8 ms·

It remains to be seen in court whether weights are even copyrightable potentially making all the various licenses and their restrictions moot.

FanaHOVA 3y ago
Yep, same with SSPL. GPL has been tested in FSF vs Cisco (2008), but none of the more restrictive licenses have.
dvdkon 3y ago
In a similar vein, the common "you may not use this model's output to improve another model" clause is AFAIK unenforceable under copyright, so it's at best a contractual clause binding a particular user. Anyone using that improved model afterward is in the clear.
- taneq 3y ago
  And of course anyone using a model improved by this is entirely unworried by these clauses if their improved model takes off hard.
- banana_feather 3y ago
  The idea is that if you violate the terms of the license to develop your own model, you lose your rights under the license and are creating an infringing derivative work. If I clone a GPL'd work and ship a derivative work under a commercial license, downstream users can't just integrate the derivative work into a product without abiding by the GPL terms and say "well we're downstream relative to the party who actually copied the GPL'd work, so the GPL terms don't apply to us".
  - blendergeek 3y ago
    If such a "derivative" model is a derivative work, then aren't all these LLMs just mass copyright infringement?
    - dTal 3y ago
      Exactly this. What's good for the goose is good for the gander!
    - banana_feather 3y ago
      At the end of the day it's not black and white, but there's a large and obvious difference in degree that would plausibly permit someone to find that one is and the other isn't. It's fairly easy to argue that using the outputs of LLM X to create a slightly more refined LLM Y creates a derivative work. The argument that a model is a derivative work relative to the training data is not so clear cut.
    - dragonwriter 3y ago
      If model weights aren’t copyrightable, derivative model weights are not a “work”, derivative or otherwise, for copyright purposes. If they are, and the license allows creating finetuned models but not using the output to improve the model, then the derived model is not a violation, but it might be a derivative work.
  - rodoxcasta 3y ago
    If the weights are not copyrighteable, you don't need a licence do use them, they are just data. There's not a right to infringe if these numbers have no author. Of course, to use openAI API you must abide to their terms. But if you publish your generations and I download them, I have nothing to do to the contract you have with openAI since I'm no part of it. They can't impede me to use it to improve my models.
  - diffeomorphism 3y ago
    Really? Your customers bought that product under license A. Afterwards it turned out that you pirated some artwork from disney. Then your customer can sue you (not disney) to make things right. The specific license of the original work seems quite irrelevant here.
    - pessimizer 3y ago
      Not at all. The reason your customer can sue you is because Disney can sue your customer. Disney would be suing your customer under the specific license of the original work. edit: you seem to see the customer as the primary victim here instead of Disney, but if Disney weren't a victim the customer wouldn't have a case.
  - dvdkon 3y ago
    Thing is, the outputs of a computer program aren't copyrightable, so it doesn't matter if your improved model is a derivative work. What you say would apply if you derived something from the weights themselves (assuming they are copyrightable, of course).
  - stale2002 3y ago
    No, because the premise of the hypothetical is that the weights aren't protected by copyright. So, no matter what they TOS says, it's not an infringing work. > Downstream users can't just integrate the derivative work into a product without abiding by the GPL terms You absolutely could do this if the original work is not protected by copyright, or if you use it in a way that is transformative and fair use.
    - mattl 3y ago
      Something under the GPL is also copyrighted. The GPL is a copyright license.
      - stale2002 3y ago
        If the underlying work is not protected by copyright, it doesn't matter what license someone tries to put on it. Similarly, if someone creates a fair use/transformative work then the license can also be ignored.
      - asadotzler 3y ago
        The GPL depends on copyright but it not itself copyright. The GPL is a license that gets its legal standing from copyright but if you don't have a copyright on something, slapping the GPL on top of it doesn't make it copyrighted to you.
        mattl 3y ago
        Absolutely.
- ljdcfsafsa 3y ago
  > it's at best a contractual clause binding a particular user. Anyone using that improved model afterward is in the clear. That's... not really accurate. See the concept of tortious interference with a contract.
  - dvdkon 3y ago
    Hm, I don't know much about common law, but I don't think this would apply if, say, an ML enthusiast trained a model from LLaMA2 outputs, made it freely available, then someone else commercialised it. The later user never caused the original developer to breach any contract, they simply profited from an existing breach. That said, doing this inside one company or with subsiduaries probably wouldn't fly.
weinzierl 3y ago
I find the idea that weights are not copyrightable very fascinating - appealing even. I have a hard time imagining a world where this is the case, though. Can you summarize why weights would not be copyrightable or give me pointers to sources that support that view.
- slimsag 3y ago
  Speculating (I am not a lawyer) I see two options: 1. Model weights are the output of mathematical principles, in the US facts are not copyrightable, so in general math is not copyrightable. 2. Model weights are the derivative work of all copyrighted works it was trained on - in which case, it would be similar to creating a new picture which contains every other picture in the world inside of it. Who is the copyright owner? Well, everyone, since it includes so many other copyright holders' works in it.
  - humanistbot 3y ago
    Your second argument, if true, disproves your first argument.
    - slimsag 3y ago
      Doesn't matter. A court decides in the end, and the two choices I presented could lead to OPs scenario. If a court decides that, they decide that, period. I'm not 'making an argument' with those points - I'm presenting options a court might choose from when setting precedent.
  - earleybird 3y ago
