[jarsin]

This is old news. The copyright office already ruled that AI generative outputs are not copyrightable in January [1].

I think many have not understood the implications of the CO ruling. This means anything you build with llms you don't own. Your company doesn't own. If your using copilot and you have a copyright notice at the top of your source file if that ever goes to court you will learn that copyright is not valid. You cant even put an open source license on the output, like the GPL, because...drumroll...you don't own the copyright.

[1] https://www.copyright.gov/ai/Copyright-and-Artificial-Intell...

[protocolture]

It doesnt say that, it says that anything thats solely produced by simply prompting is not owned. I have seen very few works that want copyright and are solely prompts.

From your own link:

"“To be sure,” the Court further explained, “the requisite level of creativity is extremely low; even a slight amount will suffice."

"The Office agrees that there is an important distinction between using AI as a tool to assist in the creation of works and using AI as a stand-in for human creativity. "

"The Office concludes that, given current generally available technology, prompts alone do not provide sufficient human control to make users of an AI system the authors of the output. "

Where the US ruling differs from others:

"Repeatedly revising prompts does not change this analysis or provide a sufficient basis for claiming copyright in the output."

Where China has had 2 cases where it supported multiple prompt changes + watermark

Also they dont rule out a change:

"There may come a time when prompts can sufficiently control expressive elements in AI-generated outputs to reflect human authorship. If further advances in technology provide users with increased control over those expressive elements, a different conclusion may be called for"

^ I would (and have) suggested that the above would likely cover the masking tools available in most image generators.

Its certainly not a case that "AI generative outputs are not copyrightable".

[Kye]

It sounds like even something minimal like choosing, applying, and adjusting one of the nodes in something like ComfyUI would be sufficient.

[protocolture]

Yes. And anyone who has stepped outside of the chat ecosystem and used something like NovelAI or Sudowrite will be familiar with the co-editing approach those tools use which is easily accounted for with the above.

[jarsin]

You left out the big "but". But if ai gets more optimized and automated our current conclusion will be more bolstered.

[protocolture]

Not really, thats covered well by:

"There may come a time when prompts can sufficiently control expressive elements in AI-generated outputs to reflect human authorship. If further advances in technology provide users with increased control over those expressive elements, a different conclusion may be called for"

Because any "advancement" in this space is predicated on getting tighter control over the requested outcome.

You can already script a local image generator to come up with random images based on text searches or LLM output. Thats already not copyrightable anywhere.

[jarsin]

The "but" is literally in response to what you quoted.

For example if I code an entire application in c by myself without ai then told ai to redo the whole thing in rust I would retain copyright.

If you just prompt the same application from scratch and accept by in large the outputs. No copyrighht. This is how the vast majority are using it to create new systems not using it as a tool to enhance majority human generated code or images or books etc.

The more it creates from pure prompts the lesser chance you have to claim copyright.

[protocolture]

>This is how the vast majority are using it

[Citation Required]

Largely covered by the other quotes. I think it would be quite difficult to create a product worth protecting using "prompts alone".

No debugging? No editing? Who put the graphics on it? Who built the database and schema?

The co-author/co-editing approach is already blessed in the document linked earlier. Code is already subject to some of the best co-editing tools in the ecosystem. Even if someone manages to avoid co-editing tools, launch a product having used "prompts alone" and monetise it, how are you going to prove that they didn't take the co-editing approach to development? And how are you planning to challenge their claimed copyright? Why would you challenge their claimed copyright instead of just generating it yourself?

I could conceive of some kind of anti copyleft organization that dedicates itself to challenging every unskilled software development firm, using the discovery phase to pull records of what tools were used. But who would fund such a witch-hunt?

Or maybe every time some firm tries to assert their copyright, we will see lawyers hit back with "Prove you coded this and didnt generate it wholecloth via LLM" clogging up the courts for decades.

>The more it creates from pure prompts the lesser chance you have to claim copyright.

Yeah but unlike image generators and media articles its going to be a lot tougher to prove.

NovelAI has a feature where it does text highlighting based on:

"User wrote this" "User edited this" "Generated"

It sets this on a per sentence basis.

I have wondered for a long time whether this will become mandatory in some jurisdictions. But even then, if you copy the text, and paste it in a new window, bam its all considered user generated again.