>We are and always have been DMCA compliant, this is our official policy. https://the-eye.eu/dmca.html However we're often hounded by false claimants to which this(mp4 the Twitter post links to) is our unofficial policy, popularised as a meme in our community that time a church tried to sue us for $22,000,000.
>Note well...Might it be that a frank conversation between adults is what is needed instead of puerile posturing?
If you're criticizing them for making a dumb joke years before there was any problem, when can they make a dumb joke? As the rest of the thread seems like a frank conversation between adults, as does their interview in the article.
I'm sorry, my previous comment was likely confusing. I was asking where was it that I said that "at no point can they ever make a joke" as you said I did. Can you please point me to the text of my comment where I say that? Thanks!
I have explained why I took your comment to mean "basically they can't ever make a joke," if you disagree then refute my explanation. It's getting difficult to assume good faith, so I'm done
This seems like the inevitable outcome. Hosting pirated media for any purpose is going to draw moneyed attention, and doing so for a purpose as controversial as AI training doesn't help. Multiple authors I follow have expressed their disapproval of their works being included in that dataset.
There are enormous digital libraries being used for training where authors will have no recourse in preventing this from happening. To continue, those performing model generation just have to either avoid certain jurisdictions or not announce what they're doing. I am not unsympathetic to author feelings on the topic, but regulation in this regard moving faster than bits and compute will be challenging to say the least. For these small dataset amounts currently discussed (<1PB), the cost to geographically relocate is obviously trivial.
TLDR Model training will go "dark" or underground, with models distributed like illicit content.
I agree, realistically this is just going to make training sets less findable in mainstream knowledge, but I think writers being against it regardless makes sense on principle. I, too, would be pissed af if someone took say, my fingerprints, and started putting my fingerprints on random stuff and places where I wasn't.
Would you even need to distribute the models that way? How would anyone know what you trained them on? Being able to occasionally emit a direct quote from a literary work doesn't tell you whether it was trained on the work itself or on various internet posts that have quoted from it, nor how the work was obtained. Is it illegal to train a model on a copy of a book that you own?
That's assuming you kept records of what you used, and that the person who trained the model is the same person distributing it, or that the former is even in your jurisdiction.
It not obvious that this even should be illegal, but a decent factor in determining that is whether making it illegal would even do any good. Unenforceable laws are all cost and no benefit.
The models themselves record the training data. ChatGPT is happy to spit out Albus Dumbledore when asked about a hypothetical magic school. If you don’t have the exact training dataset that’s not going to be an actual defense.
The only protection is specific jurisdictions ruling this stuff is fair use, which is risky when you could be found liable in literally any country.
> ChatGPT is happy to spit out Albus Dumbledore when asked about a hypothetical magic school.
How are you supposed to know if that's because it ingested the text of Harry Potter or a bunch of fan blogs talking about Harry Potter?
> The only protection is specific jurisdictions ruling this stuff is fair use, which is risky when you could be found liable in literally any country.
Saudi Arabia has quite strict blasphemy laws but people don't seem to be bothered much by it when hosting quite obvious violations of them on their servers in North America.
> or a bunch of fan blogs talking about Harry Potter?
Doesn’t actually matter here because those fan blogs are also derivative works. I may have read “Call me Ishmael” in a blog, but the quote is from Moby Dick.
ChatGPT can argue for a fair use exception, but combining lots of derivative works can be copyright infringement even if none of the things you directly copied where. IE: If you copy a lot of excerpts from a poem and recreate the poem that doesn’t mean you can now use the poem.
Being able to say Harry Potter isn’t directly in their training data is at best useful to arguing something was unintentional copyright infringement rather than not actually being copyright infringement.
> Blasphemy laws
That’s a criminal not civil issue and there aren’t a huge number of treaties on the subject. If a US company is sued in Australia for copyright infringement they don’t get to move the case to the US.
Worth noting that Fair Use is more than just transformative work. Your work can still be found to be infringing even if it's transformative if it egregiously violates other tenets of fair use, like disrupting the market for the original.
> Doesn’t actually matter here because those fan blogs are also derivative works.
Not necessarily. To be a derivative work it has to contain copyrightable elements, not just a small number of words in the same order.
> ChatGPT can argue for a fair use exception, but combining lots of derivative works can be copyright infringement even if none of the things you directly copied where. IE: If you copy a lot of excerpts from a poem and recreate the poem that doesn’t mean you can now use the poem.
