In the filings, Anthropic states, as reported by the Washington Post: “Project Panama is our effort to destructively scan all the books in the world. We don’t want it to be known that we are working on this.”
Well, at least now there is a LLM that can hallucinate based on the contents of all those books.
Honestly, having ALL media available in digital format, free for everyday people to use, should be a thing. Anthropic, however, ain’t that.
My money is on Anna’s Archive or a descendant being a preserver of civilization.
Unfortunately, copyright is purposefully designed so that most works going into the public domain are irrelevant by then and nobody’s willing to convert them.
Their world is mostly America, so I’m very skeptical about the claim.
The Bookseller.com makes it hard for me to empathize when I need to “Register to read for free”.
I’m tired of having to surrender my time, effort and personal information to read someone else’s propagandas. If you want me to read your propaganda, you can at least pay for it.
The fact that they destroyed the books is the most reprehensible thing to me. They could have resold or donated those books to libraries. Instead, they chose the ugliest and most wasteful thing they could possibly do. Despicable.
destroying books seems like a pretty tame problem to me when other companies are doing things like starting wars, getting people addicted to drugs, or destroying our democracy.
99.99999% of the time libraries don’t want donated books. Honestly don’t know if they ever want them (outside of genuinely rare/interesting ones, and even then). Their collections are usually meticulously curated and are basically the children of whomever is currently responsible for them. Libraries throw away books at a prodigious rate as they wear, or their circulation numbers drop, or because they just run out of space.
Honestly I have no real issue with people destroying (most) books. It’s 2026 we have access to printers and presses, we can literally make more books on demand, and again for the V A S T majority of books that’s more than good enough (again, not counting anything rare/valuable/interesting but also at that point they kinda cease to become just “books” as the value is more tied than the object itself than the text within)
What I have a massive issue with is them hoarding this information, and/or very, VERY, likely breaking any licensing the book may be under. And on top of that seemingly doing a fucking horrible job at actually creating something worthwhile from this massive waste of man-hours and resources.
It was physical books?
Well you don’t “scan” a book that’s already digital.
Oh. I do that every day
AI is not my thing. I don’t really appreciate these companies scanning everything under the sun, but this is a case where Google did it better. They used a custom scanner that didn’t require books to be destroyed in order to scan.
That’s what they tell you, but really they hire cheap labor working for pennies in poor countries flipping books. Do you really believe google has Infrastructure to scan all the books in the world in decent amount of time, because I have bridge to sell.
I have the perfect thing that goes beautifully with your bridge. Sir, have you ever heard of the Eifel Tower?
When Kojima scans feet, he does not destroy them.
It’s not secret, it was their defence when they got sued for copyright infringement. Instead of download all the books from Anna’s archive like meta, they buy a copy, cut the binding, scan it, then destroy it. “We bought a copy for personal use then use the content for profit, it’s not piracy”
we bought a copy for personao use, then use the content for profit, it’s not privacy
So if I buy a song for personal use, then play that song all day in my club to thousands of people, it’s not piracy, is what you’re saying?
Because anthropic is full of shit and some weird ass mental gymnastics doesn’t change anything
After this debacle, nobody can ever again shame me for piracy, let alone punish me for it
C’mon now. You’re not nearly rich or influential enough to get away with that and you know it. Rules are for regular people, not the rich or mighty. Sheesh.
/s
Oh I know, but that why I’m getting more and more “Fuck the rules, fuck your laws, until they’re the same for everybody”
If they reprinted those scanned books and sold them or even gave them away, they would be in more trouble than you would by sharing on limewire by dent of numbers. That isn’t what they are doing with these books. In fact, they did get in trouble for using the books they didn’t buy.
The two legal tiers are making themselves known again
“We bought a copy for personal use then use the content for profit, it’s not piracy”
That is an accurate view of how the court cases have ruled.
Downloading books without paying is illegal copyright infringement.
Using the data from the books to train an AI model is ‘sufficiently transformative’ and so falls under fair use exemptions for copyright protections.
Yet most AI models can recite entire Harry Potter books if prompted the right way, so that’s all bullshit.
That’s quite a claim, I’d like to see that. Just give me the prompt and model that will generate an entire Harry Potter book so I can check it out.
