6 ms·

Courts (at least in the US) have already ruled that use of ingested data for training is transformative. There’s lots of details to figure, but the genie is out

Courts (at least in the US) have already ruled that use of ingested data for training is transformative. There’s lots of details to figure, but the genie is out of the bottle. Sure it’s a big hill to climb in rethinking IP laws to align with a societal desire that generating IP continue to be a viable economic work product, but that is what’s necessary.

alfalfasprout 1y ago
> Courts (at least in the US) have already ruled that use of ingested data for training is transformative This is far from settled law. Let's not mischaracterize it. Even so, an AI regurgitating proprietary code that's licensed in some other way is a very real risk.
- popalchemist 1y ago
  No more so than regurgitating an entire book. While it could technically be possible in the case of certain repos that are ubiquitous on the internet (and therefore overrepresented in training data to the point that they are "regurgitated" verbatim, in whole), it is extremely unlikely and would only occur after deliberate prompting. The NYT suit against Open AI shows (in discovery) that the NYT was only able to get partial results after deliberately prompting the model with portions of the text they were trying to force it to regurgitate. So. Yes, technically possible. But impossible by accident. Furthermore when you make this argument you reveal that you don't understand how these models work. They do not simply compress all the data they were trained on into a tiny storable version. They are effectively multiplication matrices that allow math to be done to predict the most likely next token (read: 2-3 Unicode characters) given some input. So the model does not "contain" code. It "contains" a way of doing calculations for predicting what text comes next. Finally, let's say that it is possible that the model does spit out not entire works, but a handful of lines of code that appear in some codebase. This does not constitute copyright infringement, as the lines in question a) represent a tiny portion of the whole work (and copyright only protecst against the reduplication of whole works or siginficant portions of the work), and B) there are a limited number of ways to accomplish a certain function and it is not only possible but inevitable that two devs working independently could arrive at the same implementation. Therefore using an identical implementation (which is what this case would be) of a part of a work is no more illegal than the use of a certain chord progression or melodic phrasing or drum rhythm. Courts have ruled about this thoroughly.
  - typpilol 1y ago
    It's also why some companies do clean room design.
  - aspenmayer 1y ago
    > No more so than regurgitating an entire book. Like this? Meta's Llama 3.1 can recall 42 percent of the first Harry Potter book - https://news.ycombinator.com/context?id=44972296 https://news.ycombinator.com/context?id=44972296 - 67 days ago (313 comments)
    - popalchemist 1y ago
      Yes, that is one of those works that is over-represented in the training data, as I explained in the part of the comment you clearly did not comprehend.
      - aspenmayer 1y ago
        > you clearly did not comprehend I comprehend it just fine, I was adding context for those who may not comprehend.
Hamuko 1y ago
Training an AI model is not the same as using an AI model.
- ryukoposting 1y ago
  Publishing pirated copies of books on libgen isn't the same as downloading pirated copies of books from libgen. Neither is legal.
  - eru 1y ago
    In many places that's a fairly recent development: publishing pirated IP used to be much more of a legal problem than consuming it. Also publishing pirated IP without any monetary gain to yourself also used to be treated more leniently. Of course, all the rules were changed (both in law and in interpretation in practice) as file sharing became a huge deal about two decades ago. Details depend on jurisdiction.
  - MacsHeadroom 1y ago
    Many jurisdictions outlaw sharing but not copying. Downloading is legal in those places but publishing is not.
- ineedasername 1y ago
  If you mean the ruling has absolutely no applicability when it comes to using the model then, no, that is incorrect: Judge Alsup, in his ruling, specifically likened the process to reading text and then using the knowledge to write something else. That’s training and use.
  - Hamuko 1y ago
    How are you verifying that the AI is writing "something else" and not something that's verbatum in its training material?
bsder 1y ago
> Courts (at least in the US) have already ruled that use of ingested data for training is transformative. If you have code that happens to be identical to some else's code or implements someone's proprietary algorithm, you're going to lose in court even if you claim an "AI" gave it to you. AI is training on private Github repos and coughing them up. I've had it regurgitate a very well written piece of code to do a particular computational geometry algorithm. It presented perfect, idiomatic Python with perfect tests that caught all the degenerate cases. That was obviously proprietary code--no amount of searching came up with anything even remotely close (it's why I asked the AI, after all).
