It may be that he could be sued for license infringement for violating the GPL 3 license by feeding code and using improperly licensed code.
Idk how all the lawsuits will fall, but imo, by not disclosing AI use it jeopardizes the license requirements for everyone who ever contributed to the project. Best case the project is essentially public domain for any components edited after this change.
Yeah, I’m interested to see how it turns out. Realistically I don’t think we’ll see models training on GPL code making the model or it’s output “GPL’ed” because (I think, but I could be mistaken) there’s already been a court case about training models on copyrighted content and the court ruled that it was okay. The GPL, while extremely restrictive, is still more permissive than the default “all rights reserved” approach of copyright. That is to say, if courts ruled that copyrighted content in models is fine they’d also rule that copyleft content in models is fine. (Which sucks, and not really something I’m sure I agree with, but I’m also not a lawyer or a judge.)
My understanding is that, regardless of it was AI or not, machine output cannot be copyrighted. I’m not sure where the line is and how much tweaking you’d need to do to for it to suddenly become something you’re protected under copyright. With things like code, as opposed to images, I think we’ll likely see that devs get copyright over it. Because I think most of the time they’re tweaking it some. Generally with image generation I don’t think folks are tweaking the output, unless they themselves are an artist, and for the most part most artists I’ve seen are more opposed to AI than devs. But who knows? It’ll probably take someone copying code that was created by AI and the creator/prompter having to backup that what they did was enough to grant them protection under copyright law. But by that point, I’m really talking out of my depth, this is just a guess.
My most realistic outlook on it is fairly pessimistic. I think model creators will still be able to use copyrighted and copyleft works however they see fit and I think for all practical purposes most folks using generators will likely be tweaking or prompting creatively enough in some way to successfully argue that the result is something they made using the AI as a tool rather than something the computer just generated on its own.
Imo, I’d prefer a “contamination” approach where the strictest licenses in the training data applies to all outputs. I doubt such a rule would get through big business filters but it would maximize the public good and any country that does manage something like that would probably gain the most benefits from these companies.
It may be that he could be sued for license infringement for violating the GPL 3 license by feeding code and using improperly licensed code.
Idk how all the lawsuits will fall, but imo, by not disclosing AI use it jeopardizes the license requirements for everyone who ever contributed to the project. Best case the project is essentially public domain for any components edited after this change.
Yeah, I’m interested to see how it turns out. Realistically I don’t think we’ll see models training on GPL code making the model or it’s output “GPL’ed” because (I think, but I could be mistaken) there’s already been a court case about training models on copyrighted content and the court ruled that it was okay. The GPL, while extremely restrictive, is still more permissive than the default “all rights reserved” approach of copyright. That is to say, if courts ruled that copyrighted content in models is fine they’d also rule that copyleft content in models is fine. (Which sucks, and not really something I’m sure I agree with, but I’m also not a lawyer or a judge.)
My understanding is that, regardless of it was AI or not, machine output cannot be copyrighted. I’m not sure where the line is and how much tweaking you’d need to do to for it to suddenly become something you’re protected under copyright. With things like code, as opposed to images, I think we’ll likely see that devs get copyright over it. Because I think most of the time they’re tweaking it some. Generally with image generation I don’t think folks are tweaking the output, unless they themselves are an artist, and for the most part most artists I’ve seen are more opposed to AI than devs. But who knows? It’ll probably take someone copying code that was created by AI and the creator/prompter having to backup that what they did was enough to grant them protection under copyright law. But by that point, I’m really talking out of my depth, this is just a guess.
My most realistic outlook on it is fairly pessimistic. I think model creators will still be able to use copyrighted and copyleft works however they see fit and I think for all practical purposes most folks using generators will likely be tweaking or prompting creatively enough in some way to successfully argue that the result is something they made using the AI as a tool rather than something the computer just generated on its own.
Imo, I’d prefer a “contamination” approach where the strictest licenses in the training data applies to all outputs. I doubt such a rule would get through big business filters but it would maximize the public good and any country that does manage something like that would probably gain the most benefits from these companies.