Bruh, this is literally like scientists scavenging metal from shipwrecks that predate the era of nuclear bomb testing, cause they need steel that isn’t contaminated by nuclear fallout for certain measuring equipment 😵💫😵💫
Great analogy for the shit we’re currently saturating our digital world with
Low background bullshit training data.
Tell me more about nuclear steel.
It’s actually the low radiation stuff from before the first nuclear bombs were detonated that we want. Turns out those kinds of bombs spread fallout and now literally everything on the planet has a trace amount
nice
TIL
Fking nailed it.
The worst part is that they are destroying the books afterwards.
We also have no guarantee these books will be available at a later date in these digitized formats, nor that they won’t be altered to fit certain narratives which is of course trivial to do with digital copies. At a minimum these need public verifiable checksums to be relied upon.
I don’t like this and don’t trust it one bit.
These AI companies already showed their hand in trying to corner the PC component market with an end goal of users no longer being able to own their own PCs. Now they’re potentially trying to do this with all written recorded information.
How long until Altman’s Firemen show up at your door to burn your books?
They even hve an economical incentive to destroy the books, because by doing so, they will get the data and prevent the competition from getting it for them
The incentive is literally that it’s cheaper to scan, that’s all.
But given they’re only interested in books with ISBN’s, I recon they’re already scanned and available in digital forms. If they’re newer than the 80s, they’re most likely in a document file in a server related to the publisher.
Honestly surprised they’d rather source physical books, than just request various publishers for a price for their combined works - Only reason I can think of, is the physical scan makes other scans like a mobile photo easier to recognise and respond correctly to
Removed by mod
Literally Fahrenheit 451
Removed by mod
Reminds me of Judge Holden in Blood Meridian, who sketched ancient artifacts in his personal notebook and then destroyed the originals.
deleted by creator
…why? Why?
More that they are destroying the books in the process of digitizing them, they basically cut the pages out of the book and feed them into a scanner with a document feeder attached.
It’s worth pointing out that this is standard practice when digitising books in most places, if they’re not irreplaceable. Mass-produced books would fall under that category.
Turning the page and photographing what’s there is really only done for some books, which you can’t afford to destroy.
From what I’ve read previously, they’re destroying all the books, regardless if they’re historic and one-of-a-kind or a mass produced paperback.
Do you think they’re discriminating? Really?
At least in this case, yes. They’d just not get them, since they’d want something they can already easily process, and the source for the quote is a bookseller, who isn’t likely to sell something like that to begin with.
An old book that might disintegrate and damage the scanner and hold things up would be undesirable, compared to either a new one, or a reprint of an old one.
That’s not how most of that works.
Things would be sold in bulk. Mistakes happen. Literally the only part of what you said that makes any sense is that fucked up pages could clog the pipe.
Is it really destroying though? They’re digitizing them, and publishers still have the digital copies ready to print more at any time. So it’s not like they’re destroying the texts, they’re just shifting them.
Nobody complained when Google did this over a decade ago 🤷
When you say they’re “destroying the books” you make it sound like they’re erasing one of the last known copy of some important work when in reality, most of these books were purchased in bulk from bookstores and libraries that were planning on discarding them anyway.
Almost all these books were either headed to the dump or the recycling center. They’re just being digitized on the way.
Digitized for private consumption.
Nobody cares when Google did this or when archive.org does this, because they’re sharing the results with the world. (Idiotic shortsighted lawsuits from the authors guild notwithstanding)
in this case they often are actually destroying them. they take them apart, beause it’s easier to scan than to use a proper book scanner
Yeah that’s true of Google as well I believe. I think Archive is more careful, but I’m sure some books get damaged in the process there as well.
Having done a lot of scanning in the past, no matter how gentle you are, the book will be damaged; at least a little bit. Even if you do it manually, by hand, really slowly.
It’s the unfortunate reality, because books are not designed to be held open and pressed flat onto an unyielding surface. If you don’t press flat, you’ll get curved pages and bad quality (dark) images, which is sometimes ok and fixable by software, but often not. You can get really fancy expensive machines, but they still don’t fully solve the problem.
I did scanlation editing stuff for a while, and half my job was making the raw page scans look presentable, because the scanners didn’t want to destroy their manga (understandable).