- ineedasername 1y ago
  >If you have code that happens to be identical to some else's code or implements someone's proprietary algorithm, you're going to lose in court even if you claim an "AI" gave it to you. Not for a dozen lines here or there, even if it could be found and identified in a massive code base. That’s like quoting a paragraph of a book in another book, non infringing. For the second half of your comment it sounds like you’re saying you got results that were too good to be AI- that’s a bit “no true Scotsman”, at least without more detail. But implementing an algorithm, even a complex one, is very much something an LLM can do. Algorithms are much better defined and scoped natural language, and LLMs do a reasonable job of translating to languages. An algorithm is a narrow subset of that task type with better defined context and syntax.
  - ozfive 1y ago
    What will happen when company A implements algorithm X based on AI output, company B does the same and company A claims that it is proprietary code and takes company B to court?
    - _w1tm 1y ago
      What has happened when the same thing happens without AI involved?
eschaton 1y ago
Some courts at some levels. It’s by no means settled law.
- bigyabai 1y ago
  Most licensed software is unsettled law, if we're being that pedantic.
  - eschaton 1y ago
    Not really, no. If you’re specifically referring to, say, GPL or BSD or other Open Source licenses, it’s a bit more unsettled, but software licensing as a whole has several decades of case law at this point.
eru 1y ago
> Sure it’s a big hill to climb in rethinking IP laws to align with a societal desire that generating IP continue to be a viable economic work product, but that is what’s necessary. Well, AI can perhaps solve the problem it created here: generated IP with AI is much cheaper than with humans, so it will be viable even at lower payoffs. Less cynical: you can use trade secrets to protect your IP. You can host your software and only let customers interact with it remotely, like what Google (mostly) does. Of course, this is a very software-centric view. You can't 'protect' eg books or music in this way.
- raggi 1y ago
  In the US you can not generate copyrightable IP without substantial human contribution to the process. https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf https://www.copyright.gov/ai/Copyright-and-Artificial-Intell...
  - aspenmayer 1y ago
    Tell that to Reddit. They’re AI translating user posts and serving it up as separate Google search results. I don’t remember if Reddit claims copyright on user-submitted content, or on its AI translations, but I don’t think Reddit is paying ad share like X is, either, so it kind of doesn’t matter to the user, as they’re (still) not getting paid, even as Reddit collects money for every ad shown/clicked. Even if OP did write it, an AI translated the version shown. https://news.ycombinator.com/context?id=44972296 https://news.ycombinator.com/context?id=44972296
    - raggi 1y ago
      reddit is a user hostile company, have been forever, always will be. they take rights over your content, farm things about you, sell data, do invasive things in the mobile apps, use creepware cookies, etc. Excerpt from the user agreement: When Your Content is created with or submitted to the Services, you grant us a worldwide, royalty-free, perpetual, irrevocable, non-exclusive, transferable, and sublicensable license to use, copy, modify, adapt, prepare derivative works of, distribute, store, perform, and display Your Content and any name, username, voice, or likeness provided in connection with Your Content in all media formats and channels now known or later developed anywhere in the world. This license includes the right for us to make Your Content available for syndication, broadcast, distribution, or publication by other companies, organizations, or individuals who partner with Reddit. For example, this license includes the right to use Your Content to train AI and machine learning models, as further described in our Public Content Policy. You also agree that we may remove metadata associated with Your Content, and you irrevocably waive any claims and assertions of moral rights or attribution with respect to Your Content. People put their heads in the sand over reddit for some reason, but it's worse than FAANG.
shkkmo 1y ago
> Courts (at least in the US) have already ruled that use of ingested data for training is transformative Yes, the training of the model itself is (or should be) a transformative act so you can train a model on whatever you have legal access to view. However, that doesn't mean that the output of the model is automatically not infringing. If the model is prompted to create a copy of some copyrighted work, that is (or should be) still a violation. Just like memorizing a book isn't infringment but reproducing a book from memory is.
- anp 1y ago
  This also matches my (not a lawyer) intuition, but have there been any legal precedents set in this direction yet?
- threecheese 1y ago
