There are some curated lists of no-AI software. Would be nice to have an index / plot of how that changes in time.
I can't think of a functional reason for a no-AI policy: if it runs, it runs, regardless of who or what made it.
Also, even if you avoid AI-generated slop, you can't really avoid the human-generated or human+AI-generated slop that passes your filters.
Still, I can definitely think of good non-functional reasons: provenance, accountability, proof-of-work, encouraging people to write code themselves, empirically tracking how humans develop codebases, etc.
And some projects like WINE or ReactOS probably have to worry about that even more given they need to guarantee clean-room reverse engineering...
There are probably some subtle bugs I can't explain in the code I wrote all by myself. I sure had a few "what was I possibly thinking when I wrote this" moments working on some old code - and that's only the bits I know about. And I sure had countless times people pointing out "hey, you got this stinky here" in a code review (which is the whole point of it). Attention lapses and brain farts sure happen. Slop can be more frequent with LLMs but it's certainly not a LLM-specific issue. They're very productive, there's a literal outbreak, and by the sheer volume shadow any The Daily WTF stories.
However, I can agree that LLM-generated code most likely has higher probability of slop. But then, a policy "a human contributor MUST fully know and understand all the contents of the submitted work, in fine detail, all the way down to every single line of contributed code and documentation" would probably address that in a more functional manner. And then the code can be from an LLM or monkeys with typewriters author had seen in his sleep. That stops to matter because author takes ownership and responsibility: "here's a recognized rational agent who swears by their work". Makes non-self-authored code require a lot more effort (unless it's a trivial change for obvious reason), but arguably even more robust than self-authored code.
That is, unless the PR authors tend lie about their knowledge - but that'll be a whole different story, where LLMs will be just a background detail.
(I'm not saying Godot should be done something different - their project, their rules, let's use that as an opportunity to watch how it goes. Just musing on the matter in general, if there's any rationally explainable merit in such policy.)
For many people that’s enough of a reason.
As for functional, you can see it all up and down this comment thread. People don’t check their work and leave these massive walls of text and codebases that someone else has to audit/cleanup. It’s exhausting. Too many people offload their work to AI and put zero effort into vetting the results, which punctually means they are just offloading the work downstream. So many maintainers are simply going “no I will not do your work for you,” which is a very functional decision.
To butcher a comment I read on HN that put it very succinctly months ago: everybody wants to let AI do their work for them, but nobody wants to be downstream of AI work. It’s a seriously problematic dynamic on many levels. And that dynamic will not change until the vast majority of people start reliably vetting the results, which I don’t think is going to happen because babysitting a black box and trying to force it to output something a specific way (or constantly copy editing middling work) is not something that most of us enjoy.
There are also plenty of valid personal reasons for refusing to generate code with AI - learning-by-doing and ownership of the result being the main ones, IMO.
> everybody wants to let AI do their work for them, but nobody wants to be downstream of AI work.
This is also true in my experience. But in my work, I found that I don't care how the code or comment was generated, as long as it doesn't try to overload my brain with irrelevant and obfuscated things, and as long as the person is not pretending that it's true, verified or their own creation (when it isn't).
Agreed, but my main point is most people continue to do exactly this and simply won’t stop. They think “AI took care of it and it’s good enough” then essentially shove their work on to the recipient 30% completed. So long as that’s the way most people use LLM’s we will continue to see restrictions put in place by the recipients.
https://codeberg.org/brib/slopfree-software-index#why-care-a...
Not exactly an argument against using AI, is it? It's a bit like saying that GPS makes people worse at navigating by memory, which is true, but also not a strong argument for going back to paper maps. I feel the discourse is more about "stop using AI" and less about "how can we ensure our backup skills doesn't disappear".
But I think the point the parent commenter made is: if there's not a functional difference in the result (i.e., job satisfies the definition of done), it doesn't matter if the AI generated the code or did the diagnosis.
But I think it's also fair to say that the process matters, even if the result is the same. If something exists for our benefit (e.g., there's no real alternative to learning-by-doing, and people need to know stuff for safety/security reasons), and we're fine with the trade-off, there's no reason to just give up the process to the AI.
(And I'm unfortunately no longer as certain that "Well, that sort of thing can't happen where _I_ live." as I would have been a decade ago.)
you can see it sort of like making a list of vegan restaurants. you might not see anything wrong with other restaurants (they might even have vegan dishes) but to some people it makes all the difference because they get to choose who they support
Imagine morals.
That makes much more sense now. The inability is completely on you, and you admit it at least.
There are functional reasons for a no-AI policy. It helps the Godot Foundation function to establish a no-AI policy. Do you argue it doesn't help them function?
Do you understand the difference between functional and non-functional?
Alternatively, if you have some vague idea [1] about what you expect to see/have, and the running code satisfies that idea, then it also runs.
Obviously, there are plenty of non-functional specs (e.g., security, cleanness, readability) that a code should probably fulfill before one finds it acceptable, but these are also not somehow impossible for state-of-the-art models to satisfy.
[1] Vibe, if you prefer, tho I dislike the term. Another related term is eyeball estimation.
If you use rsync clone by an LLM to copy a million files, will you bother to verify every single one was copied correctly?
And I'm not disagreeing - it is hard to anticipate what needs verifying, regardless if it's functional or non-functional.
But if it's not a spam submission, you could probably design tests or static/dynamic analysis tools that can verify those million copies much faster than manual reviews.
So you have integration tests that verify the general specs of the software and rely on your skills to verify the finer details. But if you’re using an LLM (and not reviewing every line), you can no longer be confident about those details.
And reviewing every line kills the speed advantage of using LLM.
1. Accept quality contributions from someone who understands what they're doing
2. Cultivate a relationship with the contributor who might potentially become a core-team member. Maybe even the next maintainer
You could get a perfectly adequate instance of a PR that is easily readable and verifiable while generated by an LLM, but generally they're not.
A policy pushes the aggregate to at least what looks and communicates as a human made PR that is functionally easier to approve. Whether they are created by an LLM or not is then secondary, but it likely pushes all PRs to be better.
But I'd argue that some projects [1] could benefit from the speed (and sometimes, quality) of AI code generation without filtering by something that's difficult to identify (i.e., is it truly human-generated).
One way could be to constrain the size of each commit and PR, and invest more heavily into the review process (e.g., tests, static/dynamic analysis, sandbox deployments), so even if you get 100s of contributions, you can knock each out quickly.
Obviously, easier said than done. And at that point, you may as well use the AI to make the commits yourself, instead of relying on community contributions.
[1] Of course, this is only the case if the project's only purpose is to be a tool, and not also an educational reason for humans to learn how to code - in which case, it makes sense to invest more into identifying the "cheaters".
I'm sure some are convinced LLMs can (eventually) manage everything, and others (I'm leaning more here) are convinced that you will always need a minimum amount of people both educated in the fundamentals and the domain to steer the project, and these people wont exist down the path of non-human PRs.
> I can't think of a functional reason for a no-AI policy
These lists don't have you as an audience.
If it fails to run who is responsible to fix ? Problem with AI is that it does not have causal-models of how something works, and the reason for that is that it doesnt interact with the world. I think of it as an armchair expert who spouts recommendations without having a grounded understanding in the real world.
AI-written code is far less likely to run than human-written code. Even worse, it often gives the appearance that it will be fine, only blowing up down the line. That is an extremely strong functional reason to reject AI code.
Obviously, this won't apply to every context: I work primarily with well-known langs (e.g., Python, JS), small to medium codebases (<500k LoC, for sure), and relatively few co-developers.