OpenAI kann das Urheberrechtsverletzungsproblem von Sora nicht beheben, da es mit gestohlenen Inhalten erstellt wurde
https://www.404media.co/openai-cant-fix-soras-copyright-infringement-problem-because-it-was-built-with-stolen-content/
OpenAI kann das Urheberrechtsverletzungsproblem von Sora nicht beheben, da es mit gestohlenen Inhalten erstellt wurde
https://www.404media.co/openai-cant-fix-soras-copyright-infringement-problem-because-it-was-built-with-stolen-content/
10 Kommentare
Surprised Pikachu face.
That’s why it needs to crash and burn.
That’s pretty much every „unicorn“ tech company. Ignore the rules and deal with it later.
Wow, that’s a messy situation. Feels like no matter how smart the AI is, if it’s trained on stuff it shouldn’t have, there’s always going to be legal headaches.
This surprises literally nobody who understands how AI works
It’s not a „problem“, it’s literally the whole product.
That would be like saying that Pirate Bay has a piracy problem.
The original vibe coders can’t fix the original vibed code.
The problem isn’t only in training, it’s that the data of what it’s trained on is encoded and stored, then access to it is sold. It’s a vending machine for theft, put in the right prompts, get whatever stolen goods you want.
Pretty much.
But you don’t get to bypass copyright protections by making a program that spits outs copyrighted content. So it’s only a matter of time before the these cases wind their way through the courts and the companies either settle or find themselves functionally dead.
Therefore, it should not exist.
If I have to steal to make money, that makes me a thief.
Why not simply block the prompts for copyrighted styles and characters like they do with nsfw? It’s a problem they are choosing purposely not to address. Because it can be done.
I think a judge ruled training it was ok, because it would be transformative enough, the problem is they are allowing 1:1 copies of those characters to be generated which rules against that defense. It really isn’t that hard to fix.