What is this project's stance on AI and LLM-generated code?

[darltrash]

Does this project have any specific rules about contributing with AI, LLMs, in the form of code and/or creating issues?

[bjornbytes]

Currently there aren't any hard rules about this, no. I think I'd just say that low-effort contributions are discouraged in general, and AI stuff tends to be lower effort. Luckily LÖVR has mostly been spared from slop so far. Curious to hear what others think about this topic, will convert to discussion!

[DonaldHays]

I've dabbled with AI, but thus far only on the side, don't think I've even committed any of it to any repo.

I have a lot of mixed feelings on it. To me, it's felt like pair programming with a very, very fast junior programmer. It produces results very quickly, but the results range from fine to buggy to useless. Like it delivers work that's half as good in a tenth the time. It feels hard to argue against the cold arithmetic of that, but it also feels hard to have any pride, to say nothing of confidence, in the result.

Part of my hesitation on committing any AI code in real projects right now is the as-yet unsettled legal landscape. It's already so widespread that I doubt we'll find ourselves in a situation where AI is totally outlawed and everyone has to painstakingly remove all AI from their work. But the trend line right now seems to be that AI code itself cannot receive copyright protection, but its work product can violate others' copyrights, and you can be liable for that. So, for any project, you have to ask yourself what your risk appetite is for that.

[shakesoda]

i would really prefer any project ai involvement to stay indirect (such as research/glorified googling, explaining wikimath and so on) and avoided for anything of substance in actual code, issues and docs. i want contributions by people who have solved the problems for themselves and actually have some hope of accurately explaining what's going on in the code and why (without just making crap up constantly).

as for issues, i'm not interested in talking to chatbots. if it's just a translation or something i don't care, but the actual writing should be done by a human.

[darltrash]

It's genuinely heartbreaking to see an issue, where clearly everyone commenting and replying is just using Claude to write an answer for them.

[darltrash]

Personally, I'd want this project to be Anti-AI when it comes to actually contributing code or generating issues, it has become a prevalent problem in open source projects to slowly worsen in quality of code due to their acceptance and overall reliance of LLMs to do the tasks.

I am asking because I'm looking into adding this project to a no-ai project list that my team has been working on.

[cyrneko]

additionally, I only think it makes sense given that LÖVE also does not allow AI-generated contributions

[BobBagwill]

I think forbidding AI is a lost cause. I wonder if you could create a proof-of-work by saying the PR submitters must have a active fork on Github with successful CI builds, and only accept those PR's. I think that would eliminate lazy driveby submissions, and would ensure that at least they build. A stretch goal would be to require that their builds must pass the test suites. I have not tried it myself, but my impression is that a custom Github Action can be created to enforce this kind of work flow.

The ghostty project has started using this process:
https://github.com/ghostty-org/ghostty/blob/main/CONTRIBUTING.md

[cyrneko]

I think forbidding AI is a lost cause.

very common sentiment, but to not repeat myself, I've written about that before: https://cyrneko.eu/ai-policy.html

[stilic]

Since Love2D (which this project takes inspiration from) forbids anything made with LLMs or GenAI and is a popular framework, I don't think it will necessarily be a "lost cause" as you say.