4 ms·

i'm sorry, but you have to be extremely naive to think that all these companies actually respect people's work enough to verify that all their training data is

i'm sorry, but you have to be extremely naive to think that all these companies actually respect people's work enough to verify that all their training data is properly licensed, and deliberately hinder themselves and put themselves behind their competitors by avoiding any copyrighted work

GaggiX 2y ago
They clearly do not avoid copyrighted work, where in my reply did you infer the opposite?
- taminka 2y ago
  > where in my reply did you infer the opposite here > There is no theft
  - pixl97 2y ago
    Copyright infringement is not theft.
    - taminka 2y ago
      not legally i guess, but in spirit i'd say it is? also obviously depends on the case, copyright law is often very stupid and broken, like there's no reason why this book written 100 years ago isn't in public domain, but scraping millions of images from artists' websites kinda is
  - GaggiX 2y ago
    Well there is no theft ahah But even if by "theft" you mean "copyright infringement" there is a strong argument for fair use, we humans consume large amounts of copyrighted content and are influenced by it in everything we do, I don't find it unethical or wrong in general, and I don't see how it should be any different for a machine, unless you are against job automation, but in this case you seem to be worried about "theft". Edit: HN thinks that I'm posting too fast so my reply to the comment below here ahah: "Well, then we agree that it wouldn't be a problem if the model didn't output an image from the training dataset, which is extremely rare or essentially impossible with today's dedup steps used during dataset creation."
    - taminka 2y ago
      models will straight up return training data sometimes if you prompt them with like an image file name like IMG_1234.JPG though i'd say that this is not the equivalent of a human glancing at an image and then have the memory of details of that image have some small influence on their way of thinking and imagining and creating, i'd say that's theft

GaggiX 2y ago

They clearly do not avoid copyrighted work, where in my reply did you infer the opposite?

taminka 2y ago
> where in my reply did you infer the opposite here > There is no theft
- pixl97 2y ago
  Copyright infringement is not theft.
  - taminka 2y ago
    not legally i guess, but in spirit i'd say it is? also obviously depends on the case, copyright law is often very stupid and broken, like there's no reason why this book written 100 years ago isn't in public domain, but scraping millions of images from artists' websites kinda is
- GaggiX 2y ago
  Well there is no theft ahah But even if by "theft" you mean "copyright infringement" there is a strong argument for fair use, we humans consume large amounts of copyrighted content and are influenced by it in everything we do, I don't find it unethical or wrong in general, and I don't see how it should be any different for a machine, unless you are against job automation, but in this case you seem to be worried about "theft". Edit: HN thinks that I'm posting too fast so my reply to the comment below here ahah: "Well, then we agree that it wouldn't be a problem if the model didn't output an image from the training dataset, which is extremely rare or essentially impossible with today's dedup steps used during dataset creation."
  - taminka 2y ago
    models will straight up return training data sometimes if you prompt them with like an image file name like IMG_1234.JPG though i'd say that this is not the equivalent of a human glancing at an image and then have the memory of details of that image have some small influence on their way of thinking and imagining and creating, i'd say that's theft