Authors using a new tool to search a list of 183,000 books used to train AI are furious to find their works on the list.
I certainly hope that none of these authors have ever read a book before or have been inspired by something written by another author.
That would be a much better comparison if it was artificial intelligence, but these are just reinforcement learning models. They do not get inspired.
just reinforcement learning models
…like the naturally occuring neural networks are.
The brain does not work the way you think… (I work in the field, bio-informatics). What you call “neural networks” come from an early misunderstanding of how the brain stores information. It’s a LOT more complicated and frankly, barely understood.
Yeah, accurately simulating a single pyramidal neuron requires an eight-layer deep neural network:
that was an interesting read, thank you
It’s a LOT more complicated and frankly, barely understood.
Yet you confidently state that the brain doesn’t work the way LLMs do?
Obviously it doesn’t work exactly the same way that LLMs do, if only because of the completely different substrates. But when you get to more nebulous concepts like “creativity” and “inspiration” it’s not so clear.
people have a definite fear of being defined as machines… not sure why we think were so special…
Tell you what, you get a landmark legal decision classifying LLM as people and then we’ll talk.
Until then it’s software being fed content in a way not permitted by its license i.e. the makers of that software committing copyright infringement.
What exactly was not permitted by the license? Reading?
Using it to (create a tool to) create derivatives of the work on a massive scale.
An AI model is not a derivative work. It does not contain the copyrighted expression, just information about the copyrighted expression.
Wikipedia: In copyright law, a derivative work is an expressive creation that includes major copyrightable elements of a first, previously created original work.
I think you may be off a bit on what a derivative work is. I don’t see LLMs spouting out major copyrightable elements of books. They can give a summary sure, but Cliff Notes would like to have a word if you think that’s copyright infringement.
Well when that happens we have laws. So no problems
Would you be okay with applying that argument for any crime?
More to the point: they replicate patterns of words.
That’s an interesting take, I didn’t know software could be inspired by other people’s works. And here I thought software just did exactly as it’s instructed to do. These are language models. They were given data to train those models. Did they pay for the data that they used to train for it, or did they scrub the internet and steal all these books along with everything everyone else has said?
Well, now you know; software can be inspired by other people’s works. That’s what AIs are instructed to do during their training phase.
Does that mean software can also be afraid, or angry? What about happy software? Saying software can be inspired is like saying a rock can feel pain.
Software can do a lot of things that rocks can’t do, that’s not a good analogy.
Whether software can feel “pain” depends a lot on your definitions, but I think there are circumstances in which software can be said to feel pain. Simple worms can sense painful stimuli and react to it, a program can do the same thing.
We’ve reached the point where the simplistic prejudices about artificial intelligence common in science fiction are no longer useful guidelines for talking about real artificial intelligence. Sci-fi writers have long assumed that AIs couldn’t create art and now it turns out it’s one of the things they’re actually rather good at.
Software cannot be “inspired”
AIs in their training stages are simply just running extreme statistical analysis on the input material. They’re not “learning” they’re not “inspired” they’re not “understanding”
The anthropomorphism of these models is a major problem. They are not human, they don’t learn like humans.
They weren’t given data. They were shown data then the company spent tens of millions of dollars on cpu time to do statistical analysis of the data shown.
A computer being shown data is a computer being given data. I don’t understand your argument.
Are you saying the writers of these programs have read all these books, and were inspired by them so much they wrote millions of books? And all this software is doing is outputting the result of someone being inspired by other books?
Clearly not. He’s saying that other authors have done the same as the software does. The software creators implemented the same principle into their llm. You are being daft on purpose.
It’s not the same principle. Large language models aren’t ‘inspired’ to write new works. Software can’t be inspired. It follows instructions. Even though large language models might feel like somebody is talking back to you and giving you new information, it’s just code following instructions designed to predict output based on the input provided and the data supplied. There’s no inspiration to be had, and to attribute inspiration to language models is a huge mischaracterization of what’s happening under the hood. Can a language model, without being told what to do, actually use any of the data it was fed to create something? No. Every single large language model requires some sort of input from a user to act as a seed before any sort of response can begin.