This matches my experience. My college archive digitizes a few yearbooks a year (for whatever classes are having as big reunion) and it’s a pain. For modern yearbooks (70s and later) we have a dozen or more copies, so we take one apart and feed it through the feed scanner, which works well. We then store the loose pages in a folder, in case we need to rescan anything.
Older yearbooks (where we don’t have so many copies) we use the flatbed scanner and the pages come out warped, so we hand fix each page in Photoshop.
Yepp… memories of using the light level curve tool and white point and black point and the perspective tool. Then going in with the white brush and deleting any specks of dust that remained, and comparing to the original to make sure no lines got deleted
Google (and/or the libraries that collaborate with Google Books) and Archive.org usually scan non-destructively. Some of Google’s book scans still show the fingers holding the corners of the pages, some of them were taken mistakenly mid-flip, etc. So they clearly used whole, normally bound books.
When they destroy the physical copy they remove them from the antique stores market which often rely on circulation.
Edit: Also physical copies don’t require electricity, a device and internet access plus they are something you can own.
If you think any more than 0.1% of these physical books would ever have ended up in antique bookstores, you’re dreaming.
Think about how many books out there are things like Donald Trump’s biography, or pointless drivel from non-experts, self-help books that tell people to down “essential oils”, old editions of programming books, or just plain shitty fiction that never sold much in the first place.
It’s ok to throw trash away! Really!
At least digital storage prices have only been going down in recent months, right? /s
physical copies don’t require electricity
Good luck reading by candle light.
I guess you must live in that part of Alaska where it’s night time for half the year and don’t have daylight.
More just the part of Alaska where I work most of the daylight hours and get my reading in before bed.
Almost all these books were either headed to the dump or the recycling center.
My wife works at a law firm that contracts with “Books By The Yard”, which provides books purely for office aesthetics. For a few hundred bucks you can plaster a bookshelf full of material nobody will ever read, because they’re such a commodity.
It’s so crazy to see people ingesting this news from the conspiracy-brain perspective of “The AI companies are stealing all the knowledge!” without recognizing the more pressing reality of decade upon decade of publisher overproduction, resulting in a total devaluing of physical media for its academic importance.
Imagine going into hysterics because a warehouse full of shitty airport books went up in smoke, like it’s the Library of Alexandra that just burned down.
I imagine they were talking about destruction in a practical sense. Disassembling the book and scanning it like that is faster and cheaper than purpose build book scanning machines.
Additionally, court documents indicate they generally just throw them away afterwards. So the knowledge is retained, but that book is destroyed.
https://tagteam.harvard.edu/hub_feeds/3415/feed_items/14341220
Nobody complained when Google did this over a decade ago 🤷
Yes they did?! That was a big controversy back in the day. You are engaging in historical revisionism right now.
Almost all these books were either headed to the dump or the recycling center. They’re just being digitized on the way.
Where does it say that in this article?
ship of theseus ah shit
They’re doing everything they can to avoid it, but it’s supposed to be tHe FuTuRe
Damn AI ruined AI.
Hee hee.
Modern book burning
Destroying the books too :(
Not to defend the AI crowd, but ever go to a Goodwill Outlet store? They usually have at least 1 gaylord box (the giant box on a pallet watermelons come in) full of books. Not sure how often they cycle those out, but I’m guessing pretty regularly. Multiply that by every semi major city Goodwill has a presence in. Add in other thrift sores with a similar quantity of books. You’re talking ALOT of books. SO MANY books are thrown out everyday. Including by libraries. I don’t really have a problem if some of those get diverted for a bit to be scanned. Yes, it’s sad to see books getting destroyed. But there’s alot of stuff people just really don’t want. And, if you REALLY want a villain, maybe you can blame copyright law? That’s part of the reason AI companies destroy the books - it helps to make their scanning efforts more legal. Absent that, they could probably be convinced to spend just a little more for nondestructive scanning, assuming they could find a home for the books they scanned (the sellers of these things couldn’t, so good luck).
If the AI companies were simply picking up books that were already going to a landfill, I don’t think that would bother anybody. That isn’t what we’re talking about here, though. These books are coming from booksellers, so it’s fair to guess that most of them are at least a notch or two above what you’d find in a Goodwill bargain bin.
Bibles. There are so many fucking bibles. When I worked for a church we probably had 20 people a year buy a while new set of new bibles for the church. Like - we already had like 3,000 bibles with a weekly attendance of about 500. We didn’t need new bibles, but people just kept giving them to us.