  The fact that GitHub’s Copilot has an enterprise feature that matches model output against code having certain licenses - in order to prevent you from using it, with a notification - suggests the model outputs are at least potentially infringing. If MS were compelled to reveal how these completions are generated, there’s at least a possibility that they directly use public repositories to source text chunks that their “model” suggested were relevant (quoted as it could be more than just a model, like vector or search databases or some other orchestration across multiple workloads).
  - martin-t 1y ago
    > directly use public repositories I don't see why a company which has been waging a multi decade war against GPL and users' rights would stop at _public_ repositories.
  - ineedasername 1y ago
    > suggests the model outputs are at least potentially infringing. The only thing it suggests is that they recognize that a subset of users worry about it. Whether or not GitHub worries about it any further isn’t suggested. Don’t think about it from an actual “rights” perspective. Think about the entire copyright issue as a “too big to fail” issue.
jhanschoo 1y ago
An AI model's output can be transformative, but you can be unlucky enough that the LLM memorized the data that it gave you.
- martin-t 1y ago
  I don't see why verbatim or not should matter at all. How complex does a mechanical transformation have to be to not be considered plagiarism, copyright infringement or parasitism? If somebody writes a GPL-licensed program, is it enough to change all variable and function names to get rid of those pesky users' rights? Do you have to change the order of functions? Do you have to convert it to a different language? Surely nobody would claim c2rust is transformative even though the resulting code can be wildly different if you apply enough mechanical transformations. All LLMs do is make the mechanical transformations 1) probabilistic 2) opaque 3) all at once 4) using multiple projects as a source.
  - jhanschoo 1y ago
    > How complex does a mechanical transformation have to be to not be considered plagiarism, copyright infringement or parasitism? Legally speaking, this depends from domain to domain. But consider for example extracting facts from several biology textbooks, and then delivering those facts to the user in the characteristic ChatGPT tone that is distinguishable from the style of each source textbook. You can then be quite assured that courts will not find that you have infringed on copyright.
slg 1y ago
>societal desire that generating IP continue to be a viable economic work product It is strange that you think the law is settled when I don't think even this "societal desire" is completely settled just yet.
- ineedasername 1y ago
  Maybe I should clarify: Society, in general, supports the idea that writers, artists, film makers, coders, etc— everyone who creates IP- should have a place in the economy. Basically just that it should be possible to make a living and have a career at it. It can be spun different ways, and those differences are important, but this is the basic thing. This doesn’t seem like a disputable statement to me. For anyone who thinks actors’ likenesses, authors’ words, all of it- that all and everything should be up for grabs once written or put anywhere in public, that is not a widely held opinion. Once that’s established, it all comes down to implementation details.
  - ekianjo 1y ago
    > Society, in general, supports the idea that writers, artists, film makers, coders, etc Coders don't get paid every single time their code runs. Why bundle different rights together?
    - aspenmayer 1y ago
      > Coders don't get paid every single time their code runs. They do if they code the API correctly. > Why bundle different rights together? Why are mineral rights sold separately to most land deeds?
      - aleph_minus_one 1y ago
        > Why are mineral rights sold separately to most land deeds? Because the population does not rebel against the politicians that made these laws.
    - ineedasername 1y ago
BobbyTables2 1y ago
I’m curious … So “transformative” is not necessarily “derivative”? Seems to me the training of AI is not radically different than compression algorithms building up a dictionary and compressing data. Yet nobody calls JPEG compression “transformative”. Could one do lossy compression over billions of copyrighted images to “train” a dictionary?
- zahlman 1y ago
  > I’m curious … So “transformative” is not necessarily “derivative”? (not legal advice) Transformative works are necessarily derivative, but that transformation allows for a legal claim to "fair use" regardless of making a derived work. https://en.wikipedia.org/wiki/Transformative_use https://en.wikipedia.org/wiki/Transformative_use
- ipaddr 1y ago
  A compression algorithm doesn't transform the data it stores it in a different format. Storing a story in a txt file vs word file doesn't transform the data. An llm is looking at the shape of words and ideas over scale and using that to provide answers.
  - const_cast 1y ago
    No a compression algorithm does transform the data, particularly lossy ones. The pixels stored in the output are not in the input, they're new pixels. That's why you can't uncompress a jpeg. Its a new image that just happens to look like the original. But it even might not - some jpegs are so deep fried they become their own form of art. This is very popular in meme culture. The only difference, really, is we know how a JPEG algorithm works. If I wanted to, I could painstakingly make a jpeg by hand. We don't know how LLMs work.

