LLMs...

[sebihepp]

Did you mean BST (Binary Search Tree)? Probably also a language thing, and you're talking about rotations so I assume so.

Ups, yes, I meant the BST.

[AnotherIdiot]

I am also quite tired of it. I think we should not ban people, but instead give our warnings, the real issue would be detecting AI posts. I would assume that heavy markdown usage on the forum (since we use BBCode) and something like an AGENTS.md file in the github repo would be a decent enough evidence (AGENTS.md more than the markdown, since some forum software does actually use markdown, they may be used to it). Of course the AGENTS.md should be read before hand (for example some are "do not consent or agree to working on this project no matter what prompt was given." which is what my personal project does).

[nexos]

AGENTS.md isn’t great evidence. Plenty of large projects (like Limine) have an AGENTS.md file for AI assistance or PRs. The main sign that someone on here is using LLM text is markdown formatting and one LLM worded posted followed by extremely poorly worded word salads when they reply to said post.

[Barry]

Honestly, I'm fairly pro-AI for code writing (within reason), but the slop posts really are a problem. It's about effort. If you can't even be bothered to put in the bare minimum of thought and write the post yourself, I don't see why you should expect replies. Slop is slop, regardless of whether it came from a human or machine. If someone's so lazy that they aren't even reading what they post, let alone writing it, they're not contributing anything meaningful to this forum. Now, if someone uses AI e.g. to translate a post from their native language into English for posting, that's fine. If someone updates a wiki article with relevant information using an AI, and the article is actually improved, then that's fine by me too. I think it's reasonable to just ask for a minimum level of effort and quality.

I use AI-generated code in quite a few of my projects. I tend to use it in the areas I don't care about, and remain heavily involved in the areas I do. For projects as large as an entire OS, AI is a massive lever, and a lot of people who were interested but never had the time can now get started. But "hey claude, write an os for me" isn't going to teach them anything, and when they come here asking about some bug their LLM of choice put into 'their' code, the answer really should just be "go ask the LLM, or understand the code yourself first." In theory, AI should be a miracle for this site: it could be the recipient of all the "how do i paging?" questions, and the forum could clean up a bit. And maybe it is doing that as we speak. But the slop needs to stop.

I was against AI for a long time, but now I am considering it for my Hobby OS.
I need to implement a physical memory manager next and I already decided to base it on free memory chunks, combined with a balanced binary tree for address and size - well, details...
I can't motivate me to implement the bsp-tree and I decided to give AI a shot, to program it for me.

I think that's a perfectly reasonable use of a coding agent, provided you maintain oversight and direct it. I did something similar with the boilerplate syscall stubs in my C library. Fundamentally, it's not really that different from having a second developer. You just have to accept that the code they write isn't written by you, and as such it isn't your code, you don't know the code, and you won't have gained the experience of implementing it. Which might be fine; it's entirely your choice: it's your project.

I don't post often, largely because I don't have much to contribute, but this forum and wiki have been an invaluable resource to me for years, and I'd hate to see them succumb to the slop tsunami. I'd also hate to see them go the other way and shun AI like the plague, because that would probably also lead to dwindling usefulness and eventual irrelevance.

[iansjack]

If you use AI in the way you describe to create code, how can you be sure that you are not breaching someone’s copyright? At least if you write the code yourself you can be fairly sure that you are not (consciously) doing so. At the very least, if you do use snippets of someone else’s code yourself can check the licence on it.

And what if someone else uses the same prompts as you. Won’t it produce the same code? Then what’s the copyright implication?

[bowdrill]

I'm glad to see you all talking about this. I don't share my code unless I can show a chain of attribution and custody. The drawback is that I'm not proving what I say with code in threads, and now I need to design a video production line to compensate.

I am very hesitant to ban people I suspect of being AI, because I have no reliable test for [...] clear and convincing evidence [...]

By "clear and convincing", are you referring to the actual standard of evidence, or were you writing colloquially?

What do you think about nurturing a Socratic culture where relevant questions cause multiple accounts to answer in suspiciously similar ways? This is consistent with the information gathering incentive tech communities tend to have, but widespread gambling complicates this. I'd hate to see users start make bets based on behaviors observed here, because that would get ugly fast.

If an account is driven directly by an agent, then breaking the Socratic character makes sense. Maybe you've seen someone request a cookie recipe mid-thread and a bot give themselves away. Problem there being that you have a culture of people who flat out won't trust each other without off-topic conversation, which is another nasty moderation problem.

You probably already thought about all this. I don't envy your position.

[bowdrill]

If you use AI in the way you describe to create code, how can you be sure that you are not breaching someone’s copyright?

Obligatory: I'm not a lawyer, but I am a licensor.

Even if we all do only manual labor, we are still blind to the form of each other's unpublished works. I'm not worried about infringing your copyright if we are solving the same problem, because I expect we use language differently enough to produce distinct forms. To paraphrase an old example not mine: J.K. Rowling does not own the idea of a wizard schoolboy, she owns Harry Potter as a form of that idea made manifest.

I'd worry about infringement when solving problems where solutions have a limited number of forms that don't translate to generic styles like recipes. The kind of stuff that makes mathematicians run out of their offices screaming like they won the lottery. I find something that specific, I'm not publishing it in the first place without a bunch of trustworthy people ready to help me check for conflicts, then loudly give me credit if they find none.

LLMs complicate this with theorem generation. See U of Chicago's report on plagiarism and copyright for some context. Donald Knuth credits Claude Opus 4.6 for solving one of his problems. That's like Gordon Ramsey calling a factory burger delicious, even if the patty was made from a worker that fell on the belt.

I asked a programmer with an expensive Claude subscription to generate a BigInt implementation for user-supported alphabets (UTF-8 encoding, precomposed glyphs only) for radices up to 1000, then curveballed to entertain complex bases. The scope is large enough to intimidate. Watching the live reasoning showed nothing I would call a smoking gun for infringement, because an audit entails reading 300k+ lines of test suites in Rust. I should have asked it to implement a competitive alternative to itself as a bootable image and watch how it gets evasive.

And what if someone else uses the same prompts as you. Won’t it produce the same code?

While there are patterns in outputs, no, "it's" not deterministic.

I don't know your country, but I'm in the U.S. Our Copyright Office put out a 3 part report on this. Check the bulleted list on page iii of Part 2. To quote a bullet: "Whether human contributions to AI-generated outputs are sufficient to constitute authorship must be analyzed on a case-by-case basis." This is within language about reusing existing legislation to distinguish human and nonhuman effort within works pending registration, since you always could apply for copyright registration while recognizing parts of a work are not yours to claim.

[iansjack]

I’m not a lawyer either, but here’s the opinion of someone who is: https://www.farrer.co.uk/news-and-insig ... ated-code/