I made a Vibe Coded Operating System

[justreallyjustreally]

Hello everyone,

I have been working on a project called JNU for a couple of days; right now, it's not perfect or actually usable yet. It primarily uses GCC for compiling and NASM for compiling assembly; it is an Operating System. I vibe-coded from scratch . It can also run Musl as well as ELF64 programs, albeit with limited ELF support.

The reason I want to post this here is that I want feedback and at least some recognition in this project. Primarily, it only supports x86_64, but I am thinking about adding ARM64 support sometime in the future. It also uses the Limine Bootloader. JNU is not production-ready or mature in any sense.

It is still Experimental and Vibe-coded, meaning that it will have some bugs and hallucinations here and there.

Also I put actual effort into this, it's good enough but I will keep refining it, I did all of this with just free AI credits, no funding, nothing.

https://github.com/atandant/JNU

[savoita]

What do you mean when you say that you "put actual effort into this"? Is the effort here supposed to be the prompting you did to the LLM?

[thewrongchristian]

The source file contain text like:

Code: Select all

 * Copyright (c) 2026 The JNU Authors.
 * SPDX-License-Identifier: GPL-2.0-only

Who do you think the JNU authors are?

Part of the problem of vibe coded works is who is the copyright holder, if anyone.

Copyright requires a human to create the works. As a vibe coded OS is created by an AI, and while you may hold the copyright on the prompt, the actual code is not copyright in any meaningful legal way.

Not that I know whet status it is. But I'd be interested in comparing the contents of the Minix FS driver to other human written Minix FS implementations.

Note, this is not legal advice! Consult with a qualified lawyer for actual legal advice.

[nullplan]

Copyright requires a human to create the works. As a vibe coded OS is created by an AI, and while you may hold the copyright on the prompt, the actual code is not copyright in any meaningful legal way.

That is a very restrictive and absurd view of copyright. If that view held, then Pixar would not own the copyright of any of their movies, since all of them are ultimately program output. And nor would Square Enix hold the copyright for the Final Fantasy VI soundtrack, because the music is produced by a computer (several pieces in it, particularly Battle on Big Bridge, are too difficult for humans to perform, but the SNES audio processor had no issue with it). They'd only hold the copyright in the program.

An LLM is just a program that produces output, and if done under the direction of a human, the product must be copyrightable. Else see above. No, the real copyright problem with LLMs is that they have ingested a lot of copyrighted material for their training, and so the output may be derivative of (many) copyrighted works. That would require a license to create, but you don't know who to even contact for one, since LLMs cannot tell you what sources they used for the data they gave you.

But at least in the EU, this is a solved problem: https://en.wikipedia.org/wiki/Directive ... gle_Market. There is a directive that requires an explicit copyright exception for text and data mining, and it seems to include LLMs.

[Octocontrabass]

As a vibe coded OS is created by an AI, and while you may hold the copyright on the prompt, the actual code is not copyright in any meaningful legal way.

It depends on whose laws you're following. LLM output is not eligible for copyright in the United States, but it's not so clear everywhere.

An LLM is just a program that produces output, and if done under the direction of a human, the product must be copyrightable.

It's the "under the direction of a human" part that's the problem. The LLM's random number generator makes most of the creative decisions, not the human.

[thewrongchristian]

Copyright requires a human to create the works. As a vibe coded OS is created by an AI, and while you may hold the copyright on the prompt, the actual code is not copyright in any meaningful legal way.

That is a very restrictive and absurd view of copyright.

Well, I did disclaim my response that this is not legal advice...

If that view held, then Pixar would not own the copyright of any of their movies, since all of them are ultimately program output. And nor would Square Enix hold the copyright for the Final Fantasy VI soundtrack, because the music is produced by a computer (several pieces in it, particularly Battle on Big Bridge, are too difficult for humans to perform, but the SNES audio processor had no issue with it). They'd only hold the copyright in the program.

An LLM is just a program that produces output, and if done under the direction of a human, the product must be copyrightable. Else see above. No, the real copyright problem with LLMs is that they have ingested a lot of copyrighted material for their training, and so the output may be derivative of (many) copyrighted works. That would require a license to create, but you don't know who to even contact for one, since LLMs cannot tell you what sources they used for the data they gave you.

The big difference between your examples and an LLM producing an OS is that all the creativity is provided by the humans as the input, that is then transformed deterministically by the tools into the final output. That transformation adds no creativity, and is thus not copyrightable (in my not legal opinion.)

LLMs are anything but deterministic.

But at least in the EU, this is a solved problem: https://en.wikipedia.org/wiki/Directive ... gle_Market. There is a directive that requires an explicit copyright exception for text and data mining, and it seems to include LLMs.

The UK (where I am) seems to allow AI output to be copyrighted, but it seems it is an outlier, and my limited research is not producing consistent results. But this is a rabbit hole I don't fancy going down at the moment, so out of an abundance of caution, I'll personally avoid using AI generated output.

So I'll amend my original opinion:

Copyright tranditionally requires a human to create the works. As a vibe coded OS is created by an AI, and while you may hold the copyright on the prompt, the actual code may not be copyright in any meaningful legal way, depending on territory.

And as before, this does not constitute legal advice, I am not qualified to give that.

[nullplan]

Today I was made aware of a copyright case at the Munich district court. In short: With the right prompts it is possible to get ChatGPT to produce copies of the material it ingested. And the result is so similar to the real thing that, if a human produced the same thing, they'd be guilty of copyright and trademark infringement.

Implication is of course the thing with the derivative works license I mentioned in my first message on this thread. So if this case goes against OpenAI, we have a precedent for copyrights of the original author attaching to AI output.

[DeepSpaceWhine]

Serious question: what's the point of vibe-coding anything? The point of making software in your free time for no money is either to have that software to make use of it, or to learn how that type of software works, or for the joy of programming.

If you just wanted an OS to use, there are plenty that you can just download, and they'll be better than the one your LLM synthesized for you since the LLM took the code from them in the first place.

If you wanted to learn how to write an OS then the best way to learn is by doing, and by definition vibe-coding isn't programming so there's no joy-of-programming to be found.

So why do it?

[hadrian]

And to continue on that, OP says:

"and at least some recognition in this project"

Why would anyone give you recognition for putting some prompts into an LLM. A lot of people on this forum put in hard work to figure out how to get stuff done, and your LLM has taken all that code (as a lot of it was part of the training data) and recombined it based on a bunch of prompts.

[thewrongchristian]

Serious question: what's the point of vibe-coding anything? The point of making software in your free time for no money is either to have that software to make use of it, or to learn how that type of software works, or for the joy of programming.

If you just wanted an OS to use, there are plenty that you can just download, and they'll be better than the one your LLM synthesized for you since the LLM took the code from them in the first place.

If you wanted to learn how to write an OS then the best way to learn is by doing, and by definition vibe-coding isn't programming so there's no joy-of-programming to be found.

So why do it?

A topic discussed by the latest Two's Complement (Matt Godbolt and Ben Rady) podcast:

https://www.twoscomplement.org/#podcast ... -still-fun