    Your second question asks: "Who owns the Infinite Library[0]?" related, there was a presentation (i've lost the reference) on automatic song (tune?) generation where the presenter claimed (rather humourusly) that he'd generated all the songs that had ever been and will ever be so that while he was infringing on a large but finite number of songs, he was non infringing on an infinite number of future songs. So, on balance he was in a favourable position. [0] https://en.wikipedia.org/wiki/The_Library_of_Babel https://en.wikipedia.org/wiki/The_Library_of_Babel
humanistbot 3y ago
And it also remains to be seen if various legislatures will pass laws that explicitly declare the copyright status of model weights. It is important to remember that what is or is not copyrightable can change.
- rvcdbn 3y ago
  At least in the US copyright is established by the constitution so not sure how much it’s possible to change via the normal legislative process.
  - gpm 3y ago
    The US constitution grants congress the ability to create copyright ("To promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries"), but it doesn't create copyright law itself. That's a broad clause that gives Congress pretty free reign to change how copyright is defined.
    - rvcdbn 3y ago
      Constitutionality is also about how previous cases have been evaluated for example see the bit about how photography copyright was established here: https://constitution.congress.gov/browse/essay/artI-S8-C8-3-1/ALDE_00013063/ https://constitution.congress.gov/browse/essay/artI-S8-C8-3-...
      - rvcdbn 3y ago
        specifically: > A century later, in Feist Publications v. Rural Telephone Service Co., the Supreme Court confirmed that originality is a constitutional requirement
ljdcfsafsa 3y ago
1. Why wouldn't they be and 2. Does that even matter? If you enter into a contract saying don't do X, and you do X, you're violating the contract.
- sebzim4500 3y ago
  I assume GP was talking about a scenario in which you had not entered into a contract with Meta. E.g. if I just downloaded the weights from someone else.
- rvcdbn 3y ago
  1 - because they lack originality, see: https://constitution.congress.gov/browse/essay/artI-S8-C8-3-1/ALDE_00013063/ https://constitution.congress.gov/browse/essay/artI-S8-C8-3-...
jrockway 3y ago
It seems like a dangerous clause to me. 1) "Dear artists, the model cannot infringe upon your copyright because it's merely learning like a human does. If it accidentally outputs parts of your book, you know, it just accidentally plagiarized. We all do it haha! Our attorneys remind you that plagiarism is not illegal in the US." 2) "Dear engineers, the output of our model is copyrighted and thus if you use it to train your own model, we own it." I am not sure how both of those can be true at the same time.
- jimmaswell 3y ago
  We all truly do "accidentally plagiarize", especially artists. Many guitarists realize they accidentally copied a riff they thought they'd come up with on their own for example.
  - jrockway 3y ago
    I, for one, welcome our new plagiarism overlords. Oops. I added the "haha" in there because the probability of a human doing this kind of goes way down as the length of the text increases. Can you type, verbatim, an entire chapter of a book? I can't. But, I bet the AI can be convinced in rare cases to do that. The whole thing is very interesting to me. There was an article on here a couple days ago about using gzip as a language model. Of course, gzipping a book doesn't remove the copyright. So how low does the probability of outputting the input verbatim have to be before copyright is lost? Reading the book and benefitting from what you learned? Obviously not copyright infringement. Putting the book into gzip and sending your friend the result? Obviously copyright infringement. Now we're in the grey area and ... nobody knows what the law is, or honestly, even how to reason about what the law wants here. Fun times. (Personally, I lean towards "not copyright infringement", but I'm not a big believer in copyright myself. In the case of AI training, it just makes it impossible for small actors to compete. Google can just buy a license from every book distributor. SmolStartup can't. So if we want to make AI that is only for the rich and powerful, copyright is the perfect tool to enable that. I don't think we want that, though. My take is that the rest of society kind of hates Tech right now ("I don't really like my Facebook friends, so someone should take away Mark Zuckerberg's money."), so it's likely that protectionist laws will soon be created that ruin it for everyone. The net effect of that is that Europe and the US will simply flat-out lose to China, which doesn't care about IP.)
sangnoir 3y ago
If they are nor copyrightable, that'll be the end of publicly-released weights by for-profit companies. All subsequent models will be served behind an API.
- dragonwriter 3y ago
  > If they are nor copyrightable, that'll be the end of publicly-released weights by for-profit companies I don’t see why, for-profit companies release permissively-licensed ooen-source code all the time, and noncopyrightable models aren't practically much different than that.
  - bilbo0s 3y ago
    Because the courts will have determined their business models for them. As mercenary as it may sound, what these companies are trying to do is find a business model that is as friendly to themselves as it is hostile to their competitors. This is all part of the jockeying.
    - dragonwriter 3y ago
      And, sure, lack of copyrightability changes the parameters and will change behavior. What I think you have failed to support is that the particular change that it will induce will eliminate all such releases.
  - sangnoir 3y ago
    I debated whether to be more specific and verbose in my earlier comment and brevity won at the expense of clarity. I meant large models that cost 6 or 7 digits to train likely won't be released if the donor company can't control how the models are used. > I don’t see why, for-profit companies release permissively-licensed ooen-source code all the time I agree with this - however, they tend to open-source non-core components - Google won't release search engine code, Amazon wont release scalable-virtualization-in-a-box, etc. I'm confident that Facebook won't release a hypothetical Llama 5 in a manner that enables it to be used to improve ChatGPT 8 - the aim will be unchanged from today, byt the mechanism will shift from licensing to rate-limiting, authentication & IP-bans.

