I had a report for AI slop today. 1m LOC in 3 months. How do you even modify and distribute that? I do not think we had a rule against it.
IMHO if we allow, the community gets drowned out by AI vomit and it is harder to see handcrafted loved projects that are easy to modify and distribute.
My mind is to ban it to preserve open source.
Are folk in agreememt with this line? Would love to hear your thoughts.
A lot of people are making well-reasoned arguments about human nature, the nature of a dangerous tool, defining responsible use, and the challenges inherent in any sort of ban.
The points that look the best to me definitely include tagging, but I haven’t seen suggested any method of gatekeeping as to who has access to the means of dissemination of software. Because the damage by the tool is determined by the one who wields it, only people who have established trustworthiness should be allowed to contribute.
Yes to ban 👍
I had a report for AI slop today. 1m LOC in 3 months. How do you even modify and distribute that?
That makes some sense but does “AI slop” here refer to code with this type of problem or all code made with AI assistance?
This was a problem in !selfhosted@lemmy.world and they’ve done a good job of dealing with it. You could take a look at that past discussion to see what worked / what didn’t.
IMO there is an organized effort behind some of these new accounts pushing AI projects, and you need to deal with it in some form to stop them from coming
Agreed. TLDR, you must tag posts there with the following:
Post Tagging
There are three types of tags for project promotional / AI posts.
- [CBH] - Code By Human - A project promotional post with a project that did not use AI in any capacity. No disclosure necessary.
- [AIP] - AI Project - A project promotional post with a project that used AI in development in any capacity. Disclosure is required for how it was used.
- [AIT] - AI Topic - A discussion topic that includes AI. This may be for a discussion on hardware to self-host on, llama.cpp vs ollama, your experience using vulkan, etc.
Thanks for sharing this. Maybe correct tagging could be useful. Makes it easier to ignore but not overly heavy and possibly toxic.
Thanks for sharing this. Maybe correct tagging could be useful. Makes it easier to ignore but not overly heavy and possibly toxic.
If i am just posting these projects i found which i think are interesting (which i did a lot of). i am now suppose to spend more of my spare time researching if they include AI? I am pretty sure i won’t be able to know most of the time .
There are not a a lot of projects i think people will be interested enough to do a “high effort evaluation”. if anyone is that interested he can just ask, or check out the code which should be fairly rare.
AI has been here for a while. If it would have caused FOSS communities to be flooded by AI it probably would have happened by now. We should try and avoid the bias of risk aversion.
Looking at AI . it reminds me of how for wayland there was constant criticism even if it was just plain wrong and just made people overly scared . the plain wrong stuff was highly upvoted and i didn’t always have time to correct it, plus people being disrespectful discouraged me from discussing it because i didn’t feel it will be productive . now wayland is pretty much accepted as “good” by most of the FOSS social media communities in my experience. when a new technology appears. it takes time and effort to study it deeply and it’s advantages and disadvantages . And i suspect people in social media are often not the first to do that honestly.
Okay, then don’t post these projects if you’re not sure and don’t want to find out.
Yeh I’d prefer no slop. You’d probably also need to clarify what the definition or threshold for this might be. Just the code? Slop docs? LLM used for any aspect whatsoever? Other communities must have a definition that might help. But your goals with it as stated are sound. Let’s not drown in vomit.
Let people choose what they want to see.
People here tend to call any hint of LLM use “slop”, and arguing on a threshold is a waste of everyone’s time that will end up disappointing most. Tag it and let people choose.
Bro one million lines of code in three months is slop, without me having even read the code.
I made the report and I’m definitely pro AI when it’s used responsibly. Sadly about 95% of AI usage nowadays is irresponsible, unmaintainable, vibe-coded bullshit not even the prompters understand.
I’d like the 95% to be banned and the other 5% to be tagged.
the post is not about this 1M LoC project, it’s about creating a ban for whatever the mods decide it’s slop
What I wanted to say is: I’m pretty sure we can trust the community and moderators to recognise the difference between slop and AI assistance.
yeah, and I disagree. People have different opinions on this. A 1M LoC slop project is an extreme that is not representative of what shows up here.
