In the case of The Pile, "processing for language modelling" means "converting epub and pdf into plain text, maybe deduplicating, maybe removing some sorts of detectably malformed files"
So not a particularly lossy conversion.
Then "they're doing something amazing, they don't need permission, and the cat is already out of the bag, and similar musings".
Seriously, it's both copyright infringement, and unethical. This is why I don't use any of the popular AI tools, or even AI add-ons in Evernote, Notion, etc. They all link back to the usual suspects.
Ah, the Uber theory of law. Works surprisingly well for some reason.
I can see the problem where direct and faithful replication is possible but where it isn't is there still a problem? Or is the automatable aspect, the scale at which it can occur, that is the problem?
An AI system consumes something perfectly, then ingrains it into its weights perfectly, and becomes capable of imitating the same thing perfectly. Plus, ther are no other internal or external factors which affect these "generation" over time. Hence, it mixes and reproduces based on what it consumed, solely.
I might get inspired by people, and add my own values to it, iterate over it and diverge from what I'm inspired from ultimately to create my own style. AI doesn't work like that. Also, if I do the same amount of inspiration with the same precision and accuracy, I'll be neck deep in accusations and lawsuits (for the right reasons).
As a result, just because we fail to ask the right questions to reproduce the training data verbatim or almost verbatim doesn't mean that the information is not there. At the end, a neural network is a compression algorithm which encodes data in terms of weights. Given the correct input, you can regenerate the training data as is.
Unless you have special abilities, you can't read 500 books an hour, remember them perfectly, and generate derivative works by mashing all of them together. If I do and try to sell a novel, I'll be ridiculed no end. If I write a Ph.D. the same way and try to defend it, I'll be banned from academia for three lifetimes at least.
For more elaboration on the subject, see [0].
If you divide Stable Diffusion's file size by the number of images used to train it, you get something like 1.2 bits per image, and it is physically impossible to get this kind of a compression ratio.
The actual problem with AI is that it sometimes plagiarizes random fragments of the work it is trained on, even if it is not the user's intend, and we currently don't really know how to fully prevent this.
Same for code generating models trained on Open Source and Free Software. Tons of licenses violated, from strong copyleft to source available models and reproduced (almost) verbatim with comments intact.
Some researcher's codebase is almost completely reproducible without any licensing information just by hinting the function names.
Maybe for the image compression it's borderline impossible for now due to network size, but for text and code, generation of training data almost verbatim is very possible and straightforward.
Also in image generation models, style transfer is the bigger problem, because it completely eliminates the artist who uses/created the style in the first place. "You pioneered this, and we fine tuned this model with your images, and we can do your work for free, without you, have a nice day". However, the artist's life expenses doesn't disappear when they're transferred to an image generation model.
This is also unethical.
Just because you can doesn’t mean you should. That's what I'm trying to say.
IIRC it was like 1.4 bytes before adding in random initial and prompt. And Amiga Four-byte Burger is 4 bytes long.
[0]: https://bytecellar.com/2023/04/24/lost-amiga-four-byte-burge...
Open science repositories would take down the "dataset" immediately (or at least limit its access) if a copyright holder brings the matter to the eyes of the admins.
In terms of fair use, one of the larger factors is the 'market substitution' factor, which basically means "does this use compete with otherwise licensed uses that people would ordinarily pay for?" AI absolutely does compete with human artists for the same market. In fact, it's winning handily[1], because you don't have to pay human artists. AI art models absolutely shouldn't be trained on anything with copyright on it.
The other factors don't fare much better. Nature of the original work will differ based on the plaintiff, but the purpose and character of the AI's use of that work is very much commercial. And the amount and substantiality of the use is complete and total. I don't see AI being fair use - at least, not in every one of the many, many training lawsuits currently ongoing against OpenAI and Stability.
[0] Starting with any body of text, an LLM, and an empty context window, compute the next-token probabilities and take the highest one. If it matches the source text, output a 1 bit. If it doesn't, output 0 followed by the ID of the correct next token. Add the correct token to the context window and repeat until the text has been fully compressed. This produces a list of perplexities (wrong words) for the given text which can be used to guide the LLM to output the original work.
[1] Hey, remember when both WotC (biggest art commissioner on the planet) and Wacom (hardware vendor that sells art tools and payment terminals[2]) both got caught using AI art after making very loud and public pledges to not do that? They both wound up buying stock photography on marketplaces that are absolutely flooded with AI trash.
[2] All the credit card readers in Japan are built by Wacom, which is really funny as an artist
And that's legitimate inventions I'm talking about. Just wait until the patent trolls figure out how to get the A.I.'s to divulge patent violations to sue the A.I. suppliers.
So in that sense I understand the response that "they don't violate copyright" by studying the material. Again, I don't pretend to be a lawyer, and not every law has to follow my logic.
