4 ms·

I'm also worried that courts will decide that model weights are copyrightable and the open source free-for-all will be over.

nwienert 4y ago
If models can’t prove they are fully free of copyrighted data I don’t think they’ll have a leg to stand on there.
- yuuuuyu 4y ago
  If you are holding copyright to something, it will be on you to prove it's in there.
  - manojlds 4y ago
    I am obviously clueless, but if it's a case, can't one demand what the training data is?
    - arthurcolle 4y ago
      Based on what case law? In what jurisdiction?
      - nwienert 4y ago
        It's very easy to show the models generating near-identical data to copyrighted data, which is enough to get courts to force them to allow discovery.
        yuuuuyu 4y ago
        This not have happened all over the place yet is evidence against this being as easy as you make it sound.
- nl 4y ago
  This is clearly not a given. Search engines are good decided case law in the opposite direction.
  - Nullabillity 4y ago
    Search engines aren't a replacement for the original data, they're a way to direct traffic towards it.
    - nl 4y ago
      The business model doesn't really change the IP considerations though. (Additionally, newer LLMs like Perplexity.AI's correctly cite content sources, so that is even more similar to search engines)
      - nwienert 4y ago
        These models will readily generate near identical outputs to copyrighted data, at length. This is not comparable to search.
        nl 4y ago
        I'd invite you to read "Foundation Models and Fair Use"[1] which is a paper written as a collaboration between Standford's law school and computer science department. It talks at length about this specific problem and migration techniques for it: Existing foundation models are trained on copyrighted material. Deploying these models can pose both legal and ethical risks when data creators fail to receive appropriate attribution or compensation. In the United States and several other countries, copyrighted content may be used to build foundation models without incurring liability due to the fair use doctrine. However, there is a caveat: If the model produces output that is similar to copyrighted data, particularly in scenarios that affect the market of that data, fair use may no longer apply to the output of the model. In this work, we emphasize that fair use is not guaranteed, and additional work may be necessary to keep model development and deployment squarely in the realm of fair use. First, we survey the potential risks of developing and deploying foundation models based on copyrighted content. We review relevant U.S. case law, drawing parallels to existing and potential applications for generating text, source code, and visual art. Experiments confirm that popular foundation models can generate content considerably similar to copyrighted material. Second, we discuss technical mitigations that can help foundation models stay in line with fair use. We argue that more research is needed to align mitigation strategies with the current state of the law. [1] https://arxiv.org/abs/2303.15715 https://arxiv.org/abs/2303.15715

nwienert 4y ago

If models can’t prove they are fully free of copyrighted data I don’t think they’ll have a leg to stand on there.

yuuuuyu 4y ago
If you are holding copyright to something, it will be on you to prove it's in there.
- manojlds 4y ago
  I am obviously clueless, but if it's a case, can't one demand what the training data is?
  - arthurcolle 4y ago
    Based on what case law? In what jurisdiction?
    - nwienert 4y ago
      It's very easy to show the models generating near-identical data to copyrighted data, which is enough to get courts to force them to allow discovery.
      - yuuuuyu 4y ago
        This not have happened all over the place yet is evidence against this being as easy as you make it sound.
nl 4y ago
This is clearly not a given. Search engines are good decided case law in the opposite direction.
- Nullabillity 4y ago
  Search engines aren't a replacement for the original data, they're a way to direct traffic towards it.
  - nl 4y ago
    The business model doesn't really change the IP considerations though. (Additionally, newer LLMs like Perplexity.AI's correctly cite content sources, so that is even more similar to search engines)
    - nwienert 4y ago
      These models will readily generate near identical outputs to copyrighted data, at length. This is not comparable to search.
      - nl 4y ago
        I'd invite you to read "Foundation Models and Fair Use"[1] which is a paper written as a collaboration between Standford's law school and computer science department. It talks at length about this specific problem and migration techniques for it: Existing foundation models are trained on copyrighted material. Deploying these models can pose both legal and ethical risks when data creators fail to receive appropriate attribution or compensation. In the United States and several other countries, copyrighted content may be used to build foundation models without incurring liability due to the fair use doctrine. However, there is a caveat: If the model produces output that is similar to copyrighted data, particularly in scenarios that affect the market of that data, fair use may no longer apply to the output of the model. In this work, we emphasize that fair use is not guaranteed, and additional work may be necessary to keep model development and deployment squarely in the realm of fair use. First, we survey the potential risks of developing and deploying foundation models based on copyrighted content. We review relevant U.S. case law, drawing parallels to existing and potential applications for generating text, source code, and visual art. Experiments confirm that popular foundation models can generate content considerably similar to copyrighted material. Second, we discuss technical mitigations that can help foundation models stay in line with fair use. We argue that more research is needed to align mitigation strategies with the current state of the law. [1] https://arxiv.org/abs/2303.15715 https://arxiv.org/abs/2303.15715