This is the argument for why the model itself isn't infringing, isn't it? All of the excerpts as presented in their original context isn't the same thing as all of the excerpts purposefully rearranged back into the original work.
It's like on the one hand arguing that a dictionary is infringing because it contains all of the words in Harry Potter, and on the other hand claiming that you can reproduce the text of Harry Potter without liability because you reconstructed it by rearranging the words in a dictionary. One of these things is not like the other.
> If a US company is sued in Australia for copyright infringement they don’t get to move the case to the US.
If a US company has no operations in Australia, how does an Australian court have jurisdiction?
Fictional character names are copyrightable elements. Sure, it’s no issue for actual place names, but those aren’t the words people will be looking for.
> a dictionary is infringing because it contains all the words in Harry Potter
The argument is the model may have recreated the story of Harry Potter from being fed so much information about it.
DALL·E can’t contain a copy of every image in it’s corpus simply based on information theory, the model just isn’t that large. However it has no problem recreating the Mona Lisa because so many of its source images included it. That’s no problem for something out of copyright but it’s a big problem for major works in our culture like Harry Potter or Star Wars where potentially hundreds of thousands of derivative works get stitched together into a copy of the story.
If ask ChatGPT to summarize Harry’s first year at Hogwarts and it spits back a detailed summary of his year including dialog etc that’s wildly past the de minimis standard. I assume OpenAI has tried to prevent this, but this kind of reconstruction seems like a legal landline.
> Fictional character names are copyrightable elements.
Copyright, or trademark?
> The argument is the model may have recreated the story of Harry Potter from being fed so much information about it.
It may contain enough information to recreate it, but is that the same thing?
Suppose you're a major blog platform and everybody's blog contains the data the model was trained on, i.e. enough to recreate the story of Harry Potter if you were to read them all and then be asked to summarize it or write fan fiction. If your company isn't infringing the book's copyright, why would the model be? It certainly contains no more information than it was trained on.
The output might be infringing, in the same way as a person who read all the blogs to learn the story could write something that was. But the output isn't the model.
Always copyright, sometimes both. Really though fictional works have a really low bar for copyright infringement.
> It may contain enough information to recreate it, but is that the same thing?
Yes. The law currently doesn’t give a shit about how you got the information just it’s ultimate origin. It’s the same reason MP3’s absolutely qualify as as copies even though none of the bits map 1:1 with the source material.
An LLM that reproduces a lossy copy of some work is just another lossy format. A judge isn’t going to care about AI magic, they are just going to see the infringement. The only defense is to not create copies of copyrighted works which shouldn’t be an issue if this stuff is actually transformative.
That kind of information is present in fan fiction, Wikipedia, reviews. Plenty of other sources. You're also incredibly wrong on being able to assume a book is present in the dataset. It is up to the complainant to prove the case, you are not guilty because you didn't keep sufficient records.
Copyright is like a shit sandwich, just those two words alone prove it’s a derivative work. This is true even if you copied from a copy. Once something is demonstrated as a derivative work, proving your use is fair use is now on the person who created it. That’s normally fairly easy with a book, but harder with LLM’s if they easily spit out large chunks of a clearly copied work.
Remember lawsuits aren’t beyond a reasonable doubt the standard is much lower.
Surely this opens a door where, because an agent is unlikely to reproduce work verbatim but rather in a more compressed or decompressed wording, humans are held to a different standard? Otherwise humans would remain an effective way to launder the output of the agent.
Copyright still applies if you’re copying elements of the story, it’s just harder to detect and less obvious in court.
The value of names to copyright holders is simply how distinct they are. John Smith isn’t particularly unusual but if you name a bounty hunter Boba Fett it’s hard to argue originality. Complete originality isn’t required as long as you’re safely inside a fair use exception, but overwhelming originality is required.
So it’s generally safer for a character to talk about their favorite parts of Star Wars than have your stories plot be a group of rebels trying to destroy a planet killing doomsday weapon guarded by a sword wielding space wizards.
> Copyright still applies if you’re copying elements of the story
My point is that the meaning of "copying" as described here is unclear unclear if it's not restricted to verbatim plagiarism. As is "elements", for that matter.