This isn't about the output for content generators or about the abstract numeric weights that they operate over. That's more complex and a largely open question.
But this is literally about indiscriminately distributing copyrighted works in a large, convenient archive while arguing that it's okay because you normalized the formatting a bit and because you suspect that some people might find "fair use" value in it.
Meanwhile the complainant (such as GRRM) forgets that often passages from said articles and books are strewn throughout the Internet. Of course chat GPT can drop passages from the GoT books; there's several entire fucking wikis for that franchise that reference passages, quotes, details etc.
Same goes for news articles, passages of which are often quoted by other sources or websites.
Not that chatGPT has reproduced many works in whole, but it's an interesting logic problem for fair use law: if I have copyrighted article X, but websites ABCDEF all quote various passages of my article (ie fair use, critique, etc) and then ABCDEF is used to train an LLM, if the LLM can _reassemble_ the article from quoted passages without referencing article X itself, is it copyright infringement or fair use?
From the ruling:
> Assuming the truth of Plaintiffs’ allegations - that Defendants used Plaintiffs’ copyrighted works to train their language models for commercial profit - the Court concludes that Defendants’ conduct may constitute an unfair practice.6 Therefore, this portion of the UCL claim may proceed.
https://caselaw.findlaw.com/court/us-dis-crt-n-d-cal/1158180...
Best to operate under the presumption that The Pile is legal because we wouldn't be talking about any of this LLM stuff at all if it weren't for copyrighted text.
You can argue that the law should change, but SV has this absurd idea that it somehow can't be illegal if it is a part of a successful business model.
Yes, that's how the law works. If something is in a grey area you can do it until the government tells you not to.
You can only profitably arbitrage gaps in the law for so long before someone goes too far and there's a scandal or two, and public opinion turns against you, followed by the business-end of legislative attention.
As for the government, they are theoretically supposed to be responsive to the people, and the people have spoken loudly and unequivocally in favor of the "exploitative assholes."
It really isn't, if you don't fall for the false dilemma: taxi cartels and Uber both exploit their drivers and passengers in different ways.
> ...the people have spoken loudly and unequivocally in favor of the "exploitative assholes."
Public support is turning against tech slowly, and then rapidly. We will deserve it, going by the comments and attitudes I frequently see on HN. While I'm grateful for the high tech salaries, the prospect of wealth attracted a lot of people whose aims start and end with getting rich by any means, and will ruin it for the rest of us.
So for years, they operated illegally, gaining market share subsidized by VC funds and wage theft of driver's benefits, then changed the law to legalize what they had been doing from the start.
Q: Why do you think X is Y? A: < 10 minute exposition that doesn't address the question > ... therefore Y.
Sorry for off-topic, but I love that expression.
> I don’t understand how this can be true if set 2 contains a complete copyrighted work (say, a book) that the copyright owner hasn’t approved for such distribution.
Everything you're saying could be true, but also meaningless.
Meaningless in what sense? In my personal opinion, here are some meaningful things:
- Economic meaning: Is the Pile making anyone suffer, in an intellectually honest sense? No. Is the way it is being used making anyone suffer? No. Would anyone use the Pile if they had to pay for it? No. Would some tiny amount of money paid for it matter? No, not to anyone. The Pile is but one of many things that are going on in this pareto-inefficient world that we gain little economically by fucking with.
- Meaning in the sense of human progress. Is the Pile sincerely, substantively helping advance creative and scientific progress? Yes. Can the progress it promotes coexist peacefully, in this current status quo, with selling books on Amazon, which can also advance progress? Yes. Is there any evidence that this non-economic progress, such as sharing important stories or points of view meaningfully through text content, is harmed by the Pile? No.
The following ways of looking at this issue are meaningless:
- Psychological meaning is meaningless. Feelings that are more strongly felt by more favored parties, like beloved authors versus impetuous programmers, are not more valid. Stakeholders like authors merely exercising their collective bargaining power over something that doesn't actually make them worse off is meaningless too. Social media drama also doesn't matter.
These are just opinions. I don't go out there and say I'm a lawyer, and I'm not running for Congress, and neither are you. It's crazy to me, because when you look critically at what dog you personally have in this race, like of course you want the status quo where The Pile exists, and you and the only agitators in all of this - authors' guild authors - gain nothing from nearly every dead-on-arrival framework like "Compensation, Credit and Consent" or whatever, BESIDES meaningless psychological satisfaction of winning social media arguments and flexing the favorability of authors over programmers. Imagine if we ran the whole country this way! It's radioactive.
So this isn't a non-sequiter. Supporting the status quo, in this particular case, is the succinct way of saying everything here: that you can be right in ways that don't at all matter, which is okay, but you should have the insight or maybe if you are a programmer or an author or actually you are running for Congress, you have the duty to understand this stuff.