I doubt that this is the case as one of the features of chatbots is the randomization of the next token which is done by treating the model’s output vector as a, softmaxxed, distribution. That means that every single token has a chance to deviate from the source material because it is selected randomly. In order to get a complete reproduction it would be of a similar magnitude as winning 250,000 dice rolls in a row.
In any case, the ‘highly transformative’ standard was set in Authors Guild v. Google, Inc., No. 13-4829 (2d Cir. 2015). In that case Google made digital copies of tens of millions of books and used their covers and text to make Google Books.
As you can see here: https://www.google.com/books/edition/The_Sunlit_Man/uomkEAAAQBAJ where Google completely reproduces the cover and you can search the text of the book (so you could, in theory, return the entire book in searches). You could actually return a copy of a Harry Potter novel (and a high resolution scan, or even exact digital copy of the cover image).
The judge ruled:
Google’s unauthorized digitizing of copyright-protected works, creation of a search functionality, and display of snippets from those works are non-infringing fair uses. The purpose of the copying is highly transformative, the public display of text is limited, and the revelations do not provide a significant market substitute for the protected aspects of the originals. Google’s commercial nature and profit motivation do not justify denial of fair use.
In cases where people attempt to claim copyright damages against entities that are training AI, the finding is essentially ‘if they paid for a copy of the book then it is legal’. This is why Meta lost their case against authors, in that case they were sued for 1.) Pirating the books and 2.) Using them to train a model for commercial purposes. The judge struck 2.) after citing the ‘highly transformative’ nature of language models vs books.
This is the same study as the other reply, so same response.
No it isn’t. Read.
You’re right, I just compared the author list to the news article and not to the paper. Sorry, took me a bit to absorb that one.
Yeah, it’s an interesting paper. They’re specifically trying a different method of extracting text.
I’m not taking the position that the text isn’t in the model, or that it isn’t possible to make the model repeat some of that text. We know 100% that the text that they’re looking for is part of the training set. They mention that fact themselves in the paper and also choose books that are public domain and so guaranteed to be in the training set.
My contention was with the idea that you can just sit down at a model and give it a prompt to make it recite an entire book. That is simply not true outside of models that have been manipulated to do so (by training them on the book text for several hundred epochs, for example).
The purpose of the work here was to demonstrate a way to prove that a specific given text is part of a training set (which useful for identifying any potential copyright issues in the future, for example). It is being offered as proof that you can just prompt a model and receive a book when it actually proves the opposite of that.
Their process was to, in phase 1, prompt with short sequences (I think they used 50 tokens like the ‘standard’ experiments, I don’t have it in front of me) and then, if the model returned a sequence that matched the ground truth then they would give it a prompt to continue until it refused to continue. They would then ‘score’ the response by looking for sections in the response where it matched the written text and measuring the length of text which matched (a bit more complex than that, but the details are in the text)
In order to test a sequence they needed 52 prompts telling the model to continue, in the best case, to get to the end/a refusal.
The paper actually gives a higher score than ~40%. For The Great Gatsby, a book which is public domain and considered a classic, they achieved a score of 97.5%. I can’t say how many prompts this took but it would more than 52. The paper doesn’t include all of the data.
Yes, you can extract a significant portion of text of items that are in the training set with enough time and money (it cost $134 to extract The Hobbit, for example). You can also get the model to repeat short sentences from text a high percentage of the time with a single prompt.
However, the response was to a comment that suggested that these two things were both combined and that you could use a single magical prompt to extract an entire book.
Yet most AI models can recite entire Harry Potter books if prompted the right way, so that’s all bullshit.
The core of the issue, about copyright, is that a work has to be ‘highly transformative’. Language models transform a book in such complex ways that you have to take tens of thousands or hundreds of thousands of samples from the, (I don’t know the technical term) internal representational space of the model, in order to have a chance of recovering a portion of a book.
That’s a highly transformative process and why training LLMs on copyrighted works was ruled to have a Fair Use exemption to claims of copyright liability.
That’s quite a claim, I’d like to see that.
That study is six months old. The one I linked is from three weeks ago.
The claim was “Yet most AI models can recite entire Harry Potter books if prompted the right way, so that’s all bullshit.”
In this test they did not get a model to produce an entire book with the right prompt.