This is why it’s so stupid to call this shit AI, because people start thinking it’s actual intelligence. Really, It’s just a fancy illusion.
They purchased their books to get inspiration from, the original author gets paid, and the author consented to selling it. That’s the difference.
Also the LLM can post entire snippets or chapters of books, which of course you’ll take at face value even if it hallucinates and makes the author look like a worse author then they are.
These are machines, though, not human beings.
I guess I’d have to be an author to find out how I’d feel about it, to be fair.
Machines that aren’t reproducing or distributing works
If an AI “reproduces” a work it was trained on it is a failure of an AI. Why would anyone want to spend millions of dollars and devote oodles of computing power to build something that just does what a simple copy/paste operation can accomplish?
When an AI spits out something that’s too close to one of the original training set that’s called “overfitting” and it is considered an error to be corrected. Most overfitting that’s been detected has been a result of duplication in the training set - when you hammer an AI image generator in training with thousands of copies of the Mona Lisa it eventually goes “alright, I get it already, when you say ‘Mona Lisa’ you want that exact pattern!” And will try its best to replicate that pattern when you ask it to later. That’s why training sets need to be de-duplicated.
AIs are meant to produce new things.
These are machines, though, not human beings.
What’s the difference? On the most fundamental level it’s all the same.
The same thing as with tooooooons of things: scale.
Nobody cares if one dude steals office supplies at work. Now, if everyone stats doing it, or if the single guy steals everything, then action is taken.
Nobody cares if a random person draws in the same style and with same characters as you, but if they start to sell them, or god forbid, out-sell you, then there is a problem.
Nobody cares (except police I guess) if a random driver drives double the speed limit and annoys people living next to the road on the weekends, but when tons of people do it, you get speed bumps.
Nobody cares if few people pirate movies, but when it gets to mainstream and companies notice that there might be money being lost. Then you get whatever we have now.
Nobody cares if the mudhill behind your house erodes a bit and you get mud on your shoes. Have a bunch of that erode and you realise the danger…
You have been fine-tuning your own writing style for a decade and random schmuck starts to write similarly, you probably don’t care. No harm done. Now, get an AI to write 10 000 books in a weekend and someone starts to sell them… well now you have a completely different problem.
On a fundamental level the exact same thing is happening, yet action is only taken after a certain threshold is step over.
Bingo.
A human, regardless of how many books they read, will have personal experiences that are undeniably unique to themselves. They will interpret the works they read differently from each other based on their worldly experiences. Their writing, no matter how many books they read and get inspired on, will always be influenced by their own personal lives. They can experience love, hate, heartbreak, empathy, sadness, and happiness.
This is something a LLM does not have, and in my opinion, is a massive distinguishing factor. So on a “fundamental” level, it is not the same. It is no where near the same.
do you really think we are that far off… from giving a foundational memory and motivation layers to these LLMs, that could mimic… or even… generate the generic thoughts youre indicating?
i dont think so. you seem to imply its impossibility, i expect its inevitability. the human brain will not be a black box forever… it still exists in a world of physics we can emulate, even if rudimentary.
Unless you think theres no difference between killing a person and closing a program, I think we can agree they should be treated differently in the eyes of the law.
And so theres a difference between a person reading a book and being inspired by it, and someone writing a program that automatically transforms the book in data that can create new books.
Wait. Are human beings machines?
Please do not take this as support of ai use of copyrighted works (I don’t), but as far as I can tell, yes we are machines. This rant is just me being aspie atm, so feel free to ignore it.
We are thinking machines programmed by our genetics, predispositions, experiences, and circumstances. A 2 part explanation of how humans are merely products of their circumstances was once put forward to me. The first part is that humans can do anything, but only the thing we want to do most.
For instance, a common rebuttal is that people can choose go to the gym even when they find the experience of exercise undesirable. However, when that happens, it’s merely a case of other wants out balancing the want to not go to the gym, typically they want to be fit.
