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.
deleted by creator
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.
deleted by creator
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
should be open source by law.
That doesn’t make sense. The “source” of the AI model is the publically available information, which the creators have no right to redistribute.
The rules of Open Source simply do not work for AI models. You’d have to come up with some other rules.
deleted by creator
The whole legal situation around AI models isn’t clear and common Open Source licenses are an ill fit for them because you aren’t distributing the source, but just a binary blob. You can’t just take any random accumulation of data and slap a Open Source license on it, especially when that accumulation is the result of proprietary data, incompatible licenses and all that.
Most people don’t care and just remix everything as they please, but just because you can download for free something doesn’t make it Open Source. Furthermore a lot of the models exclude commercial use or otherwise restrict the use in ways that are incompatible with the Open Source definition.
Has any of the model made it into Debian yet?
deleted by creator
They are, just like a lot of other people. For example they cite LLaMa 2 as a popular open source model, while it restricts commercial use.
What do you define as “source” for an AI model? Training code? Training data set?
Yeah, it ought to be owned by the people who contributed the work that trained it. But that’s socialism. … No really, that would literally be socialism.
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.
The part where brain and neural net differ is in the learning via backpropagation, that seem to be done different in the brain, as there is no mechanism to go backwards through the network and jiggle the weights.
That aside, they seem to work very similar once they are trained, as the knowledge they are able to extract from data ends up being basically the same that a human would be able to extract. There is surprisingly little weirdness in AI and a surprising amount of human-like capabilities.
people have a definite fear of being defined as machines… not sure why we think were so special…
so its barely understood, but this definitely is not it. got it.
But you, random stranger on the internet, knows better than the guy that literally works in the field. Got it.
i do? where did i claim that?
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.
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.
An AI model is not a derivative work. It does not contain the copyrighted expression, just information about the copyrighted expression.
Better tell that Google and their search index, book scanning project and knowledge graph.
I didn’t know those were LLMs, TIL.
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.
So do humans.
That’s a Bingo!
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.
Does that mean software can also be afraid, or angry?
If it is programmed/trained that way, sure. I recommend having a listen to Geoffrey Hinton on the topic (41:50).
Saying software can be inspired is like saying a rock can feel pain.
The rock doesn’t do anything similar to pain. The LLM on the other side does a lot of things similar to inspiration. I can give the LLM a very trivial question and it will answer with a mountain of text. Did my question or the books it was trained on “inspire” the LLM to write that? Maybe, depends of course how far reaching you want to define the word. But either way, the LLM produced something by itself, that was neither a copy of my prompt nor the training data.
Here is an alternative Piped link(s):
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I’m open-source; check me out at GitHub.
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.
The anthropomorphism of these models is a major problem.
People attributing any kind of person hood or sentience is certainly a problem, the models are fundamentally not capable of that (no loops, no hidden thought). At least for now. However what you are doing isn’t really much better, just utterly wrong in the opposite direction.
Those models are very definitely do “learn” and “understand” by every definition of the word. Simply playing around with that will quickly show that and it’s baffling that anybody would try to claim otherwise. Yes, there are limits to what they can understand and there are plenty things that they can’t do, but the amount of questions they can answer goes far beyond what is directly in the training data. Heck, even the fact that they hallucinate is proof that they understand, since it would be impossible to make completely plausible, but incorrect, stuff up without having a deep understanding of the topics. Also humans make mistakes too and they’ll also make stuff up, so this isn’t even anything AI specific.
Yeah, that’s just flat out wrong
Hallucinations happen when there’s gaps in the training data and it’s just statistically picking what’s most likely to be next. It becomes incomprehensible when the model breaks down and doesn’t know where to go. However, the model doesn’t see a difference between hallucinating nonsense and a coherent sentence. They’re exactly the same to the model.
The model does not learn or understand anything. It statistically knows what the next word is. It doesn’t need to have seen something before to know that. It doesn’t understand what it’s outputting, it’s just outputting a long string that is gibberish to it.
I have formal training in AI and 90%+ of what I see people claiming AI can do is a complete misunderstanding of the tech.
I have formal training in AI
Than why do you keep talking such bullshit? You sound like you never even tried ChatGPT.
It statistically knows what the next word is.
Yes, that’s understanding. What do you think your brain does differently? Please define whatever weird definition you have of “understand”.
You are aware of Emergent World Representations? Or have a listen to what Ilya Sutskever has to say on the topic, one of the people behind GPT-4 and AlexNet.
It doesn’t understand what it’s outputting, it’s just outputting a long string that is gibberish to it.
Which is obviously nonsense, as I can ask it questions about its output. It can find mistakes in its own output and all that. It obviously understands what it is doing.
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.
The data is gone by the time a user interacts with the AI. ChatGPT has no access to any books.
And here I thought software just did exactly as it’s instructed to do.
AI isn’t software. Everything the AI knows is from the books. There is no human instructing the AI what to do. All the human does is build the scaffolding to let the AI learn, everything else is in the data.
deleted by creator
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.
This is why it’s so stupid to call this shit AI
It is using the term as defined. Maybe stop being a stupid parrot just repeating crap you heard else where and use your brain for a moment. I am losing hope that humans are capable of thought reading all this junk.
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.
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.
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.
A human, regardless of how many books they read, will have personal experiences that are undeniably unique to themselves.
So will every AI. ChatGPT will give you different answers than Bard or WizardLM, since they are all trained on different books. And every StableDiffusion model creates different images, different styles, different topics, etc. It’s all in the data they “experienced”.
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.
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.
Biological machines, yes.
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.
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.
It shares popular quotes from books, it can’t reproduce arbitrary content from a book. The content needs to be heavily duplicated in the training data to stick around (e.g. from book reviews), and even than half of it might still end up being made up on the spot.
Also request for copyrighted content will be blocked by ChatGPT and just receive the stock “I can’d do that” response anyway.
If you have some damning examples that show the opposite, show them.
Being blocked by ChatGPT just means that the interaction layer you see doesn’t show the output, not that the output wasn’t generated.
Everything you see that’s public facing and interfacing with an AI is an extreme filtering layer for what is output. There’s tons of checks that happen to ensure that they don’t output illegal content or any of a million other undesirable things.
I’m too lazy and care too little but you can basically get it to roleplay as a book expert or something and to “remind” you of certain passages. It gets around the filter pretty easily, that’s how jailbreaks work.
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?
deleted by creator
LLMs can’t reprint their entire training data on demand. They rarely even remember quotes.
I mean, yeah? They were running to a concrete description. That is not valid. My brain has most of Terry Pratchett’s works.
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.
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.
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
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?
I know, right? It’s weird.
?
You appear to be saying that:
the AI just tells them what happens
Is a violation of the author’s copyright.
Everyone’s a fan of fair use until it’s their work that is transformed.
deleted by creator
Absolutely agree
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.
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.
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/
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?
Authors Guild, Inc. v. Google, Inc. decided that it is fair use to scan books and make large parts of them available verbatim on the net. What AI does is far more transformative than that, as very little of a book can be reproduced verbatim with AI (e.g. popular quotes), you really just get “knowledge” from the books. The sources are however lost in the process, unlike with Google, which by itself however also makes it difficult to argue for copyright violation, since you can’t point at what was actually copied.
