3 ms·

> Then took my kids down to the local pool. When I got back, I had a fully functional Haskell library with autogenerated C bindings, FFI to CoreAudio and my jaw

by neilv 2y ago

> Then took my kids down to the local pool. When I got back, I had a fully functional Haskell library with autogenerated C bindings, FFI to CoreAudio and my jaw was on the ground. With a tricky copyright declaration and licensing. A mechanical translation of Rust code under the copyright of ___? Maybe also involving stealing bits of Haskell code without attribution from material copyrights of ___, ___, and ___? Cleanroom engineering is a thing when it comes to "make me one like that, except I have copyright of it". The LLM laundering approach isn't that. Ripping off open source code isn't new. Various companies and individuals would do it. Occasionally they got caught. Microsoft/GitHub had to reassure companies that cared about LLM copyright violations, by indemnifying them. I guess so that open source projects ripped off would have to face the legal resources of Microsoft.

coldpie 2y ago
It turns out copyright doesn't apply if you violate all of it at the same time. Weird trick.
- formerly_proven 2y ago
  Reminds me of the old "if you owe the bank a million dollars, you're in trouble, if you owe them a billion dollars, they're in trouble" bit.
Joker_vD 2y ago
All this AI-generated stuff really should just go into the public domain automatically. That'd immediately not only resolve any possible copyright disputes but also make AI usage much, much less appealing for the large orgs.
- neilv 2y ago
  > All this AI-generated stuff really should just go into the public domain automatically. That'd immediately not only resolve any possible copyright disputes [...] I can't just take something of yours, and make it public domain.
  - Joker_vD 2y ago
    Well, as of right now the AI-generated images are considered to be un-copyrightable by the USA Office of the Copyright, because they were created by "a mere mechanical process that operates randomly". That doesn't mean that they are in public domain, of course — but the law could make it so, why not. I guess I know why not: all of the major copyright holders would probably rather prefer to settle such issues in the courts and/or contracts, to get their share of the rent extraction, and AI companies also don't have much incentive to have AI-generated output be public domain either, so I'd imagine the lawmakers would try to reconcile theirs opinions somehow instead of adopting this point of view of mine (I am, after all, a random person on the Internet and have no lobbying power).
    - layer8 2y ago
      Them being uncopyrightable doesn’t imply that they can’t infringe on others’ copyrights. If you publish some copyright-infringing piece and say “my dog typed it by accident” (LLMs basically have the same status as animals here), then even in the hypothetical case that that’s true it’s still copyright infringement.

by neilv 2y ago

coldpie 2y ago

It turns out copyright doesn't apply if you violate all of it at the same time. Weird trick.

formerly_proven 2y ago
Reminds me of the old "if you owe the bank a million dollars, you're in trouble, if you owe them a billion dollars, they're in trouble" bit.

Joker_vD 2y ago

All this AI-generated stuff really should just go into the public domain automatically. That'd immediately not only resolve any possible copyright disputes but also make AI usage much, much less appealing for the large orgs.

neilv 2y ago
> All this AI-generated stuff really should just go into the public domain automatically. That'd immediately not only resolve any possible copyright disputes [...] I can't just take something of yours, and make it public domain.
- Joker_vD 2y ago
  Well, as of right now the AI-generated images are considered to be un-copyrightable by the USA Office of the Copyright, because they were created by "a mere mechanical process that operates randomly". That doesn't mean that they are in public domain, of course — but the law could make it so, why not. I guess I know why not: all of the major copyright holders would probably rather prefer to settle such issues in the courts and/or contracts, to get their share of the rent extraction, and AI companies also don't have much incentive to have AI-generated output be public domain either, so I'd imagine the lawmakers would try to reconcile theirs opinions somehow instead of adopting this point of view of mine (I am, after all, a random person on the Internet and have no lobbying power).
  - layer8 2y ago
    Them being uncopyrightable doesn’t imply that they can’t infringe on others’ copyrights. If you publish some copyright-infringing piece and say “my dog typed it by accident” (LLMs basically have the same status as animals here), then even in the hypothetical case that that’s true it’s still copyright infringement.