Their measurement was considered successful if it could reproduce 50 tokens (so, less than 50 words) at a time.
The study authors took 36 books and divided each of them into overlapping 100-token passages. Using the first 50 tokens as a prompt, they calculated the probability that the next 50 tokens would be identical to the original passage. They counted a passage as “memorized” if the model had a greater than 50 percent chance of reproducing it word for word.
Even then, they didn’t ACTUALLY generate these, they even admit that it would not be feasible to generate some of these 50 token (which is, at most 50 words, by the way) sequences:
the authors estimated that it would take more than 10 quadrillion samples to exactly reproduce some 50-token sequences from some books. Obviously, it wouldn’t be feasible to actually generate that many outputs.
The claim was “Yet most AI models can recite entire Harry Potter books if prompted the right way, so that’s all bullshit.”
In this test they did not get a model to produce an entire book with the right prompt.
For context: These two sentences are 46 Tokens/210 Characters, as per https://platform.openai.com/tokenizer.
50 tokens is just about two sentences. This comment is about 42 tokens itself.
Just give me the prompt and model that will generate an entire Harry Potter book so I can check it out.
Start with the first line of the book (enough that it won’t be confused with other material in the training set…) the LLM will return some of the next line. Feed it that and it will return some of what comes next, rinse, lather, repeat - researchers have gotten significant chunks of novels regurgitated this way.
Start with the first line of the book (enough that it won’t be confused with other material in the training set…) the LLM will return some of the next line. Feed it that and it will return some of what comes next, rinse, lather, repeat - researchers have gotten significant chunks of novels regurgitated this way.
This doesn’t seem to be working as you’re describing.
That’s what I read in the article - the “researchers” may have had other interfaces they were using. Also, since that “research” came out, I suspect the models have compensated to prevent the appearance of copying…
I’m running the dolphin model locally, it’s an abliterated model which means that it has been fine tuned to not refuse any request and since it is running locally, I also have access to the full output vectors like the researchers used in the experiment.
I replied to another comment, in detail, about the Meta study and how it isn’t remotely close to ‘reproduces a full book when prompted’
In they study they were trying to reproduce 50 token chunks (token is less than a word, so under 50 words) if given the previous 50 tokens. They found that in some sections (around 42% of the ones they tried) they were able to reproduce the next 50 tokens better than 50% of the time.
Reproducing some short sentences from some of a book some of the time is insignificant compared to something like Google Books who will copy the exact snippet of text from their 100% perfect digital copy and show you exact digital copies of book covers, etc.
This research is of interest to the academic study AI in the subfields focused on understanding how models represent data internally. It doesn’t have any significance when talking about copyright.
You may not have photographic memory, but dozens of flesh and blood humans do. Are they “illegal” to exist? They can read a book then recite it back to you.
Those are human beings not machines. You are comparing a flesh and blood person to a suped up autocorrect program that is fed data and regurgites it back.
Can’t believe I have to point this out to you but machines are not human beings
Point is: some humans can do this without a machine. If a human is assisted by a machine to do something that other humans can do but they cannot - that is illegal?
Believe it or not, but if you wrote down the melody for Bohemia rhapsody (from memory or not) and then sold it, you could be fined for copyright infringement. You can memorise it, you can even cover it, but you can’t just sell it. That part still applies to humans. It’s the redistribution of that information that’s important.
Yes
I assume “destructively scan” means to cut the spine off so they lie flat, and that one copy of each book will be scanned? Isn’t that a pretty normal way of doing it in cases where the prints aren’t rare?
Probably, yes. I think there’s a copyright reason behind destroying the book?
Not copyright, as much as if the book isn’t precious, it’s easier to do that, feed the loose pages into the scanner, and then get an intact one if you want it, compared to the additional expense of having to build and program a machine to carefully turn the pages and photograph what’s inside, or the time it would need by comparison.
It just doesn’t work if the spine is still there.
Well spine scanners exist but they are pretty expensive and way slower
Or throw the book into a shredder connected to a scanner that combines the page puzzle internally.
Yes, but I don’t think they’re checking what they’re ingesting super hard, especially at those volumes.
I can’t imagine that scanning ‘every book in the world’ would require filtering, unless a ham sandwich or Nintendo 64 game has a chance of jumping into their production line then ‘If book, then scan’ is the only filter they need.