We want to not spend money, but we want to not rush going to jail for stealing more, usually. We want to not work overtime, but sometimes we want the extra cash more than that.
The second part of the argument is that we can’t choose what we want. When someone talks themselves out of the slice of cheesecake, they aren’t changing what they want, they’re resolving said want against the larger want they have to lose weight.
And if we make decisions by our wants, while said wants are not decided by us, then despite appearances we are little more than complex automata.
I don’t think anyone is faulting the machines for this, just the people who instruct the machines to do it.
But terminator said neural networks
Damn.
Did you write a comment on Reddit before 2015? If so, your copyrighted content was used without your permission to train today’s LLMs, so you absolutely get to feel one way or another about it.
The idea that these authors were somehow the backbone of the models when any individual contribution was like spitting in the ocean and model weights would have considered 100 pages of Twilight fan fiction equivalent to 100 pages from Twilight is honestly one of the negative impacts of the extensive coverage these suits are getting.
Pretty much everyone who has ever written anything indexed online is a tiny part of today’s LLMs.
Thank you for your reply.
On a completely separate note, it’s funny to think that there exists Twilight fan fiction when
Twilight itself started as fan fiction work.Edit: I dun goofed.
Pretty sure it’s the other way around.
Fifty Shades of Gray started out as Twilight fanfiction before becoming its own thing.
AFAIK Twilight was always just its own pulp fiction.
Oh true! My memory was fuzzy on the details. Thanks for the correction.
Generally they probably bought the books they read though.
If George RR Martin torrented Tolkien, wouldn’t he be infringing on the copyright no matter how he subsequently incorporated it into future output?
I completely agree that the training as infringement argument is ludicrous.
But OpenAI exposed themselves to IP infringement by sailing the high seas in how they obtained the works in the first place.
I hate that the world we live in is one where so much data is gated behind paywalls, but the law is what it is, and if the government was going to come down hard on Aaron Swartz for trying to bypass paywalls for massive amounts of written text, it’s not exactly fair if there’s a double standard for OpenAI doing the same thing in an even more closed fashion.
But yes, the degree of entitled focus on the premise of training an AI as equivalent of infringing is weird as heck to see from authors drawing quite clearly from earlier works in their own output.
I have to assume that openAI also paid for the books. if yes then i consider it the same as me reciting passages from memory or coming up with derivative text.
if no, then by all means, go after them and any model trainer for the cost of one book.
Asking an LLM to recite an entire novel isn’t even vaguely a thing yet.
Well, here’s straight from one of the suits against them:
“The OpenAI Books2 dataset can be estimated to contain about 294,000 titles. The only ‘internet-based books corpora’ that have ever offered that much material are notorious ‘shadow library’ websites like Library Genesis (aka LibGen), Z-Library (aka B-ok), Sci-Hub, and Bibliotik. The books aggregated by these websites have also been available in bulk via torrent systems.”
I’m not even sure how they would have logistically gone about purchasing 294,000 books in bulk in digital form to be fed into training. Using the existing collections seems much more likely, but I suppose we’ll see what turns up in litigation.
Also, the penalty for downloading copyrighted material if willful infringement is up to $250,000 per work. So it’s quite a bit more than the cost of one book on the line…
God that Aaron/jstor thing makes me see red every time. Swartz was scraping jstor to publish it for the benefit of everyone, openai is doing it to make billions of dollars. Don’t forget who the bad guys are (and donate to sci-hub)
Any AI model that uses publically available information for training should be open source by law.
This business where corporations (that includes authors, who are published by huge corporations) fight over who “owns” ideas is assinine. When it comes down to it, this is a fight about money being wrapped in an argument about “ideas.”
AI models were developed with the collective knowledge and wisdom of society. They’re like libraries and should be public like libraries. OpenAI, Google, all those fucks should be forced to open source their models, end of story.
Trick is educating the octogenarians in the senate to understand any of what you just wrote.
One less to educate now. Hopefully replaced by someone that doesn’t need diapers.
Yup! My ideas about what should happen are so far removed from what will actually happen they could be Planet X.
