Post
According to @pluralistic, copyright offices around the world have published clarifications that media created by generative models is not subject to copyright law. Not text, not images, not audio or video, and not code.
So every generative model that outputs media ought to add a public domain notice (eg CC0) to anything it produces. Either visibly within the media, or in metadata.
@strypey That may not hold up. Yes, you need a human creator to secure copyright, as evidenced by the "monkey selfie" debacle. However, LLMs sort of extrude "content slurry," so their output is *also* not eligible for copyright because it's just other stuff, analogous to photographing a page of a novel.
You're on the hook if the LLM violated somebody's copyright, which isn't a great match for public licenses.
@jcolag
> You're on the hook if the LLM violated somebody's copyright
IANAL, but I've seen no strong argument for why copyright enforcement ought to apply to the creation of generative models. See Cory Doctorow's piece on the pros and cons of policing web scraping;
https://pluralistic.net/2023/09/25/deep-scrape/#steering-with-the-windshield-wipers
> which isn't a great match for public licenses
Which is why I proposed using a public domain notice. They're not licenses, they just disclaim any copyright restriction on the new work itself.
@strypey Yes, I read it when it came out. It's (probably) Fair Use to train an AI. But it's almost never Fair Use, though, to use elements of works to create a replacement for those works for non-personal use.
It's fine for you to take that risk. But the companies advising you to risk it opens them up to inciting "copyright laundering."
(And that ignores the "if a person didn't find it worth making, it's not worth taking seriously" holy wars...)
@jcolag
> Yes, I read it when it came out. It's (probably) Fair Use to train an AI
> But it's almost never Fair Use, though, to use elements of works to create a replacement for those works for non-personal use
Pick one. If it's Fair Use to create a generative model, surely it's also Fair Use to use whatever it generates? What legal basis would there be for a copyright claim on that output? Especially if the output itself is not subject to copyright restrictions.
@strypey To be blunt, if you think that Fair Use has nothing to do with context, then we're not having a serious conversation and I'm wasting my time.
@jcolag
> if you think that Fair Use has nothing to do with context, then we're not having a serious conversation
Of course it does. What I'm asking is why you think it can be Fair Use to develop a model but not to use its output. What's the difference in your mind between these two contexts, as it pertains to copyright.
@strypey @pluralistic whaaaaa! Mind blown, is this true? Attribution please.
@gwilymgj
> Attribution please
"... the outputs of generative AI can be protected by copyright only where a human author has determined sufficient expressive elements. This can include situations where a human-authored work is perceptible in an AI output, or a human makes creative arrangements or modifications of the output, but not the mere provision of prompts."
#USCopyrightOffice, Jan 2025
A space for Bonfire maintainers and contributors to communicate
