being shredded.
On the contrary these companies should be publicly listing the name/info of every book they destroy and use as training data.
If the books are really worthless as you say then their case would be proved transparently.
This article presumably had to maintain confidentiality to protect the seller who agreed to place a tracking device in the shipment.
I would say the burden of proof is on the companies destroying human cultural heritage en masse, not the handful of journalists calling for attention to the matter.
> I would say the burden of proof is on the companies destroying human cultural heritage en masse, not the handful of journalists calling for attention to the matter.
You're literally assuming the conclusion. The exact topic under contention is whether they are "destroying human cultural heritage".
I do work adjacent to the AI book scan-shred pipeline. There are definitely significant books that aren't "Windows 95 for Dummies" which are getting down to single-digit remaining copies.
I just looked for one novel, which wasn't a fantastic book, but it is the first use of a pithy and fun phrase that is so ubiquitous that you'll probably read it a couple of times today. I argued with Claude, GPT and Gemini for ten minutes just now, even knowing the title of the book, to even prove the book exists. It took me years to find a copy originally and then I lost it in a move. I found one more copy today from a rare book seller, but it just sold (to Amazon?).
Is the book valuable? Not particularly, but I feel it's noteworthy and important. I don't want to name it either, because now I have some searches out and the next copy that pops up I'll scan and put on IA. There can only have been a few thousand copies originally published in 1947, it's only in hardcover. I know of a couple of other copies in private hands, so it's not zero copies, but it has to be single-digits.
I have one periodical issue that I know of only one other existing copy (Worthpoint only shows one copy ever sold in their database) and if you look on collector sites there is a blank because nobody even knows what the cover looks like. I can't explain it, since the publication routinely printed hundreds of thousands of copies of each issue, but here we are. Perhaps all the copies were withdrawn and pulped immediately after publication for some reason? It's in my scan pile, so I'll have it uploaded soon. Is it significant? Not hugely, but every other issue of this title has been scanned already, so it's scratching an itch to get this one done.
There's definitely rare stuff getting scanned and shredded. Someone in a comment above said it's not like Nazi book-burning since they were trying to destroy information. But it is like that if you consider there remain no other physical copies and all the electronic copies are locked up in a way that nobody can access except to trick an LLM to spit out a paraphrased copy from its training data.