I like this idea. Tag it. Mandatory disclosure. People should just be honest about the amount of AI they used and how.
“I vibecoded this awesome thing with 100k LOC in a weekend, check it out” vs “I used AI to help with planning and architecture, but wrote the code myself” are two very different things.
You have it backwards. It’s better to keep a tight grip on planning and architecture than implementation of individual functions.
I’m not making a value call, I’m saying they are two very different things.
Let people choose by their owns. In my client, I can filter by texts; I already used to hide some topics. Maybe it’s more convenient to include a tag in the title. For example: if your contribution/project is made by or include generated content from a LLM, include the tag [AI]. That way, anyone can hide these contributions by filtering by the text “[AI]”.
Yeah I don’t think we need slop projects being posted here. I know a lot of people are using these tools even for personal projects nowadays but tbh I’m not really interested in exploring any project heavily written by AI.
I’m conflicted. I’m anti-AI due to distrust in capitalism and our pseudo-democratic plutocracies, the profit motive to replace the human workforce, the reckless criminal-incompetence of its implementation, and apparent stupidity of the average hairless primate in understanding that you can never trust a correlation engine… but it’s just a tool, and I’ve also found it incredibly useful in my limited and controlled usage; getting it to write small functions or blocks of code, then amending and composing them myself.
How do we determine what is “heavily written by ai”, and all the obvious slop factory bullshit, without banning strong projects who are just leveraging it for legitimate dev work; as a glorified autocomplete, test autofill, etc but still read and validate every line of code?
Like most people, I’m finding it increasingly difficult to find trustworthy projects amongst the see of amateur slop, but I think banning it out-right would likely result in many false positives, and ultimately be counter-productive.
As much as I don’t want to depend on metrics, I think a better approach would be to define a bunch of heuristics based on historically-competent projects. Things like defining an upper limit based on LoC vs commit time and number of devs, project age, etc. Ultimately I feel the industry should do what cybersecurity should be doing; moving to some sort of cryptograhically-verified social trust model where engineers/professionals have to build trust in the real world, and take turns in “vouching” for projects and code releases.
Yeah, exactly.
As someone who has been working on a large open source project for literally years, while trying to support a family, the hardest thing is time to devote to this project (which, absent capitalism, is what I’d spend my life doing).
AI can allow me to instead increase the speed at which I’m delivering the project I’d already architected, designed, and laid out with my other devs. I don’t have the money to pay myself to work on it, much less other devs - and my bank doesn’t take repo stars as mortgage payments unfortunately.
I hate the AI industry - but that’s a problem with the economics, not the tool.
I’m happy to label that my project uses AI assistance, but bans just make me realize that wherever is banning it isn’t more concerned with shipping a solution instead of a crusade against LLMs.
“distrust in capitalism and our pseudo-democratic plutocracies, the profit motive to replace the human workforce, the reckless criminal-incompetence of its implementation, and apparent stupidity of the average hairless primate”
I have hair, thank you very much, but seriously get out of my head. The hubris these days is unbelievable.
I’m anti-AI due to distrust in capitalism and our pseudo-democratic plutocracies, the profit motive to replace the human workforce, the reckless criminal-incompetence of its implementation, and apparent stupidity of the average hairless primate in understanding that you can never trust a correlation engine… but it’s just a tool
Doesn’t that whole preamble prove that “AI” isn’t in fact just a tool? Or rather, that it’s a tool designed for all the wrong things? There’s a well known Audre Lorde quote that springs to mind.
the problem with “just a tool” is that using it will inherently change how you act.
Best concise way i can put it is that it’s like how better hardware has largely just resulted in software not really caring about efficiency any longer.It is just a tool, but human nature makes it a net negative. There are a lot of tools like that, and history shows that once the cat is out of the bag, humanity just has to live with the damage that tool causes. There’s literally nothing that can be done to stop it on a large scale, but small communities can always dictate what they will tolerate.
I’m fine with AI. I’m not fine with slop. Ban slop, not LLMs.
Fair enough, I don’t have said decades of experience
You are the same.
LLMs are slop.