FanaHOVA 3y ago

Yep, same with SSPL. GPL has been tested in FSF vs Cisco (2008), but none of the more restrictive licenses have.

dvdkon 3y ago

In a similar vein, the common "you may not use this model's output to improve another model" clause is AFAIK unenforceable under copyright, so it's at best a contractual clause binding a particular user. Anyone using that improved model afterward is in the clear.

taneq 3y ago
And of course anyone using a model improved by this is entirely unworried by these clauses if their improved model takes off hard.
banana_feather 3y ago
The idea is that if you violate the terms of the license to develop your own model, you lose your rights under the license and are creating an infringing derivative work. If I clone a GPL'd work and ship a derivative work under a commercial license, downstream users can't just integrate the derivative work into a product without abiding by the GPL terms and say "well we're downstream relative to the party who actually copied the GPL'd work, so the GPL terms don't apply to us".
- blendergeek 3y ago
  If such a "derivative" model is a derivative work, then aren't all these LLMs just mass copyright infringement?
  - dTal 3y ago
    Exactly this. What's good for the goose is good for the gander!
  - banana_feather 3y ago
    At the end of the day it's not black and white, but there's a large and obvious difference in degree that would plausibly permit someone to find that one is and the other isn't. It's fairly easy to argue that using the outputs of LLM X to create a slightly more refined LLM Y creates a derivative work. The argument that a model is a derivative work relative to the training data is not so clear cut.
  - dragonwriter 3y ago
    If model weights aren’t copyrightable, derivative model weights are not a “work”, derivative or otherwise, for copyright purposes. If they are, and the license allows creating finetuned models but not using the output to improve the model, then the derived model is not a violation, but it might be a derivative work.
- rodoxcasta 3y ago
  If the weights are not copyrighteable, you don't need a licence do use them, they are just data. There's not a right to infringe if these numbers have no author. Of course, to use openAI API you must abide to their terms. But if you publish your generations and I download them, I have nothing to do to the contract you have with openAI since I'm no part of it. They can't impede me to use it to improve my models.
- diffeomorphism 3y ago
  Really? Your customers bought that product under license A. Afterwards it turned out that you pirated some artwork from disney. Then your customer can sue you (not disney) to make things right. The specific license of the original work seems quite irrelevant here.
  - pessimizer 3y ago
    Not at all. The reason your customer can sue you is because Disney can sue your customer. Disney would be suing your customer under the specific license of the original work. edit: you seem to see the customer as the primary victim here instead of Disney, but if Disney weren't a victim the customer wouldn't have a case.
- dvdkon 3y ago
  Thing is, the outputs of a computer program aren't copyrightable, so it doesn't matter if your improved model is a derivative work. What you say would apply if you derived something from the weights themselves (assuming they are copyrightable, of course).
- stale2002 3y ago
  No, because the premise of the hypothetical is that the weights aren't protected by copyright. So, no matter what they TOS says, it's not an infringing work. > Downstream users can't just integrate the derivative work into a product without abiding by the GPL terms You absolutely could do this if the original work is not protected by copyright, or if you use it in a way that is transformative and fair use.
  - mattl 3y ago
    Something under the GPL is also copyrighted. The GPL is a copyright license.
    - stale2002 3y ago
      If the underlying work is not protected by copyright, it doesn't matter what license someone tries to put on it. Similarly, if someone creates a fair use/transformative work then the license can also be ignored.
