• 0 posts
  • 8 comments
Joined 1 year ago
Cake day: August 26th, 2025
  • Okay, you just seem to understand how copyright works, and I already said that raw LLM output isn’t copyrightable. Prompts needn’t be poetry either; they can be branding design guidelines for example.

    I don’t feel like giving you a lecture that you probably won’t understand either, so I’m done here. Go ask an attorney or read the wikipedia page or copright.gov PDF I linked previously if you care to learn more. The post I linked has a direct quote with some bulletpoints that contradict your understanding. The full document goes into more detail.

  • Nothing of what I said is controversial in the legal world, and it’s all pretty well understood. Yes, your example of a one sentence prompt is likely to not be defensible; but more elaborate prompts likely are, depending on specificity and the merit of individual claims. This is a case by case thing. No court decision will be or even can be universal on something as broad as LLM prompts.

    But regardless, all of this is irrelevant to the original post.

  • What little rep I have here is probably seen as pro-AI; even I’ll say “yeah, no shit.”

    Did anyone expect same companies that stole everyone’s data and flagrantly ignored copyright (among other laws) to build their models would suddenly start caring about the privacy of their users? (Esp. When agentic session prompts/data is some of the most valuable right now)

    No. No one did. ZDR is just compliance LARPing to shut up legal, so you can shift blame by pointing at the inference vendor policy when shit gets exposed, rather your decision to use cloud AI.

    If this is about what I’m thinking it’s about at first glance: this is a Chinese lab on top of all that. Openrouter et al are all insisting the fault is on the backend and not their forwarding, in which case it’s the lab itself, which should be a surprise to no one. Unless this is a completely different incident, in which case it’s also still not a surprise.

  • Architecture is absolutely copyrightable. That’s a creative form of expression. And IP extends beyond mere digital assets.

    As explained in the quote in my link: raw LLM output isn’t copyrightable, as verbatim text, but other aspects of the project may be.

    Also, you’re confusing LLM prompts with generated output; (human-authored) prompts are copyrightable.

    Edit: Also, this project isn’t source-available (I don’t think?). So the license is actually applying to the distributed bundle. But I was speaking in generalities anyways.

  • More or less, you’re correct. It’s a bit nuanced and the lines get blurry. These are things that get argued out in court and are upheld on a case-by-case basis. If I see slop with a GPL license, I’m probably not going to risk violating it on a hope that it isn’t covered…

    https://www.copyright.gov/ai/

    https://en.wikipedia.org/wiki/Artificial_intelligence_and_copyright

    Quote from the former

    The use of AI tools to assist rather than stand in for human creativity does not affect
    the availability of copyright protection for the output.
    • Copyright protects the original expression in a work created by a human author,
    even if the work also includes AI-generated material.
    • Copyright does not extend to purely AI-generated material, or material where there
    is insufficient human control over the expressive elements.
    • Whether human contributions to AI-generated outputs are sufficient to constitute
    authorship must be analyzed on a case-by-case basis.
    • Based on the functioning of current generally available technology, prompts do not
    alone provide sufficient control.
    • Human authors are entitled to copyright in their works of authorship that are
    perceptible in AI-generated outputs, as well as the creative selection, coordination,
    or arrangement of material in the outputs, or creative modifications of the outputs.
    • The case has not been made for additional copyright or sui generis protection for AI-
    generated content

    Repo licenses, for example, apply where ever they may be applicable; they aren’t necessarily a blanket claim that everything in the repo strictly falls under that license (e.g. vendored dependencies may have their own licenses. Modules, components, files, or even individual functions or lines of code may have differing licenses. Its messy when you get pedantic about it)

    So you can slap a license on slop… But whether or not it holds, and to which parts specifically, is variable

  • This. One of the big concerns from AI, especially in the early days, was ‘license laundering’; taking copy-left code (e.g. GPL), pumping it through an LLM, and putting it into a proprietary close-sourced product. Slop can’t be copyrighted, under most definitions; but (where applicable) putting slop under such a license reduces those concerns.