7 ms·

Fair use. The models learn patterns, not exact copies.

by natch 3y ago

amelius 3y ago
Define "not exact". A Xerox copier produces not exact copies in my definition.
- natch 3y ago
  We are talking about AI, not Xerox copiers. AI today doesn’t come close to the word for word, pixel for pixel copy you are talking about.
hakdbha 3y ago
[dead]
stefan_ 3y ago
Fair use is an admission of copyright violation. If this is your starting point, you should probably look for new lawyers. In any case, exact reproduction of the full novels as part of a dataset is of course not covered by fair use and proving that through discovery will be trivial.
- Ukv 3y ago
  > Fair use is an admission of copyright violation. If this is your starting point, you should probably look for new lawyers. It's an admission to copying when there's a prima facie case of copyright infringement, but wouldn't generally be considered actual infringement if the fair use defense is successful. It's perfectly fine to rely on fair use - we'll have to wait and see how it applies to machine learning, but it's not some obscure implausible theory. > In any case, exact reproduction of the full novels as part of a dataset is of course not covered by fair use It definitely can be. Consider in Authors Guild v. Google, where Google copied millions of in-copyright books, stored them internally in full, and made them accessible as snippets: "verbatim intermediate copying has consistently been upheld as fair use if the copy is ‘not reveal[ed] . . . to the public.’"
- Turing_Machine 3y ago
  > Fair use is an admission of copyright violation. No, it is not. It's an admission of copying, but not an admission of copyright violation. Legal copying (which includes fair use) is not a "violation" of any kind. For example, in general it is unlawful to use controlled substances (e.g., morphine). However, if one has a prescription for morphine and uses it in accordance with that prescription (i.e., legally), that isn't an "admission" of "violating" the laws against narcotics use. > If this is your starting point, you should probably look for new lawyers. Someone needs new lawyers, but it's not the OP.
dahart 3y ago
Wasn’t OpenAI just seeking a copyright exception because they admitted they’re sometimes producing exact copies? There are too many counter-examples of AI memorizing and reproducing specific works now to hide behind the ‘models learn patterns’ argument.
- baobabKoodaa 3y ago
  Reproducing exact or near exact copies is a failure mode called overfitting. It's not what these tools are intended for, and when it happens it's typically to the detriment of the users.
  - dahart 3y ago
    It seems like one of Stable Diffusion’s most popular use cases is give me <situation> in the style of <famous_artist>. I mean the whole point of using AI at all is to get something to give you high quality stuff for free, where the high quality has been lifted from a high quality training set. Getting sorta-kinda copies but ideally without breaking laws kinda is what these tools are explicitly intended and designed for.
    - EMIRELADERO 3y ago
      Styles aren't copyrightable.
      - dahart 3y ago
        That’s right, correct. You’re just a lot more likely to infringe on a specific “fixation” of a copyrighted work when the purpose and use centers around getting as close as possible to something that already exists, right?
    - baobabKoodaa 3y ago
      > It seems like one of Stable Diffusion’s most popular use cases is give me <situation> in the style of <famous_artist>. You know what's the most popular <style> for Stable Diffusion? Photorealism. And the second most popular <style>? Anime. At the risk of stating the obvious, I will note that no artist can hold the honor of having invented generic "photography" and generic "anime". But sure, I will give you 2 internet points for accurately spotting one of the extremely rare use cases of Stable Diffusion where an artists' style is copied. I agree with you that is not all kosher. > I mean the whole point of using AI at all is to get something to give you high quality stuff for free, where the high quality has been lifted from a high quality training set. No, it's not. It's a failure mode. It's called "overfitting". We went over this. The vast majority of Stable Diffusion's usage is about producing something novel, and reproducing something that looks like a training image is considered a failure.
BadHumans 3y ago
I don't think you understand fair use.

Fair use. The models learn patterns, not exact copies.

by natch 3y ago

Fair use. The models learn patterns, not exact copies.

amelius 3y ago

Define "not exact". A Xerox copier produces not exact copies in my definition.

natch 3y ago
We are talking about AI, not Xerox copiers. AI today doesn’t come close to the word for word, pixel for pixel copy you are talking about.

hakdbha 3y ago

[dead]

stefan_ 3y ago

Fair use is an admission of copyright violation. If this is your starting point, you should probably look for new lawyers. In any case, exact reproduction of the full novels as part of a dataset is of course not covered by fair use and proving that through discovery will be trivial.

Ukv 3y ago
> Fair use is an admission of copyright violation. If this is your starting point, you should probably look for new lawyers. It's an admission to copying when there's a prima facie case of copyright infringement, but wouldn't generally be considered actual infringement if the fair use defense is successful. It's perfectly fine to rely on fair use - we'll have to wait and see how it applies to machine learning, but it's not some obscure implausible theory. > In any case, exact reproduction of the full novels as part of a dataset is of course not covered by fair use It definitely can be. Consider in Authors Guild v. Google, where Google copied millions of in-copyright books, stored them internally in full, and made them accessible as snippets: "verbatim intermediate copying has consistently been upheld as fair use if the copy is ‘not reveal[ed] . . . to the public.’"
Turing_Machine 3y ago
> Fair use is an admission of copyright violation. No, it is not. It's an admission of copying, but not an admission of copyright violation. Legal copying (which includes fair use) is not a "violation" of any kind. For example, in general it is unlawful to use controlled substances (e.g., morphine). However, if one has a prescription for morphine and uses it in accordance with that prescription (i.e., legally), that isn't an "admission" of "violating" the laws against narcotics use. > If this is your starting point, you should probably look for new lawyers. Someone needs new lawyers, but it's not the OP.

dahart 3y ago

Wasn’t OpenAI just seeking a copyright exception because they admitted they’re sometimes producing exact copies? There are too many counter-examples of AI memorizing and reproducing specific works now to hide behind the ‘models learn patterns’ argument.

baobabKoodaa 3y ago
Reproducing exact or near exact copies is a failure mode called overfitting. It's not what these tools are intended for, and when it happens it's typically to the detriment of the users.
- dahart 3y ago
  It seems like one of Stable Diffusion’s most popular use cases is give me <situation> in the style of <famous_artist>. I mean the whole point of using AI at all is to get something to give you high quality stuff for free, where the high quality has been lifted from a high quality training set. Getting sorta-kinda copies but ideally without breaking laws kinda is what these tools are explicitly intended and designed for.
  - EMIRELADERO 3y ago
    Styles aren't copyrightable.
    - dahart 3y ago
      That’s right, correct. You’re just a lot more likely to infringe on a specific “fixation” of a copyrighted work when the purpose and use centers around getting as close as possible to something that already exists, right?
  - baobabKoodaa 3y ago
    > It seems like one of Stable Diffusion’s most popular use cases is give me <situation> in the style of <famous_artist>. You know what's the most popular <style> for Stable Diffusion? Photorealism. And the second most popular <style>? Anime. At the risk of stating the obvious, I will note that no artist can hold the honor of having invented generic "photography" and generic "anime". But sure, I will give you 2 internet points for accurately spotting one of the extremely rare use cases of Stable Diffusion where an artists' style is copied. I agree with you that is not all kosher. > I mean the whole point of using AI at all is to get something to give you high quality stuff for free, where the high quality has been lifted from a high quality training set. No, it's not. It's a failure mode. It's called "overfitting". We went over this. The vast majority of Stable Diffusion's usage is about producing something novel, and reproducing something that looks like a training image is considered a failure.

BadHumans 3y ago

I don't think you understand fair use.