Embed this noticeAlex Gleason (alex@gleasonator.com)'s status on Monday, 17-Jul-2023 02:32:37 JST
Alex GleasonChatGPT, Dall-E 2, Midjourney - all proprietary software, BUT all copyright for anything it produces is assigned to you. This is a superpower, and probably the best opportunity Free Culture has ever had. Generate a bunch of useful graphics and CC license it, or public domain. There is a huge freedom prospect even though the code is proprietary.
@joshix In my experience none of these AI services infringe on copyright. They does not produce derivative works, they produce original works inspired by the vast library of training data.
They DO potentially infringe trademarks (depending on how you use the resulting images, of course). Like, you can get an AI to draw you a picture of Mario. But it would not be based on any previously existing Mario picture, it would be an all new image of Mario. This is similar to fan art.
@alex nope, that's not how it works. The models get trained with data (images + text). Then when you feed it data it uses a compressed form of the input data and the prompt to generate an image. There is nothing intelligent about that process. There is nothing really new created.
Of course the images have probably never existed before. But you didn't produce something new if you just merge existing images.
@joshix For the prompt "edible diamond", you can't tell me this isn't creative. I'm not sure whether the task of being creative just isn't as special or hard as we thought, or if the AI is just advanced enough to be creative. But the outcome from AI is at least equal to what a creative human can do, and far more consistent.
@alex >all copyright for anything it produces is assigned to you. The whole idea is the copyright can only be applied to creative works authored by a human.
Either these automated plagiarism tools infringe copyright, or their output doesn't qualify for copyright.
>In my experience none of these AI services infringe on copyright Do you have a source for that claim better than "just believe me bro"?
>They do not produce derivative works, they produce original works inspired by the vast library of training data. Computers are meant to be deterministic and therefore cannot produce anything original - all they can do is calculations, although those calculations can be used to copy data and do transformations on it.
What you're saying is that you have access to some software that you can put non-random numbers into and get truly random numbers out - that's preposterous.
>Like, you can get an AI to draw you a picture of Mario. But it would not be based on any previously existing Mario picture How on earth is a computer meant to output a picture of mario without basing it off pictures on mario inputted into it?
>it would be an all new image of Mario. This is similar to fan art. The thing is, fan works have been found to be derivative works of the original works many times in court.
In countries with fair use, provided you meet the fair use requirements, fan art of mario is permitted, otherwise it is forbidden (although if you draw an original character that looks somewhat like mario, but clearly isn't, that may not be a derivative work).