But that doesn’t make me wrong, dammit!
I’d say they should have to follow the most-restrictive license of all of their training data, and that existing CC/FOSS licenses don’t count because they were designed for use in a pre-LLM world.
It seems like a pretty reasonable request. But people like free stuff, and when they think about who will get screwed by this they like to imagine that they’re sticking it to the biggest publishers of mass media.
But IRL, those publishers are giddy with the idea that instead of scouting artists and bullying them into signing over their IP, they can just summon IP on demand.
The people who will suffer are the independents who refused to sign over their IP. They never got their payday, and now they never will either.
I think we just need to ban the ability to copyright any AI output. Unless you can prove you created, and or paid for the rights for every piece of training data, I don’t see how it’s fair. Even then, there are still arguments against letting AI create IP.
The people I’m seeing outraged are artists and authors who did not sign their ideas over for public access or for disingenuous use. not a faceless publisher with cloth bags and dollar signs painted on them. Also I don’t think you understand what public and private ownership means. A person is allowed to privately own their own creation. They don’t owe that to the world. The world isn’t entitled to it.
You can do this open source right now
Here’s an idea, legally force companies like OpenAI to rely on opt-in data, rather then build their entire company on stealing massive amounts of data. That includes requiring to retrain from scratch. Sam Altman was crying for regulations for scary AI, right?
Would search engines only be allowed to show search results for sources that had opted in? They “train” their search engine on public data too, after all.
They aren’t reselling their information, they’re linking you to the source which then the website decides what to do with your traffic. Which they usually want your traffic, that’s the point of a public site.
That’s like trying to say it’s bad to point to where a book store is so someone can buy from it. Whereas the LLM is stealing from that bookstore and selling it to you in a back alley.
AI isn’t either. It’s selling statistical data about the books.
It literally shares passages verbatim
So does any site that quotes the book. Just being trained on a work doesn’t give the model the ability to cite it word for word. For most of the books in this set you wouldn’t even be able to get a single accurate quote out of most models. The models gain the ability to cite passages from training on other sources citing these same passages.
That claim is disingenuous at best, and misinformed otherwise.
That’s maybe an issue. I mirror speech a lot, though. How large are the passages?
First: There are mechanisms to opt out (robots.txt and meta noindex)
Second: There is some foreknowledge on the part of the web author. Even in the early days of the web — before you could’ve predicted the concept of search engines — in order to distribute anything you had to understand the basics of hypermedia, among which is the idea that anything can link to anything else and clients can be users or machines alike.
Third: Even though you are correct that search engines are tokenizing text and doing statistical analysis to recombine the tokens into novel forms in order to rank against queries, those novel forms are never presented to the user. Only direct quotes. So a user never gets a false reference to the supposed content of a page (unless the page itself lies to crawler requests).
Fourth: All of the technical points above are pretty much meaningless, because we are social creatures and our norms don’t stem from a mechanical flow chart divorced from real-world context.
Creators are generally okay with their content being copied into search DBs, because they know it’s going to lead to users finding the true author of those words, which will advance their creative pursuits either through collaboration or monetary support.
Creators are complaining about content being copied into LLMs, because their work will be presented out of context, often cited incorrectly, keep people away from the author of those words, and undermine the lifeblood of their creative pursuits – be it attracting new collaborators or making sales.
Whether it technically counts as IP infringement or not under current law? Who really cares? Current IP law is a fucking scam, designed to bully creators out of their own creations and assign full control to holding companies who see culture as nothing more than a financial instrument to be optimized. We desperately need to change IP law anyway – something that I think even many strident “AI” supporters agree with – so using it as a justification for the ethics of LLMs reveals just how weak the group’s position truly is.
LLM vendors see an opportunity for profit, if they can get away with it. They are offering consumers a utopian vision of infinite access to content while creating an IP chokepoint that they can enshittify once it blows past critical mass. It’s the same tactics the social media companies used 15 years ago, and it weighs heavy on my heart that so many Lemmy users are falling for it once again while the lesson is still so fresh.
