3 ms·

I would agree that these tools are almost inevitably going to fail. If anything, they'll only improve the models' abilities to handle edge cases accurately. Ho

I would agree that these tools are almost inevitably going to fail. If anything, they'll only improve the models' abilities to handle edge cases accurately. However, I'm surprised at how little empathy is being displayed towards artists by so many here. Having a model train itself on your work is not the same as having another human being inspired by or even copying your work, despite the fact that both involve some sort of learning. I am not an artist but I build all my solo programming projects "in the open". My programs are mostly niche stats models, and frequently people message me with questions and ask for help integrating my work into their software. I'm happy to do so. I don't care if they credit me in their final product, but knowing that humans went through my work and took time to appreciate it is a big motivator for me. I would imagine that many artists feel similarly. On the other hand, knowing that OpenAI, Github/Copilot, etc, train models on my work and turn it into some pay-to-play API, without a human ever seeing my work during the whole process is a nasty feeling. At that point, I've just been turned into faceless cog to generate profits for big tech shareholders. Luckily, these are just side projects for me and I can just make these repos private, but of course artists are forced to "build in the open" by the very nature of their work.

khimaros 4y ago
my personal take is that it is okay (for code and other copyrightable works), and possibly for the greater common good if, and only if, the resulting models are permissively licensed.
- alwayslikethis 4y ago
  Yeah, I think it would make sense to have a special condition like the following: You get to train models on public data under fair use if and only if you release the resulting models (it is not clear whether the models themselves are copyrightable at this point) to the public and do not claim copyright on them.
easyThrowaway 4y ago
Because the main objective of commercial AI generative projects is the "Uberization" of the Arts. Massively dumping on prices, becoming gatekeepers on the creation process, ultimately forcing themselves between clients and the artists, who then become disposable "content creators" for the training model.

khimaros 4y ago

my personal take is that it is okay (for code and other copyrightable works), and possibly for the greater common good if, and only if, the resulting models are permissively licensed.

alwayslikethis 4y ago
Yeah, I think it would make sense to have a special condition like the following: You get to train models on public data under fair use if and only if you release the resulting models (it is not clear whether the models themselves are copyrightable at this point) to the public and do not claim copyright on them.

easyThrowaway 4y ago

Because the main objective of commercial AI generative projects is the "Uberization" of the Arts. Massively dumping on prices, becoming gatekeepers on the creation process, ultimately forcing themselves between clients and the artists, who then become disposable "content creators" for the training model.