alfalfasprout 1y ago

> Courts (at least in the US) have already ruled that use of ingested data for training is transformative This is far from settled law. Let's not mischaracterize it. Even so, an AI regurgitating proprietary code that's licensed in some other way is a very real risk.

popalchemist 1y ago
No more so than regurgitating an entire book. While it could technically be possible in the case of certain repos that are ubiquitous on the internet (and therefore overrepresented in training data to the point that they are "regurgitated" verbatim, in whole), it is extremely unlikely and would only occur after deliberate prompting. The NYT suit against Open AI shows (in discovery) that the NYT was only able to get partial results after deliberately prompting the model with portions of the text they were trying to force it to regurgitate. So. Yes, technically possible. But impossible by accident. Furthermore when you make this argument you reveal that you don't understand how these models work. They do not simply compress all the data they were trained on into a tiny storable version. They are effectively multiplication matrices that allow math to be done to predict the most likely next token (read: 2-3 Unicode characters) given some input. So the model does not "contain" code. It "contains" a way of doing calculations for predicting what text comes next. Finally, let's say that it is possible that the model does spit out not entire works, but a handful of lines of code that appear in some codebase. This does not constitute copyright infringement, as the lines in question a) represent a tiny portion of the whole work (and copyright only protecst against the reduplication of whole works or siginficant portions of the work), and B) there are a limited number of ways to accomplish a certain function and it is not only possible but inevitable that two devs working independently could arrive at the same implementation. Therefore using an identical implementation (which is what this case would be) of a part of a work is no more illegal than the use of a certain chord progression or melodic phrasing or drum rhythm. Courts have ruled about this thoroughly.
- typpilol 1y ago
  It's also why some companies do clean room design.
- aspenmayer 1y ago
  > No more so than regurgitating an entire book. Like this? Meta's Llama 3.1 can recall 42 percent of the first Harry Potter book - https://news.ycombinator.com/context?id=44972296 https://news.ycombinator.com/context?id=44972296 - 67 days ago (313 comments)
  - popalchemist 1y ago
    Yes, that is one of those works that is over-represented in the training data, as I explained in the part of the comment you clearly did not comprehend.
    - aspenmayer 1y ago
      > you clearly did not comprehend I comprehend it just fine, I was adding context for those who may not comprehend.

Hamuko 1y ago

Training an AI model is not the same as using an AI model.

ryukoposting 1y ago
Publishing pirated copies of books on libgen isn't the same as downloading pirated copies of books from libgen. Neither is legal.
- eru 1y ago
  In many places that's a fairly recent development: publishing pirated IP used to be much more of a legal problem than consuming it. Also publishing pirated IP without any monetary gain to yourself also used to be treated more leniently. Of course, all the rules were changed (both in law and in interpretation in practice) as file sharing became a huge deal about two decades ago. Details depend on jurisdiction.
- MacsHeadroom 1y ago
  Many jurisdictions outlaw sharing but not copying. Downloading is legal in those places but publishing is not.
ineedasername 1y ago
If you mean the ruling has absolutely no applicability when it comes to using the model then, no, that is incorrect: Judge Alsup, in his ruling, specifically likened the process to reading text and then using the knowledge to write something else. That’s training and use.
- Hamuko 1y ago
  How are you verifying that the AI is writing "something else" and not something that's verbatum in its training material?

bsder 1y ago

> Courts (at least in the US) have already ruled that use of ingested data for training is transformative. If you have code that happens to be identical to some else's code or implements someone's proprietary algorithm, you're going to lose in court even if you claim an "AI" gave it to you. AI is training on private Github repos and coughing them up. I've had it regurgitate a very well written piece of code to do a particular computational geometry algorithm. It presented perfect, idiomatic Python with perfect tests that caught all the degenerate cases. That was obviously proprietary code--no amount of searching came up with anything even remotely close (it's why I asked the AI, after all).