This is no different than every other capitalist enterprise. The whole system works on taking a public resource, claiming private ownership of it, and then selling it back to the public for profit.
First it was farmland, then coal and minerals, oil, seafood, and now ideas. Its how the system works and is the whole reason people have been trying to stop it for the past 150 years.
The people making the laws are there because they and/or their parents and/or grandparents did the exact same thing. As despicable and corrupt as it is you won’t change it by complaining and no-one is going to make a law to stop it.
God damned right. Every “new” thing tends to be stolen. In more event history, its stolen from other capital, or from innovation with a free license, rather than artwork. Publishers might actually be able to make a problem out of this.
I hope they can at least get compensated.
So where can I check to see if my book was used? I published a book.
Did you ever comment on Reddit before 2015? If so, your copyrighted material was used to train the modern LLMs even if your published book wasn’t used at all.
Yes I did my account is almost 11years old on Reddit. But I was talking about my novel that was never on Reddit.
The database is here. You’ll have to sign up for a free trial if you’re not a subscriber to The Atlantic already. https://www.theatlantic.com/technology/archive/2023/09/books3-database-generative-ai-training-copyright-infringement/675363/
they were compensated when the company using the book, purchased the book. you can’t tell me what to do with the words written in the book once I’ve purchased it. nor do you own the ideas or things I come up with as a result of your words in your book. of course this argument only holds up if they purchased the book. if it was “stolen” then they are entitled to the $24.95 their book costs.
That’s the thing – they weren’t.
The case has two prongs.
One is that training the AI on copyrighted material is somehow infringement, which is total BS and a dangerous path for the world to go down.
The other is that copyrighted material was illegally downloaded by OpenAI, which is pretty much an open and shut case, as they didn’t buy up copies of 100k books, they basically torrented them.
And because of ridiculous IP laws bought by industry lobbyists in the dawn of the digital age, the damages are more like $250,000 per book if willful infringement, not $24.95.
Had they purchased them, these cases would very likely be headed for the dumpster heap.
That said, there’s a certain irony to Lemmy having pirate subs as one of the most popular while also generally being aggressively pro-enforcement on IP infringement.
Training AI on copyrighted material is infringement and I’ll die on that hill. It’s use of copyrighted material to create a commercial product. Doesn’t get any more clear cut than that.
I know as an artist/musician/photographer I’d rather not put my creations out there at all if it means some corporation is going to be able to steal it.
Courts look at how the party claiming fair use is using the copyrighted work, and are more likely to find that nonprofit educational and noncommercial uses are fair.
This does not mean, however, that all nonprofit education and noncommercial uses are fair and all commercial uses are not fair; instead, courts will balance the purpose and character of the use against the other factors below.
Additionally, “transformative” uses are more likely to be considered fair. Transformative uses are those that add something new, with a further purpose or different character, and do not substitute for the original use of the work.
You can stand wherever you like on any hill you’d like, but the question of nonprofit use vs commercial use is only one part of determining fair use, and where your stance is going to have serious trouble is the fact that the result of the training is extremely transformed from the training data, with an entirely different purpose and character and cannot even reproduce any of the works used in training in their entirety. And the areas where they can reproduce in part are likely not even the direct result of using the work itself in training, but additional reinforcement from other additional secondary uses and quotations of the reproducible parts of works in question.
And don’t worry. Within about a year or so (by the time any legal decision gets finalized or new legislation is passed) no one is going to care about ‘stealing’ your or anyone else’s creations, as training is almost certainly moving towards using primarily synthetic data and curated content creation to balance out edge cases.
Use of preexisting works was a stepping stone hack that acted like jumper cables starting the engine. Now that it’s running, there’s a rapidly diminishing need for the other engine.
Edit: And you’d have a very hard time convincing me that StableDiffusion using Studio Ghibli movies to train a neural network that can produce new and different images in that style is infringement while Weiden+Kennedy commercially making money off of producing this ad is not.
Good point. I guess this aspect is much different from the AI Art scene, where the producers of the dataset are usually not compensated for their drawings.
Yea I hope they get their $1 for their contribution
What about my Reddit history?