Using those two words though is still fair use. And even activities such as producing fan fiction in the world of something like Harry Potter is also protected fair use. To infringe on copyright the complainant still has to PROVE that there was a copyright violation.
Without details of what specific data a language model was trained on, there is no way to differentiate it having retrieved that name from a review of the book that was used with permission from the author (which is their own copyright and fair use on its own) or from the books itself (which would be copyright infringement).
If those records don't exist, the only mechanism you have left is to try and get the language model to spit out sufficient verbatim text from the source material to cross the fair use threshold. This doesn't work in the case of "obvious extrapolation" either which is a whole other defense that could be used depending on what that body produced was (If knowing only that the main character is named Harry, who is a wizard going to a wizarding school in the British countryside is required to produce a couple of close looking paragraphs, you need a much closer match with the original text over those two paragraphs for it to be infringing).
How is this different from running pirated software for example? Big business doesn‘t do it. And if they do they get fined heavily. Same should be true for illegal AI training data. Surely this is a problem that can be solved.
But it also feels like standing at the edge of the sea complaining about the tide coming in...I'm not sure there's really much that can / will be done about it.
I actually think this kind of use is purer to the original intent of the web, where everything on the internet was freely available and consumption was encouraged. "Information wants to be free" used to be the rallying cry.
That being said, it feels like there's also a shade of perspective from the old quote:
"In its majestic equality, the law forbids rich and poor alike to sleep under bridges, beg in the streets and steal loaves of bread." - assuming everything in public is fair game, then everyone is welcome to build a multi-petabyte database of text and use millions of dollars worth of GPUs to train an AI on it.
That last point is great. I've definitely seen a lot of people talking about how we need to let the little guy develop their own AI, with very little attention paid to the actual realistic costs of doing so. GPT-4 I believe cost $100 million ish to train.
Your argument seems to be: why stop the little guys from doing something illegal if the big guys are doing it too? We should ideally stop them all, but it isn't surprising that the blatant examples of illegality are stopped first.
If datasets are not shared externally, like how deepmind etc. does it, it is much more of a grey area. Training may well end up being fair use. So yes, the efforts of Shawn compiling books3 just levels the playing field a bit.
It is a very good question, what should be protected by copyright and what shouldn’t be.
I understand that in Japan, for example, it is legal to apply deep analysis techniques, including deep learning and generative models to any data, regardless the copyright.
I would guess that in Japan, they’ve looked at China, where copyrights are also not an issue. And decided that it’s a really bad idea to stop research because of copyrights and let China to be first at the AGI race.
If a lot of time will get wasted on the GDPR and copyrights discussions in Europe and United States, instead of actually doing research, the democratic world will get behind.
Giant corporations that are scrambling to keep R&D going are quite law-abiding actually. And are currently struggling, as most of the datasets out there can’t be used, due to unclear copyright and licensing regulations.
Looking to China for ethical business practice seems pretty bad to me. I don't think some imaginary race to agi is a good excuse for violating the rights of thousands of people. If you're getting value out of someone's data and they're not being compensated, some kind of shitty externality is happening. Hurting authors and artists to the point that they make fewer works that can be fed to the AI also hurts AI companies in the long run, and also it sucks a lot.
The question here could be who is first to AGI. If it is an authoritarian regime with leadership standing side by side with Putin, things would likely suck a lot more, even if we consider training neural networks from books (learning from books) to be a copyright violation.
I agree that this database looked blatantly illegal as the ones using it never actually bought the books. You can clearly sue them for counterfeiting.
I still would like to see if legally a judge would consider the training illegal. I do not think this got tested and i would not consider the answer as clear cut. If a company buys a million ebooks, can it simply train an AI on it? If no employee reads the books and only the AI "reads" it you could argue that the company bought it only for a single reader/entity. They will probably need to update the copyright system to take this case into account. We will see in a few years how this plays out.
In the end these authors will eventually lose their copyright after their death. And nothing will stop the AI training at that point. This only delays the inevitable by a few years. The public domain already has a lot of amazing books and resources. Google even has access to very rare books nobody else has access to digitally i think:
Using the term 'anti-piracy' makes the speaker a shill for 'pro-censorhip' or 'anti-knowledge' or 'elitist' or whatever groups. Look at the effectiveness of well-crafted public communications: The people you don't like are 'pirates' - criminals; threats to order; dangerous to civilians; wild, drunken and low status. And 'our' side is against piracy!