It’s the most-printed book of all time, and they just don’t stop making them. It’s insane.
I mean even when you go to Goodwill, it’s not like 99% of the books are rare or irreplaceable. How many copies of Patterson books does the world need to begin with?
You can only stave off model collapse so long.
It’s only a matter of time before most of the data going in to LLMs came out of one too.
Like I’m not even good at math or anything, much less computer science, but that seems obvious and inevitable to me.
Maybe someone smarter than me can chime in.
It’s kind of sort of already happening. Emdashes are fragments of it, a token saving strategy introduced by already trained on old books in their data sources. Model collapse is inevitable.
Idk. Prose books generally have tons of em dashes.
Yes exactly the point being made in the link. AI picks up unintended behaviours already, it’s not too hard to imagine that the general direction is towards model collapse.
Anna’s Archive: …uh, well okay, then.
I’m waiting for the day when desperate LLM companies start paying people to post real human content, only for those people to just ask ChatGPT to do it.
Already happening for years.
Amazon mechanical turk, Outlier.ai, scale ai, are all gig platforms that pay people to produce real human content for LLM training.
At first they were paying people to actually produce content, like worked solutions to maths problems or translations, then they pivoted more to rating LLM output.
And as with every enshittification cycle, wages and work rapidly dried up, so people responded by asking LLMs for the answer just to meet deadlines and get by.
“now you only posted 4 times today Jenny, do you just want to do the minimum?”
This is already happening.
They are paying people bottom market rates to generate unique written content and people are just having AI write it for them.
Some kind of AI ouroboros situation.
(Non)Human centipede.
Reverse centaur centipede?
It’s simultaneously unethical as an employee and based to be feeding AI it’s own tail
more than likely they are scraping other AI bot posts, there isnt that many users on most of the social media sites. its 50/50 AI/BOTS for most of them.
Supposedly the people paid to rate responses are already using AI
Big AI mostly switched to synthetic training data anyway. The books they’re digitizing are being used to gather knowledge, not writing styles or logic (mostly).
As in, when you ask ChatGPT how long some book is, it can just go check (if it’s in the database). It’s also useful if you ask about that book or about knowledge contained in that book. It’ll even reference books now (if you demand that in your prompt).
It’s not the same as earlier LLM tech which relied on scanned text to figure out how to respond to any given prompt (from a language standpoint). The “language” part of LLMs is a solved problem now (thanks to the synthetic training). At least for English 🤷
The article quotes a post from ISBNdb saying the issue is model collapse from training on synthetic data.
Total collapse is a solution.
It’s also unavoidable.
Yes, what’s the issue?
That’s like saying, “they had some failure modes from the synthetic data, so they should just obviously stop trying forever.”
They’ll just fix the edge cases and move on. Like any programming task.
The problem is that synthetic data is not fit for that purpose. The more of it you use, the worse at dealing with the edge cases LLMs get.
Think of it like this… You feed a language model a bunch of genuine human-written content. Great. Now it can produce the most likely text in a lot of cases. Word combinations that rarely appear in written language rarely get generated, so most of its synthetic data lacks those rare - but still valid - combinations.
Train it on this synthetic data, and now more outliers and rare combinations get filed off. Rinse and repeat.
It’s like saving a lossy file over and over again. It degrades each time.
Nice that they’re actually buying them this time instead of just downloading torrents of book collections again.
I imagine they already burned through literally all of the torrentable books
That would be all the books.
Far from it. I think you vastly underestimate how many books have been published, and in what small quantities.
Estimates are around 15% of books is digitised, and less than a third of that is searchable. I’m guessing a small fraction of that is torrentable (yes that’s a word I just made up).
Ah fuck… No wonder old maths texts are a nightmare to get hold of recently…
They first crashed RAM prices, the SSDs, and now books too?! Oh fuck 😭
Why? So their LLM’s can ingest the books and use them to churn out more AI slop?
Don’t you know that the solution to any and all problems with AI just need more training data to fix? If we can teach it to be just a little more human, it’ll surely stop going off the rails. We just need to feed the machine more data More Data MORE DATA MORE DATA
they arnt already? some yt videos churns AI scifi
Yeah, pretty awful on so many different levels.
Brilliant decision to train their models on the worst content to begin with.
deleted by creator