Arguably there’s more of my text there that was used to train these LLMs than most authors in that list.
The comment elsewhere in this thread about models built on broad public data needing to be public in turn is a salient one.
IP laws were designed to foster innovation, not hold it back.
I’d much rather see a world where we have open access models trained broadly and accelerating us towards greener pastures than one where book publishers get a few extra cents from less capable closed models that take longer for us to reach the heyday where LLMs can do things like review the past 20 years of cancer research in order to identify promising trends in allocation of future resources.
OpenAI should probably rightfully be dinged for downloading copyrighted media the same way any average user would be sued when caught doing the same.
But the popular arguments these days for making training infringement are ass backwards and a slippery slope to a far more dystopian future than the alternative.
I would be proud, but you do you.
But would you get paid?
Yes. People wouldn’t be able to pirate my story through an AI, it wouldn’t spit it out verbatim. They’d still need to buy or pirate it other ways.
They don’t need to, the AI just tells them what happens. Why are you against the author being able to consent for their work to be trained on and being compensated?
You think spoilers should be illegal?
?
You appear to be saying that:
the AI just tells them what happens
Is a violation of the author’s copyright.
I know, right? It’s weird.
do they also complain when their books are used to train wet networks in public schools? those networks are also later exploited by corporations who dont give back the writers. hmmmmmmm
They do get paid for that, however. They get a share of the value of each book sold. Those schools are paying for the books.
There is also the catch that those wet networks are of finite lifespan and are output throttled. This limits the losses caused. A lot of authors also consider improving those networks a big part of why they write.
It’s the difference between someone hand drawing a Micky mouse birthday card for their sibling, and hallmark mass producing them for sale. The former is considered acceptable, the latter is grounds for a law suit.
Everyone’s a fan of fair use until it’s their work that is transformed.
Ok so it’s been stealing art now it’s coming for authors. At what point do we hold the coalition who started this shit culpable for numerous accounts of plagiarism?
TIL “culpable” is an English word too. Culpable means guilty in Spanish and I thought you were a Spanish speaker doing spanglish. Now I know you’re just a man of culture.
There’s an idea by Barath Raghavan about an AI dividend that companies pay each netizen a share for the data they use to train these models.
I am into this idea if companies can’t even do a simple opt-in mechanism.
Does this fall under fair-use part of copyright?
It hasn’t been tested in court yet but I don’t see why it shouldn’t.
Fair use is any copying of copyrighted material done for a limited and “transformative” purpose, such as to comment upon, criticize, or parody a copyrighted work.
I don’t see why it should.
The creation of the AI model is transformative. The AI’s model does not contain a literal copy of the copyrighted work.
No, but the training data does contain a copy. And making a model is not criticising, commenting upon, or creating a parody of it.
That list is not exclusive, it’s just a list of examples of fair use.
The training data is not distributed with the AI model.
it’s just a list of examples of fair use.
Yes, it’s a list of quite similar ways of commenting upon a work. Please explain how training an LLM is like any of those things, and thus, how Fair use would apply.
I’m not saying that training an LLM is like any of those things. I’m saying it doesn’t have to be like those things in order for it to still be fair use.
Pay for every bit of information you’ve read and regurgitated on exams.
AI is not human and should not be treated like a human
It’s not. The humans that trained it (assumably) purchased the material used to train it. What’s the problem?
The training argument is probably going to come up dry by the time the court works its way through expert testimony, as the underlying argument for training as infringement is insane.
But where OpenAI is probably in hot water is that torrenting 100k books in the first place runs afoul of existing copyright legislation.
Everyone is debating the training in these suits, but the real meat and potatoes is going to be the initial infringement of obtaining the books, not how they were subsequently used.
Curious if the AI company actually bought those books or if they just came across them by pirating.
Oh, they’re 100% pirated. Sorry this isn’t open, but the preview should give you enough information. The database is available elsewhere, IIRC. https://www.theatlantic.com/technology/archive/2023/09/books3-database-generative-ai-training-copyright-infringement/675363/