We need neutral terminology for both sides. Even 'intellectual property' tries to transform a time-limited monopolistic license from government for intangible goods into eternal real estate. Any ideas?
This is a really good question. Can language be de-weaponised?
Wittgenstein and Marcuse thought not - that words stand in as tools
of intent. It is our hostile times that make weapons of words.
But it's frustrating because neutral prose is hard to reach. We seem
in an age where it's hard to say anything succinctly, in a fashion
that will pique the wretched attention span of the ordinary reader,
that isn't either drenched in political bias or ungainly.
I think the media industry absolutely embarrassed themselves by
misappropriating the word "piracy". Even after 30 years it still jars
as awful, childish cringe. When I see a grown man in a suit talking of
"piracy" I can't help but see a pitiful clown before me.
But one cannot police other's language. They must be allowed to make
themselves known through the words they choose.
Words can be reclaimed by social movements and used neutrally or positively instead of negatively within them. "Gay" used to be a slur, as did "slut", and language is culturally dependent on top of that - I'm lead to believe that "cunt" is a generic reference to another person in Australia, while it's not a particular term of affectation in the US. So it depends on the circles you run in. "Pirate" is villianized if you're in the shipping business due to actual piracy, but if your morals are a bit flexible, paying $100/month to have all the streaming services is definitively a choice; the word "piracy" just brings to mind that person's preferred torrent site, and is seen on the same level as breaking the speed limit on the highway.
The particular tone those two words aren't a direct equivalent is that boffin bears a respectful tone. Nerd or geek (and lately, in polarized online US political discussions, scientist and expert) don't always have the same sort of caché or respect in all circles.
Oh good point. Caution must be used when trying to use words outside the context within which where they work. "Hacker" is fine to use in these parts, but having that be your profession on an immigration form is not likely to go over well!
I mean, except if they are clearly copying works they don't have copyright to, they are violating the law and effectively taking some amount of money out of the pocket of authors and rights holders. You can try and demand people use more "neutral" language to disucss copyright violations, but good luck.
I hope you never know the misfortune of having assets devalued by people that feel they don't need to operate within the boundaries of the rules.
That assumes the laws are fairly made and just, and therefore people who don't conform to them are doing something harmful.
If a company successfully lobbied and bribed the legislature to pass a law gving the company a monopoly on book publishing and sales, and someone bought books elsewhere, would that someone be doing something wrong? Are they harming that poor victim of a company?
Life isn't as simple as a hypothetical, of course, but it's also not as simple as assuming everything is just and fair.
You're giving a lot of credit to the "pro-piracy side", which in this case are massive corporate surveillance platforms generating huge amounts of revenue with data they had no part in creating. There's gotta be a middle ground here that doesn't turn all artists and writers into paste for the AI machine.
No, there's no ideas. It's already trivial to devour the works of many good writers for the cost of an internet connection and the capital outlay of 'a laptop'.
Piracy is killing the motive to work. I've seen it happen countless times and nobody says a thing about it.
How about copyright becomes so pervasive and all enforced that massive corporations like Disney cannot bully the small fry into legally moving into their turf.
Copyright is an hellish nightmare because it used by big corporations to devour, rather than used to emancipate every single person.
ChatGPT ravaging and pillaging the internet was the last straw for "faith in freedom".
I'm putting all my faith and work into walking on water, establishing the cleansing task of getting out from under the sawblade.
People think to write is to compile and package in on itself like a computer program or a spirit-sucking business seminar. Typing on keyboards does that to a man.
It is known that LLMs are possible, so there is no stoping it. It's too late to think about if we like it. All ideas and skills are fair game now, but LLMs lack in depth in all of them. No LLM system is better than a domain expert.
Maybe with better training data we will have better models. There are some models trained only on public domain and synthetic (generated) data. So probably LLMs won't learn directly from our works. They need better data than internet stuff, human text is not good enough. See the Microsoft model Phi-1.5 [1], trained on diverse synthetic text of high quality.
If copyright cannot be technically held any more, then it should rather be not held for everyone? Not just the small fry who have no recourse, but also for giants like Disney.
Since sharing of information is not a zero-sum game, areas where limiting access to an intellectual product is hard to achieve tend to build open and cooperative products instead. Of course it's much slower, and the output is smaller.
