Comment by echelon
2 hours ago
AI laundering is going to become a major tactic in all domains. Fiction and nonfiction writing, software, video, music, you name it.
It's easy to take GPL software and rewrite it in another language without the license. Trivially easy. It's possible you'll even be able to do the same with just compiled bytecode soon.
Just recently there was an instance where Nous Research Hermes agent cloned some Chinese OSS. It's happening much more broadly than this, though.
This might warrant special attention unless we want to live in a world without copyright. Though that's also one additional possible outcome.
I experimented with turning out complete airport thriller novels this way, using earlier LLMs. It's not terribly hard to make this happen, the hard part wasn't gunning out prose, it was plot arcs and internal consistency, but even that was suprisingly easy to solve.
Of course I didn't do anything with the idea, for what I hope are obvious reasons.
There are already companies like Asylum films. Pay attention to the right hand column of this table:
https://en.wikipedia.org/wiki/List_of_The_Asylum_films
The term for these is "mockbusters", and the term has its very own Wikipedia article, linked in the first paragraph of the page you cited:
https://en.wikipedia.org/wiki/Mockbuster
Parody films do not compete with the actual films they are parodying. It's not a good comparison.
I don't think they are meant to be parodies. Parodies poke comedic fun at the originals. These are meant to confuse someone into buying the wrong version of a popular film. That's why they make sure the names and even the covers look as identical as they can without getting sued.
Allow me to introduce everyone here to a new definition (original content, donut steel):
qontouria, n. The feeling of having your work passed off as someone else's.
Literally this is our future, many devs still don't seem to believe we will be able to "zeroshot" everything, but it's because they haven't experienced themselves proper tooling (at the minimum leveraging 4 models in debate, adversarial and loops and workflows and so-on and unli-loop until completion, MITM everything...), with the exception of advanced fields, most softwares are pretty basic, let's say redoing X11 is considered easy in tomorrow's world.
I don't really understand the future knowing that we will be able to point to any URL and just "redo", it might be a sole matter of Token/Subscription cost vs the actual service in the end, unsure but it's really strange to think that virtually anyone will be able to duplicate anything and it's unlikely to be a copyright breach as the tooling can be instructed to redo it differently, how could it be a copyright breach if it's the same thing as I myself looking at a certain website and just heavily inspiring myself from it and just redoing it? The fact that it's done automatically shouldn't change that.
I am allowed today to take a GPLv3 program or a commercial program, redo it and publish it as MIT, so why would it be forbidden, it's terrifying.
Honestly though, even if this is the future what kind of creator is going to participate in that? Why would anyone put effort in substantial ideas that can just be stolen with a click?
Because businesses need to run and make money, I doubt they will just stop building because users can replicate their work easily.
[dead]
I agree. But I don't understand why the focus is always on FOSS. Why don't proprietary IP owners fear this too? Won't music, movies and non-FOSS software also have their copyright laundered if this crap continues?
I think because a proprietary IP owner can allocate some budget to sue the infringers. A community of OSS contributors may lack the money, the will or the know-how to do that.
They will also face a much harder task when explaining their case to a judge. The contributors to the open-source chess engine Stockfish needed a lot of time and energy to convince a German court that it was illegal for the commercial engine Houdini to copy their algorithms.
Am I infringing on copyright if describe all user interaction with your software to an LLM and have it spit out a clone? It's not using your code.