    - asadotzler 3y ago
      The GPL depends on copyright but it not itself copyright. The GPL is a license that gets its legal standing from copyright but if you don't have a copyright on something, slapping the GPL on top of it doesn't make it copyrighted to you.
      - mattl 3y ago
        Absolutely.
ljdcfsafsa 3y ago
> it's at best a contractual clause binding a particular user. Anyone using that improved model afterward is in the clear. That's... not really accurate. See the concept of tortious interference with a contract.
- dvdkon 3y ago
  Hm, I don't know much about common law, but I don't think this would apply if, say, an ML enthusiast trained a model from LLaMA2 outputs, made it freely available, then someone else commercialised it. The later user never caused the original developer to breach any contract, they simply profited from an existing breach. That said, doing this inside one company or with subsiduaries probably wouldn't fly.

weinzierl 3y ago

I find the idea that weights are not copyrightable very fascinating - appealing even. I have a hard time imagining a world where this is the case, though. Can you summarize why weights would not be copyrightable or give me pointers to sources that support that view.

slimsag 3y ago
Speculating (I am not a lawyer) I see two options: 1. Model weights are the output of mathematical principles, in the US facts are not copyrightable, so in general math is not copyrightable. 2. Model weights are the derivative work of all copyrighted works it was trained on - in which case, it would be similar to creating a new picture which contains every other picture in the world inside of it. Who is the copyright owner? Well, everyone, since it includes so many other copyright holders' works in it.
- humanistbot 3y ago
  Your second argument, if true, disproves your first argument.
  - slimsag 3y ago
    Doesn't matter. A court decides in the end, and the two choices I presented could lead to OPs scenario. If a court decides that, they decide that, period. I'm not 'making an argument' with those points - I'm presenting options a court might choose from when setting precedent.
- earleybird 3y ago
  Your second question asks: "Who owns the Infinite Library[0]?" related, there was a presentation (i've lost the reference) on automatic song (tune?) generation where the presenter claimed (rather humourusly) that he'd generated all the songs that had ever been and will ever be so that while he was infringing on a large but finite number of songs, he was non infringing on an infinite number of future songs. So, on balance he was in a favourable position. [0] https://en.wikipedia.org/wiki/The_Library_of_Babel https://en.wikipedia.org/wiki/The_Library_of_Babel

humanistbot 3y ago

And it also remains to be seen if various legislatures will pass laws that explicitly declare the copyright status of model weights. It is important to remember that what is or is not copyrightable can change.

rvcdbn 3y ago
At least in the US copyright is established by the constitution so not sure how much it’s possible to change via the normal legislative process.
- gpm 3y ago
  The US constitution grants congress the ability to create copyright ("To promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries"), but it doesn't create copyright law itself. That's a broad clause that gives Congress pretty free reign to change how copyright is defined.
  - rvcdbn 3y ago
    Constitutionality is also about how previous cases have been evaluated for example see the bit about how photography copyright was established here: https://constitution.congress.gov/browse/essay/artI-S8-C8-3-1/ALDE_00013063/ https://constitution.congress.gov/browse/essay/artI-S8-C8-3-...
    - rvcdbn 3y ago
      specifically: > A century later, in Feist Publications v. Rural Telephone Service Co., the Supreme Court confirmed that originality is a constitutional requirement

ljdcfsafsa 3y ago

1. Why wouldn't they be and 2. Does that even matter? If you enter into a contract saying don't do X, and you do X, you're violating the contract.