I’m at the point where “piracy” isn’t even a Bad Word anymore, because it always refers to pirating from big companies where there’s either no way to legitimately buy, buying is unreasonably expensive (e.g. due to different seasons being on different platforms), or it’s research where the scientists don’t care about their work being pirated because they don’t get paid for it anyways.
You’d have to reword it to something like “anti-digital theft” to bring back the negative connotation. Or have more articles about how piracy is destroying small businesses, which are suspiciously rare.
> Using the term 'anti-piracy' makes the speaker a shill
No, it absolutely does not. "Piracy" is a reasonably neutral word with a specific definition (it's slightly different than "theft") - you are artificially assigning nonexistent emotional weight to it in order to push your own political agenda.
"Is this the right thing to have done or not?" is an important question obviously but the more impactful question for our times is: "is the right thing being done consistently?"
So, why is similar action not being taken against the actual big offenders -- Google, OpenAI, rest of big-tech?
So they "burned" down a collection of books larger than the old Library of Alexandria of yore, specifically to deny scientists the ability to progress the art of AI.
Are we sure that copyright is really still suitable for purpose when it comes to promoting the Progress of Science and useful Arts?
96 comments
[ 3.4 ms ] story [ 282 ms ] threadhttps://twitter.com/theshawwn/status/1320282155457646592
Might it be that a frank conversation between adults is what is needed instead of puerile posturing?
>We are and always have been DMCA compliant, this is our official policy. https://the-eye.eu/dmca.html However we're often hounded by false claimants to which this(mp4 the Twitter post links to) is our unofficial policy, popularised as a meme in our community that time a church tried to sue us for $22,000,000.
If you're criticizing them for making a dumb joke years before there was any problem, when can they make a dumb joke? As the rest of the thread seems like a frank conversation between adults, as does their interview in the article.
TLDR Model training will go "dark" or underground, with models distributed like illicit content.
https://news.ycombinator.com/item?id=37386478
The only protection is specific jurisdictions ruling this stuff is fair use, which is risky when you could be found liable in literally any country.
How are you supposed to know if that's because it ingested the text of Harry Potter or a bunch of fan blogs talking about Harry Potter?
> The only protection is specific jurisdictions ruling this stuff is fair use, which is risky when you could be found liable in literally any country.
Saudi Arabia has quite strict blasphemy laws but people don't seem to be bothered much by it when hosting quite obvious violations of them on their servers in North America.
Doesn’t actually matter here because those fan blogs are also derivative works. I may have read “Call me Ishmael” in a blog, but the quote is from Moby Dick.
ChatGPT can argue for a fair use exception, but combining lots of derivative works can be copyright infringement even if none of the things you directly copied where. IE: If you copy a lot of excerpts from a poem and recreate the poem that doesn’t mean you can now use the poem.
Being able to say Harry Potter isn’t directly in their training data is at best useful to arguing something was unintentional copyright infringement rather than not actually being copyright infringement.
> Blasphemy laws
That’s a criminal not civil issue and there aren’t a huge number of treaties on the subject. If a US company is sued in Australia for copyright infringement they don’t get to move the case to the US.
Not necessarily. To be a derivative work it has to contain copyrightable elements, not just a small number of words in the same order.
> ChatGPT can argue for a fair use exception, but combining lots of derivative works can be copyright infringement even if none of the things you directly copied where. IE: If you copy a lot of excerpts from a poem and recreate the poem that doesn’t mean you can now use the poem.
This is the argument for why the model itself isn't infringing, isn't it? All of the excerpts as presented in their original context isn't the same thing as all of the excerpts purposefully rearranged back into the original work.
It's like on the one hand arguing that a dictionary is infringing because it contains all of the words in Harry Potter, and on the other hand claiming that you can reproduce the text of Harry Potter without liability because you reconstructed it by rearranging the words in a dictionary. One of these things is not like the other.
> If a US company is sued in Australia for copyright infringement they don’t get to move the case to the US.
If a US company has no operations in Australia, how does an Australian court have jurisdiction?
Fictional character names are copyrightable elements. Sure, it’s no issue for actual place names, but those aren’t the words people will be looking for.
> a dictionary is infringing because it contains all the words in Harry Potter
The argument is the model may have recreated the story of Harry Potter from being fed so much information about it.
