✨ Nora Tindall  🪐

To specialize the GPT-3 network in our domain we're going to use thousands of movie scripts. We're using movies from Disney, Terminator, Start Wars. A lot of licensed movies (this is already happening, as I've said) After training our network, it generates a movie script (+)

I understand it is already happening and we can't stop it, but this will fundamentally affect how copyright law works, and not by choice. This will be a one-sided push for things like GitHub Copilot by Big Tech until people accept this as not being copyright infringement

This is very similar to the taxi industry being disrupted by Uber years ago: we can agree it is a better model, but traditional taxi drivers who paid big money for their license were sacrificed in the process. Laws were changed by the force of things, not by "choice"

If all of sudden all source code made publicly available can be used to train AI code assistant tools with complete disregard for the license, this will fundamentally change how OSS is developed: the only way to protect your code will be to make it private

I don't completely agree with you on that point. Not all open-source developers care about that. Some developers try to protect their code with a strict license, but some of them don't. As a open-source dev myself, I'm one of those who don't care about it

Okay. But I'm one who does. And I have a right to have my license respected. Yes?

Yes, but right now it isn't clear what is the boundary between what violates and what doesn't your license. If I copy a function from your code, modify it a little and use it. I would be violating your license?

That's literally what copyright law is. We are not going to rederive all of copyright law in a Twitter conversation. Stop trying to argue everything from first principles, our society has thought about this.