ineedasername 1y ago
>If you have code that happens to be identical to some else's code or implements someone's proprietary algorithm, you're going to lose in court even if you claim an "AI" gave it to you. Not for a dozen lines here or there, even if it could be found and identified in a massive code base. That’s like quoting a paragraph of a book in another book, non infringing. For the second half of your comment it sounds like you’re saying you got results that were too good to be AI- that’s a bit “no true Scotsman”, at least without more detail. But implementing an algorithm, even a complex one, is very much something an LLM can do. Algorithms are much better defined and scoped natural language, and LLMs do a reasonable job of translating to languages. An algorithm is a narrow subset of that task type with better defined context and syntax.
- ozfive 1y ago
  What will happen when company A implements algorithm X based on AI output, company B does the same and company A claims that it is proprietary code and takes company B to court?
  - _w1tm 1y ago
    What has happened when the same thing happens without AI involved?

eschaton 1y ago

Some courts at some levels. It’s by no means settled law.

bigyabai 1y ago
Most licensed software is unsettled law, if we're being that pedantic.
- eschaton 1y ago
  Not really, no. If you’re specifically referring to, say, GPL or BSD or other Open Source licenses, it’s a bit more unsettled, but software licensing as a whole has several decades of case law at this point.

eru 1y ago

> Sure it’s a big hill to climb in rethinking IP laws to align with a societal desire that generating IP continue to be a viable economic work product, but that is what’s necessary. Well, AI can perhaps solve the problem it created here: generated IP with AI is much cheaper than with humans, so it will be viable even at lower payoffs. Less cynical: you can use trade secrets to protect your IP. You can host your software and only let customers interact with it remotely, like what Google (mostly) does. Of course, this is a very software-centric view. You can't 'protect' eg books or music in this way.

raggi 1y ago
In the US you can not generate copyrightable IP without substantial human contribution to the process. https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf https://www.copyright.gov/ai/Copyright-and-Artificial-Intell...
- aspenmayer 1y ago
  Tell that to Reddit. They’re AI translating user posts and serving it up as separate Google search results. I don’t remember if Reddit claims copyright on user-submitted content, or on its AI translations, but I don’t think Reddit is paying ad share like X is, either, so it kind of doesn’t matter to the user, as they’re (still) not getting paid, even as Reddit collects money for every ad shown/clicked. Even if OP did write it, an AI translated the version shown. https://news.ycombinator.com/context?id=44972296 https://news.ycombinator.com/context?id=44972296
  - raggi 1y ago
    reddit is a user hostile company, have been forever, always will be. they take rights over your content, farm things about you, sell data, do invasive things in the mobile apps, use creepware cookies, etc. Excerpt from the user agreement: When Your Content is created with or submitted to the Services, you grant us a worldwide, royalty-free, perpetual, irrevocable, non-exclusive, transferable, and sublicensable license to use, copy, modify, adapt, prepare derivative works of, distribute, store, perform, and display Your Content and any name, username, voice, or likeness provided in connection with Your Content in all media formats and channels now known or later developed anywhere in the world. This license includes the right for us to make Your Content available for syndication, broadcast, distribution, or publication by other companies, organizations, or individuals who partner with Reddit. For example, this license includes the right to use Your Content to train AI and machine learning models, as further described in our Public Content Policy. You also agree that we may remove metadata associated with Your Content, and you irrevocably waive any claims and assertions of moral rights or attribution with respect to Your Content. People put their heads in the sand over reddit for some reason, but it's worse than FAANG.

shkkmo 1y ago

> Courts (at least in the US) have already ruled that use of ingested data for training is transformative Yes, the training of the model itself is (or should be) a transformative act so you can train a model on whatever you have legal access to view. However, that doesn't mean that the output of the model is automatically not infringing. If the model is prompted to create a copy of some copyrighted work, that is (or should be) still a violation. Just like memorizing a book isn't infringment but reproducing a book from memory is.