DALL·E can’t contain a copy of every image in it’s corpus simply based on information theory, the model just isn’t that large. However it has no problem recreating the Mona Lisa because so many of its source images included it. That’s no problem for something out of copyright but it’s a big problem for major works in our culture like Harry Potter or Star Wars where potentially hundreds of thousands of derivative works get stitched together into a copy of the story.
If ask ChatGPT to summarize Harry’s first year at Hogwarts and it spits back a detailed summary of his year including dialog etc that’s wildly past the de minimis standard. I assume OpenAI has tried to prevent this, but this kind of reconstruction seems like a legal landline.
Copyright, or trademark?
> The argument is the model may have recreated the story of Harry Potter from being fed so much information about it.
It may contain enough information to recreate it, but is that the same thing?
Suppose you're a major blog platform and everybody's blog contains the data the model was trained on, i.e. enough to recreate the story of Harry Potter if you were to read them all and then be asked to summarize it or write fan fiction. If your company isn't infringing the book's copyright, why would the model be? It certainly contains no more information than it was trained on.
The output might be infringing, in the same way as a person who read all the blogs to learn the story could write something that was. But the output isn't the model.
Always copyright, sometimes both. Really though fictional works have a really low bar for copyright infringement.
> It may contain enough information to recreate it, but is that the same thing?
Yes. The law currently doesn’t give a shit about how you got the information just it’s ultimate origin. It’s the same reason MP3’s absolutely qualify as as copies even though none of the bits map 1:1 with the source material.
An LLM that reproduces a lossy copy of some work is just another lossy format. A judge isn’t going to care about AI magic, they are just going to see the infringement. The only defense is to not create copies of copyrighted works which shouldn’t be an issue if this stuff is actually transformative.
Remember lawsuits aren’t beyond a reasonable doubt the standard is much lower.
The value of names to copyright holders is simply how distinct they are. John Smith isn’t particularly unusual but if you name a bounty hunter Boba Fett it’s hard to argue originality. Complete originality isn’t required as long as you’re safely inside a fair use exception, but overwhelming originality is required.
So it’s generally safer for a character to talk about their favorite parts of Star Wars than have your stories plot be a group of rebels trying to destroy a planet killing doomsday weapon guarded by a sword wielding space wizards.
My point is that the meaning of "copying" as described here is unclear unclear if it's not restricted to verbatim plagiarism. As is "elements", for that matter.
Without details of what specific data a language model was trained on, there is no way to differentiate it having retrieved that name from a review of the book that was used with permission from the author (which is their own copyright and fair use on its own) or from the books itself (which would be copyright infringement).
If those records don't exist, the only mechanism you have left is to try and get the language model to spit out sufficient verbatim text from the source material to cross the fair use threshold. This doesn't work in the case of "obvious extrapolation" either which is a whole other defense that could be used depending on what that body produced was (If knowing only that the main character is named Harry, who is a wizard going to a wizarding school in the British countryside is required to produce a couple of close looking paragraphs, you need a much closer match with the original text over those two paragraphs for it to be infringing).
Do you think that DeepMind, OpenAI, etc. don't have this dataset copied 10 times over? Think again!
All the current LLMs are trained on data scraped from random websites, with no regard for the website's copyright, aren't they?
Presumably because these big businesses have a theory training an LLM is 'fair use'.
Why wouldn't they treat books the same way they treat web pages?
But it also feels like standing at the edge of the sea complaining about the tide coming in...I'm not sure there's really much that can / will be done about it.
That being said, it feels like there's also a shade of perspective from the old quote:
"In its majestic equality, the law forbids rich and poor alike to sleep under bridges, beg in the streets and steal loaves of bread." - assuming everything in public is fair game, then everyone is welcome to build a multi-petabyte database of text and use millions of dollars worth of GPUs to train an AI on it.
But saying "we steal to stay relevant, because we don't have as much funds as our competitors" is not the answer.
I understand that in Japan, for example, it is legal to apply deep analysis techniques, including deep learning and generative models to any data, regardless the copyright.
I would guess that in Japan, they’ve looked at China, where copyrights are also not an issue. And decided that it’s a really bad idea to stop research because of copyrights and let China to be first at the AGI race.
If a lot of time will get wasted on the GDPR and copyrights discussions in Europe and United States, instead of actually doing research, the democratic world will get behind.