sebzim4500 3y ago
I assume GP was talking about a scenario in which you had not entered into a contract with Meta. E.g. if I just downloaded the weights from someone else.
rvcdbn 3y ago
1 - because they lack originality, see: https://constitution.congress.gov/browse/essay/artI-S8-C8-3-1/ALDE_00013063/ https://constitution.congress.gov/browse/essay/artI-S8-C8-3-...

jrockway 3y ago

It seems like a dangerous clause to me. 1) "Dear artists, the model cannot infringe upon your copyright because it's merely learning like a human does. If it accidentally outputs parts of your book, you know, it just accidentally plagiarized. We all do it haha! Our attorneys remind you that plagiarism is not illegal in the US." 2) "Dear engineers, the output of our model is copyrighted and thus if you use it to train your own model, we own it." I am not sure how both of those can be true at the same time.

jimmaswell 3y ago
We all truly do "accidentally plagiarize", especially artists. Many guitarists realize they accidentally copied a riff they thought they'd come up with on their own for example.
- jrockway 3y ago
  I, for one, welcome our new plagiarism overlords. Oops. I added the "haha" in there because the probability of a human doing this kind of goes way down as the length of the text increases. Can you type, verbatim, an entire chapter of a book? I can't. But, I bet the AI can be convinced in rare cases to do that. The whole thing is very interesting to me. There was an article on here a couple days ago about using gzip as a language model. Of course, gzipping a book doesn't remove the copyright. So how low does the probability of outputting the input verbatim have to be before copyright is lost? Reading the book and benefitting from what you learned? Obviously not copyright infringement. Putting the book into gzip and sending your friend the result? Obviously copyright infringement. Now we're in the grey area and ... nobody knows what the law is, or honestly, even how to reason about what the law wants here. Fun times. (Personally, I lean towards "not copyright infringement", but I'm not a big believer in copyright myself. In the case of AI training, it just makes it impossible for small actors to compete. Google can just buy a license from every book distributor. SmolStartup can't. So if we want to make AI that is only for the rich and powerful, copyright is the perfect tool to enable that. I don't think we want that, though. My take is that the rest of society kind of hates Tech right now ("I don't really like my Facebook friends, so someone should take away Mark Zuckerberg's money."), so it's likely that protectionist laws will soon be created that ruin it for everyone. The net effect of that is that Europe and the US will simply flat-out lose to China, which doesn't care about IP.)

sangnoir 3y ago

If they are nor copyrightable, that'll be the end of publicly-released weights by for-profit companies. All subsequent models will be served behind an API.

dragonwriter 3y ago
> If they are nor copyrightable, that'll be the end of publicly-released weights by for-profit companies I don’t see why, for-profit companies release permissively-licensed ooen-source code all the time, and noncopyrightable models aren't practically much different than that.
- bilbo0s 3y ago
  Because the courts will have determined their business models for them. As mercenary as it may sound, what these companies are trying to do is find a business model that is as friendly to themselves as it is hostile to their competitors. This is all part of the jockeying.
  - dragonwriter 3y ago
    And, sure, lack of copyrightability changes the parameters and will change behavior. What I think you have failed to support is that the particular change that it will induce will eliminate all such releases.
- sangnoir 3y ago
  I debated whether to be more specific and verbose in my earlier comment and brevity won at the expense of clarity. I meant large models that cost 6 or 7 digits to train likely won't be released if the donor company can't control how the models are used. > I don’t see why, for-profit companies release permissively-licensed ooen-source code all the time I agree with this - however, they tend to open-source non-core components - Google won't release search engine code, Amazon wont release scalable-virtualization-in-a-box, etc. I'm confident that Facebook won't release a hypothetical Llama 5 in a manner that enables it to be used to improve ChatGPT 8 - the aim will be unchanged from today, byt the mechanism will shift from licensing to rate-limiting, authentication & IP-bans.