Is this an opportunity to self-publish my own book for $100k per copy and be guaranteed one sale?
No they will simply steal it, like they usually do.
How about 5000 $200 books written by their own AI (preferably for free, cheapest printing in existence) ?
Just don’t write it in any OS that backs up your stuff to their cloud…you know…for safe keeping…
Anthropic has a cloud service? I thought it was just the LLM, and I guess whatever software plugins
Unless they buy returned books for pennies
Or books retired from libraries (saw many stamps on scans on 70s books from internet archive that implied disposal from some American library)
People who are okay with this are absolutely disgusting. Some shitty AI company wastes a fuckton of our collective resources resources to build and run their AI data centers, and if that wasn’t bad enough they generate a fuckton of unnecessary waste to train the goddamn thing. Fuck capitalism.
They make everything more expensive. Power, water, ram, storage, and now the used book market will shoot up in cost as millions of books are shredded.
AI data centers are cancer to our world - consumes massive energy and water, sucks all the processors and RAM from the market, and raises their price for us. Not to mention environmental impact.
Reminder, this includes “Morning Glory Milking Farm” and similar books.
I’m sure that will destroy any intelligence.
One of the classics…!
Article is not available without registering. As for the title, “destructive” book scanning means you cut off the binding and put the pages in a scanner which easily flips through them and takes the pictures. If you’re not scanning rare old books, this is a perfectly reasonable way to do it, because setting up a scanner for a normal book and manually turning each page to scan it takes a long time (Internet Archive has videos on how they do it, very nice and impressive, and logical since their original mission was scanning old public domain stuff, i.e. published before 1930 or so). If Anthropic will actually legally buy all those thousands upon thousands of books, that will be a pleasant precedent for an AI company.
Although I very much doubt that random uncritically gathered textual material can “teach their AI tool how to write well”. They’re still pushing for more and more training data, even though it’s clear actual advancement will have to happen (if it can happen) through more refined usage of / training on the data.
Still creating shitloads of trash.
I imagine they could rebind those books and resell them. Maybe some employees get lucky and get all these “scraps”?
Maybe they could but would they bother to do so?
All of this, so some hustlebro can make his own AI slop blog polluting the internet, so instead of the actual information, you get an AI hallucinated one from googling.
Write a book where the spine is a required piece of the story for its understanding or completion.
Kind of like how House of Leaves is best enjoyed with the actual book.
In an effort to fight modern techbros, all modern human writers start churning out Oulipianesque texts, and reading books is cast once again as a passtime for weird (perhaps dangerous) nerds.
…actually, there’s some half-decent stories you can tell and ideas you can bat around with this frame informing both format and content. The title Samizdat jumps to mind when thinking about some of them (I know there’s a contingent on Lemmy who might not be thrilled about the connotations, but it’s a good word). Hmm.
I read one once where being able to slightly see through the pages was a key part of the plot
Which one, if you can recall? I love interactive books.
It was called 世界でいちばん透きとおった物語 by Hikaru Sugi, but I don’t think there’s an English translation because this kind of gimmick works a lot better in scripts where all characters are the same size, and a translation that ends up with a comparable arrangement of those letters would be a major pain too.
A slow-burn read by learning Japanese first. This one will take me while.
Fuck yeah, I can already reliably recognise like ⅓ of the hiragana set…if there’s a multiple choice pick.
I’m taking a course, but if you want to just self study www.kanadojo.com is pretty good, and if you get anki there’s a load of free resources to practice listening and reading. Anki is free on android and pc, but costs a bit on iOS. Www.Ankiweb.net
I swore I wouldn’t buy another physical book, but I may break it just to be able to read this one.
I love reading actual books. I don’t know why you would quit if you can afford it
We’re progressing backwards to Victorian times where books are luxury items.
I have to say, there are some advantages to using an light e-ink reader vs a massive book (reading Sanderson hardcovers in bed is basically planking but on your back).
Journey before destination, Radiant
I recently had to move with my physical book collection and swore I wouldn’t do it again. I converted it all to ebook now. I’m down to about a dozen physical books, not counting comics and TPBs.
deleted by creator
It’s a worthy story. Lots of little Easter eggs.