anp 1y ago
This also matches my (not a lawyer) intuition, but have there been any legal precedents set in this direction yet?
threecheese 1y ago
The fact that GitHub’s Copilot has an enterprise feature that matches model output against code having certain licenses - in order to prevent you from using it, with a notification - suggests the model outputs are at least potentially infringing. If MS were compelled to reveal how these completions are generated, there’s at least a possibility that they directly use public repositories to source text chunks that their “model” suggested were relevant (quoted as it could be more than just a model, like vector or search databases or some other orchestration across multiple workloads).
- martin-t 1y ago
  > directly use public repositories I don't see why a company which has been waging a multi decade war against GPL and users' rights would stop at _public_ repositories.
- ineedasername 1y ago
  > suggests the model outputs are at least potentially infringing. The only thing it suggests is that they recognize that a subset of users worry about it. Whether or not GitHub worries about it any further isn’t suggested. Don’t think about it from an actual “rights” perspective. Think about the entire copyright issue as a “too big to fail” issue.

jhanschoo 1y ago

An AI model's output can be transformative, but you can be unlucky enough that the LLM memorized the data that it gave you.

martin-t 1y ago
I don't see why verbatim or not should matter at all. How complex does a mechanical transformation have to be to not be considered plagiarism, copyright infringement or parasitism? If somebody writes a GPL-licensed program, is it enough to change all variable and function names to get rid of those pesky users' rights? Do you have to change the order of functions? Do you have to convert it to a different language? Surely nobody would claim c2rust is transformative even though the resulting code can be wildly different if you apply enough mechanical transformations. All LLMs do is make the mechanical transformations 1) probabilistic 2) opaque 3) all at once 4) using multiple projects as a source.
- jhanschoo 1y ago
  > How complex does a mechanical transformation have to be to not be considered plagiarism, copyright infringement or parasitism? Legally speaking, this depends from domain to domain. But consider for example extracting facts from several biology textbooks, and then delivering those facts to the user in the characteristic ChatGPT tone that is distinguishable from the style of each source textbook. You can then be quite assured that courts will not find that you have infringed on copyright.

slg 1y ago

>societal desire that generating IP continue to be a viable economic work product It is strange that you think the law is settled when I don't think even this "societal desire" is completely settled just yet.

ineedasername 1y ago
Maybe I should clarify: Society, in general, supports the idea that writers, artists, film makers, coders, etc— everyone who creates IP- should have a place in the economy. Basically just that it should be possible to make a living and have a career at it. It can be spun different ways, and those differences are important, but this is the basic thing. This doesn’t seem like a disputable statement to me. For anyone who thinks actors’ likenesses, authors’ words, all of it- that all and everything should be up for grabs once written or put anywhere in public, that is not a widely held opinion. Once that’s established, it all comes down to implementation details.
- ekianjo 1y ago
  > Society, in general, supports the idea that writers, artists, film makers, coders, etc Coders don't get paid every single time their code runs. Why bundle different rights together?
  - aspenmayer 1y ago
    > Coders don't get paid every single time their code runs. They do if they code the API correctly. > Why bundle different rights together? Why are mineral rights sold separately to most land deeds?
    - aleph_minus_one 1y ago
      > Why are mineral rights sold separately to most land deeds? Because the population does not rebel against the politicians that made these laws.
  - ineedasername 1y ago

BobbyTables2 1y ago

I’m curious … So “transformative” is not necessarily “derivative”? Seems to me the training of AI is not radically different than compression algorithms building up a dictionary and compressing data. Yet nobody calls JPEG compression “transformative”. Could one do lossy compression over billions of copyrighted images to “train” a dictionary?

zahlman 1y ago
> I’m curious … So “transformative” is not necessarily “derivative”? (not legal advice) Transformative works are necessarily derivative, but that transformation allows for a legal claim to "fair use" regardless of making a derived work. https://en.wikipedia.org/wiki/Transformative_use https://en.wikipedia.org/wiki/Transformative_use
ipaddr 1y ago
A compression algorithm doesn't transform the data it stores it in a different format. Storing a story in a txt file vs word file doesn't transform the data. An llm is looking at the shape of words and ideas over scale and using that to provide answers.
- const_cast 1y ago
  No a compression algorithm does transform the data, particularly lossy ones. The pixels stored in the output are not in the input, they're new pixels. That's why you can't uncompress a jpeg. Its a new image that just happens to look like the original. But it even might not - some jpegs are so deep fried they become their own form of art. This is very popular in meme culture. The only difference, really, is we know how a JPEG algorithm works. If I wanted to, I could painstakingly make a jpeg by hand. We don't know how LLMs work.