How should raylib as a project handle AI-generated contributions?

[ch0ccyra1n]

Hi! I've been enjoying raylib for some time and have used it in some unreleased projects (I'll finish them some day I swear!) in the past. With the usage of AI-generated contributions in at least some open source game development tools, I figured I'd check to see what raylib's policy is, and based on CONTRIBUTING.md there doesn't seem to be one at the moment. Thus, I figured I'd ask the community what we think. My personal views aside, having any policy at all is certainly going to be better than having no clear policy.

Sorry for the possible flamewar this might start, but it is important.

EDIT: For an example of a simple disclosure policy, the Linux kernel currently has such a policy.

[raysan5]

@ch0ccyra1n No policy, AI PRs will be reviewed as any other PR. raylib is a C project, changes require hard testing, it's a domain where AI-developers rarely try to contribute. In any case, PRs are reviewed by me, as any other PR.

[RobinsAviary]

Thank you for always being open with the community about things like this and making it a poll. I personally don't see a place for AI in stuff like code, where a lot of times, especially in open source, people do it as a labor of love. It also, in my opinion, just doesn't make for good programming, as I find the final quality of a lot of scripts to be questionable, and I've heard that frequent use makes one worse at actually being able to understand and edit them after the fact.

[ChristosMaragkos]

I think requiring the common courtesy of disclosure of AI use is the perfect middle ground, personally.

[gamblingmaster69]

I don't think 100% AI generated code should be allowed, Maybe for repetition or some suggestions for the author but it should at least reviewed by author and include a full human written explanation

[bauripalash]

Mentioning usage of AI should be hard requirement.

[korzewarrior]

The people reading these comments 5, 10, 15, 20+ years from now are laughing at us for wasting our time arguing about such inane topics.

[Tertle950]

Are we surviving that long?

[bauripalash]

No

[jsuarez5341]

Ban people who PR low-effort trash, hand-coded or AI. Leniency towards beginners who are producing low quality work but putting in effort and trying to learn

[rutexd]

AI can be good tool in a complex algorithms. But it can become quite bad at the same time, even if best model is used.

I think best way is to: 1. Require AI usage announcements in pr. 2. Strict requirement for a medium - big changes for a quality check on side of a contributor. Not just tests, but the codebase quality and it's alignment with the project and human readability / understanding too.

Low effort and Ai misuse should be closed without time waste on the analyze and discussing.

[helioscout]

Developers should write code, AI should not write code.

[Tertle950]

Sorry for the possible flamewar this might start, but it is important.

Unfortunately the flamewar is made possible by how important it is.

[sarbesh0x01]

The thing about AI writtten code is, people will write explanatation with AI too and try to pass it as human written, it should be no AI or disclosure of AI with later its matter of will the code meet the standard and do the contributer understand part of the code they sent the PR about

[HumanBodyMacAndCheese]

Probably not going to end well for me, but I believe that AI-generated contributions should be mostly or fully rewritten by a human. During training, an AI model sources its code from us without (presumably) any credit.

[Ternarian]

The scope of raylib thus far is focused on being reliable fundamental building blocks for graphics programming. It's deliberate, thoughtful, and organized in an intuitive but specific way. AI could easily write functioning code that muddies the waters as far as that organization and stylistic consistency goes. I've had to look at raylib source a few times, and that consistency really helped me to understand how things worked and to find the answers I needed.

I think this is one of the triumphs of raylib as a lower-level tool that is very beginner friendly (but that is capable of doing just about anything). Keep that strength. No AI!

[thehuglet]

If the output of both AI or 100% human code is the same AND the dev can explain exactly what's going on, to me that's fair game.

It's hard to know for sure whether someone is lying about AI usage, but it's fairly easy to poke holes in someone's explanation when they don't understand the code they generated using AI.

I'd lean on a "fair game until there are signs of AI slop that the dev can't explain", this way if someone clearly shows signs they can be banned from future contributions. C projects are way less likely to have the tsunamis of AI slop PRs that JS/Py projects suffer from.

And obviously screw fully AI generated PRs, I don't know how anyone can even defend those...

[SHARKgamestudio]

I think AI contributions should be banned. Especially 100% AI generated contributions.

I'm aware you do review contributions yourself Ray,
but I still think allowing AI generated code threatens the consistency and code quality of the library as a whole.

Because let's bet honest, you cannot review everything in a PR.
If it takes more time to review a PR than writing the feature took.. Than i don't see the point.

On a more personal note. The force of Raylib, and the reason we as a community are all here, is because of the humanity of this project and how much it brings people together, the context, the person and the maintainers behind it.

There are enough libraries powered by sterile AI contributions as it is. Let's not make Raylib one.

I dont want to feel like my favorite library is beeing contributed by something as sterile as generative AI.
Let it be alive, let it be human.

I apologise for my bad language & writing skills, english is not my native language.. :(

[JamesB7]

You should absolutely not accept AI contributions. They plagiarize far and wide, and when (not if) copyright is settled correctly in this regard, your project will be in a legally dubious state. You've built it just fine without it. There is no need and nothing to gain.

[JamesB7]

When someone submits a patch, it means they wrote it. When someone submits an AI patch, it doesn't mean that at all. You can't know that the licensing is even valid. You wouldn't let someone submit a patch they copy-and-pasted from GPL code, would you? Because it's worse than doing that.

[raysan5]

raylib had +800 contributors along the years, how could I know if any of the submited code has not been copied from a GPL codebase, from Stack Overflow, from AI or from any other place, if not stated on the PR by the author?

[maiconpintoabreu]

My opinion on this take is to block Al fully automated AI issue and PR creations as it is impossible to know if even the code from stack over was not previous generated by AI or another repo that the person got references from.

[spaciouskarter78]

I disagree with allowing vibe coded PRs. LLMs are still incredibly poor with lower abstraction languages like C because humans can handle lower level logic better. They cook up crap that somehow works but makes no sense.

I support banning vibe coded commits altogether but allowing LLM assisted contributions (only for fixing bugs that are PITA) and LLM usage should be disclosed while filing a PR.

We can't avoid LLM contributions anymore as the tech took over every piece of our life but we can have a good amount of quality control by limiting it.

[McDread]

No matter personal opinions on AI. Steam and Itch have started requiring AI disclosure. Even if it is not the case today, any developer using Raylib (and other tools that allow AI contributions) will probably have to disclose AI usage tomorrow on these platforms. I agree with OP that this question is relevant for any potential user, and I would like to add that it might be a good idea to not have the "no policy" policy hidden away in this thread.

[orcmid]

I don't know how I missed this poll until now.

I had to say "something else". While signalling that a commit somehow AI-assisted is fine, I think it is more to indicate in the code that there has been AI-assistance in some simple manner. Not that every line had some sort of assistance. I would not expect substantial portions of code to be so assisted.