The first agitators were Free Software developers, namely the Software Freedom Conservatory and Hector Martin, who pointed out that GitHub's proprietary cloud hosted code completion LLM was trained on shittons of GPL code and regurgitated a lot of it[0]. The only reason why book authors, visual artists, and so on are suing AI companies is because the Free Software people - who otherwise oppose software copyrightability - wrote the rhetorical framework that everyone else is using.
[0] My favorite example is that commenting "evil floating point level bit hack" produces the entire Quake lighting function. Or at least it did until Microsoft specifically filtered that phrase out from code completion.
Do you really expect anyone to engage with you in good faith if you feel this way?
Copyright law isn't perfect, but most of the great creative works from the last few centuries wouldn't exist if their authors didn't have legal protections allowing them to monetize their work. Redistributing existing media without worrying about copyright might be great in the short term, but the next century of art will probably look pretty barren if that becomes the status quo.
Maybe we'll reach a point where the benefits of training AI outweigh the benefits of all human creative output. But you don't need to be a stooge of some author's guild to not feel like we're there today.
No, and a class action lawsuit against that practice doesn't make anyone better beyond "meaningless psychological satisfaction". But it's still something we do as a society, for good reasons, right?
Here, you have a number of commercial entities that are poised to make a ton of money off individually small but entirely non-consensual contributions of other people. You also have a "data laundering" industry where the datasets of webpages, books, and images are published by notionally non-profit entities that are in one way or another bankrolled by the commercial players, keeping their hands clean.
Framing the debate as being about "human progress" is sort of goofy in that world. We're not talking about academics tinkering with wacky ideas. We're talking about an industry that takes your work without asking and monetizes it. Yes, the sum of it is greater than the component pieces, but that doesn't mean you get to do whatever you want.
No. Nobody cares.
> But it's still something we do as a society, for good reasons, right?
I really don't think whatever you're talking about matters. I don't think anyone cares about class action lawsuits over pennies except assholes. But I enjoy that you are reading these comments.
I also think you are ascribing way too much meaning to the legal process. Process and procedure matter to lawyers. There are places where you can never get relief from the law, and there is no process and procedure, but oftentimes, people will concede that they are governed by rules and laws.
In the US the starkest example was probably the BP oil spill, where they had video of oil coming out of the thing, there was nothing to investigate, and in 15m a meeting between political and business executives concluded that the fine would be large but not ruin BP. The next 2 years of process for that disaster were meaningless under the same framework of meaning I gave you. It provided psychological satisfaction to all sorts of people, maybe especially to the people who had to feel good about taking a big check from BP for destroying the ecosystem except by fishing instead of by extracting oil. But it didn't matter, they could have also not done 2 years of process and nothing would change.
> Here, you have a number of commercial entities that are poised to make a ton of money off individually small but entirely non-consensual contributions of other people.
The Pile is doing the exact opposite, and unauthorized but nonetheless legal use of copyrighted material for LLM training is the only way you will see a world where non-commercial entities, such as authors, can use the technology for free, in whatever way they choose it to be aligned, under whatever rules.
The DMCA already made movie piracy cost $100,000 a pop more than 2 decades ago, and yet here we are. Who. Cares. Did any of it work? You can find lots of bad guys in movie piracy, and extremely few good guys, and it turns out none of it matters. Surely if we can tolerate them we can tolerate the Pile.
> We're not talking about academics tinkering with wacky ideas.
You might mean that they specifically are not graduate students or professors at Universities, sure, but they are all Ivory tower, bone dry Ivy League or adjacent academics. Some of the people at the very top may have dropped out of Stanford or Harvard or whatever, and some of them may have some nuanced or fringe beliefs about specific journeys through learning for specific people, but they are very much the kind of person who would be called "academic," as an insult, by a layperson.
This is just to say that you are immediately reaching for the good versus bad people appeals. Like it's the same shit, it's the same energy as saying the authors are right and the weirdos are wrong. You have some favorable idea in your head of nice humanities PhDs, and yes, they are nice, and this disfavorable view of mean compsci bachelors, but does it matter? In a real sense they are cut from a similar cloth, but in the imaginary setup in your head they're different.
> We're talking about an industry that takes your work without asking and monetizes it.
This is the same appeal. Authors Guild is the industry. Writers Guild is the industry. You are mixing up your antagonists and protagonists. You need to get an independent opinion, colorfully we could call it a "contrarian" one, about who all these people really are. And then you'll see, man, it's not so black and white, and I wonder why people are arguing about this shit that does not matter. If you are a guilded writer, I will tell you now, you are wasting your time worrying about this stuff, because monetizing via traditional publishing or screenplays has been bad forever, and the Pile changes nothing.