Giant corporations that are scrambling to keep R&D going are quite law-abiding actually. And are currently struggling, as most of the datasets out there can’t be used, due to unclear copyright and licensing regulations.
Copyright ethics are something we have invented.
China chooses not to follow western copyright regulations. That’s not unethical, to them anyway.
Indeed, Copyright law was originally intended to allow people to legally copy things, not protect corporate interests for an eternity!
I still would like to see if legally a judge would consider the training illegal. I do not think this got tested and i would not consider the answer as clear cut. If a company buys a million ebooks, can it simply train an AI on it? If no employee reads the books and only the AI "reads" it you could argue that the company bought it only for a single reader/entity. They will probably need to update the copyright system to take this case into account. We will see in a few years how this plays out.
In the end these authors will eventually lose their copyright after their death. And nothing will stop the AI training at that point. This only delays the inevitable by a few years. The public domain already has a lot of amazing books and resources. Google even has access to very rare books nobody else has access to digitally i think:
https://blog.google/outreach-initiatives/arts-culture/in-beg...
122 points | 29 days ago | 130 comments
https://www.thenose.cc/wiki/Overview
Group Takes AI Training Dataset: good good good
We need neutral terminology for both sides. Even 'intellectual property' tries to transform a time-limited monopolistic license from government for intangible goods into eternal real estate. Any ideas?
Wittgenstein and Marcuse thought not - that words stand in as tools of intent. It is our hostile times that make weapons of words.
But it's frustrating because neutral prose is hard to reach. We seem in an age where it's hard to say anything succinctly, in a fashion that will pique the wretched attention span of the ordinary reader, that isn't either drenched in political bias or ungainly.
I think the media industry absolutely embarrassed themselves by misappropriating the word "piracy". Even after 30 years it still jars as awful, childish cringe. When I see a grown man in a suit talking of "piracy" I can't help but see a pitiful clown before me.
But one cannot police other's language. They must be allowed to make themselves known through the words they choose.
Saving my fellow nonUK people a click.
Is the answer to this situation a neutral position? I don't think so.
Intellectual monopoly is bullshit. I see no value in deescalating the narrative.
What we need is change. Change can't be made by taking a step back and deescalating the narrative.
I hope you never know the misfortune of having assets devalued by people that feel they don't need to operate within the boundaries of the rules.
If a company successfully lobbied and bribed the legislature to pass a law gving the company a monopoly on book publishing and sales, and someone bought books elsewhere, would that someone be doing something wrong? Are they harming that poor victim of a company?
Life isn't as simple as a hypothetical, of course, but it's also not as simple as assuming everything is just and fair.
Piracy is killing the motive to work. I've seen it happen countless times and nobody says a thing about it.
How about copyright becomes so pervasive and all enforced that massive corporations like Disney cannot bully the small fry into legally moving into their turf.
Copyright is an hellish nightmare because it used by big corporations to devour, rather than used to emancipate every single person.
ChatGPT ravaging and pillaging the internet was the last straw for "faith in freedom".
I'm putting all my faith and work into walking on water, establishing the cleansing task of getting out from under the sawblade.
People think to write is to compile and package in on itself like a computer program or a spirit-sucking business seminar. Typing on keyboards does that to a man.
Maybe with better training data we will have better models. There are some models trained only on public domain and synthetic (generated) data. So probably LLMs won't learn directly from our works. They need better data than internet stuff, human text is not good enough. See the Microsoft model Phi-1.5 [1], trained on diverse synthetic text of high quality.
[1] https://youtu.be/0lF3g4JtY9k?t=531
Since sharing of information is not a zero-sum game, areas where limiting access to an intellectual product is hard to achieve tend to build open and cooperative products instead. Of course it's much slower, and the output is smaller.
You’d have to reword it to something like “anti-digital theft” to bring back the negative connotation. Or have more articles about how piracy is destroying small businesses, which are suspiciously rare.
No, it absolutely does not. "Piracy" is a reasonably neutral word with a specific definition (it's slightly different than "theft") - you are artificially assigning nonexistent emotional weight to it in order to push your own political agenda.
Also, dupe, also old https://news.ycombinator.com/item?id=37154633
So, why is similar action not being taken against the actual big offenders -- Google, OpenAI, rest of big-tech?
Are we sure that copyright is really still suitable for purpose when it comes to promoting the Progress of Science and useful Arts?