Sorry about the downvotes, Lemmy has a big coding community (naturally, in this comm), and that seems to be the sector where people are the most hypocritical about “AI” usage.
- someone with zero experience using AI to blindly create something barely functional, feeding in errors until it gets a working version. No checking for vulnerabilities, testing, error catching, no thought on architecture or scalability.
- someone carefully using AI in a controlled manner to create open source software. They spend days making sure that is reliable, scalable, secure, and that the project is quality.
You believe these two are them same? That both are equal slop? Quite frankly, I don’t care if an AI made most of the code for an open source project as long as it is made well, and allows me to get away from awful proprietary alternatives. If the output of an AI is distinguishable from a quality programmer, idk how that is also slop.
By definition a quality programmer wouldn’t want to use an LLM. Just like a quality architect wouldn’t want to build a house by making monkeys throw bricks in a pile, with the excuse that he goes after them to fix it.
I’d actually say that the “quality” dev that vibe codes is worse, because it often comes with an ego that leads to them wanting to hide that they use AI, or to hide the fact that no, they don’t master their code nearly as much as if they wrote it by hand, and it leads to more crap.
I don’t know how many more studies proving that LLMs are crap we’ll need before the slop stops.
Just like a quality architect wouldn’t want to build a house by making monkeys throw bricks in a pile, with the excuse that he goes after them to fix it.
If the result of the monkeys throwing bricks in a pile is indistinguishable from a team of builders, what does it matter how it got there. If the end result is a pile of garbage, of course it’s unusable and a waste of resources. It’s about the end result.
I don’t know how many more studies proving that LLMs are crap we’ll need before the slop stops.
Imagine if in the year 2000 someone said “how many studies does it take to show computers are slow before we stop wasting time on them”. It’s a technology. Whether you like it or not, this moment is the worst the technology will ever be. The studies you reference are about outdated technology
There’s some nuance for sure, but personally I’m a lot less interested in heavily LLM-programmed projects
that’s completely fair - would labeling then be ideal for you?
Well, fuck… how could I possibly win such a well-versed argument like that?!
No, you!
Yes. They are useless. Get rid of them.
Most important step is to decide exactly where you want to draw the line.
Agreed ! I think a complete AI ban on every element of the project is too strict. But I also have no interest in a massive AI repository with no supervision (and who can supervise a huge AI commit ?)
For me the most important thing is that something is free software. I’ve seen people advertise proprietary stuff in this community and have reported it, and nothing has been done. I don’t know if that means my reports don’t federate properly or if it’s tacitly approved by the moderators. I know in the degoogle community there are "trusted’ proprietary apps that are often recommended but this should not be the case in a free software or open source community. In any case, any rule that bans “AI slop” while allowing proprietary products is in my mind backwards.
Regarding “AI slop” I am generally in agreement with Codeberg’s policy which states,
You must not share projects that mostly consist of code written by “generative AI”-tools (including services such as Claude, OpenAI Codex).
I suppose “mostly” is up to interpretation but for me this specifically targets vibecoded apps where there is no or very minimal human involvement. This may be heterodoxy but I actually don’t think “an otherwise human-written codebase with a single line of LLM code” is the same thing as an entirely LLM generated codebase. I think there’s room for nuance especially if the usage is disclosed (I don’t think punishing transparency should be encouraged).
While I’m personally against AI coded projects I’ve noticed more and more that the best projects for my homelab have either turned into AI coded projects or started as one. I first and foremost look into alternatives and weigh in if the features in the AI coded projects are needed over the one that isn’t. But it’s been harder and harder over the years and months to find projects without it.
While I see the different spectrums of fully supporting it, to seeing it as a tool to outright hate it and disregarding everything with it completely I’m mostly curious where does the line go?
When can you spot a good developer that uses it and when can you spot a bad one?
1 million lines of code in one month from a hugely experienced coder vs a developer with none?
Without deeply vetting the developer and their history and their other projects code how would it be best done?
I only really do SvelteKit stuff for example, some bash scripting too, so I can’t spot code efficiency in other languages like Rust or Go.
I’m on the stance that I always avoid to the best of my ability AI coded projects because I don’t believe that AI can generate code that is efficient and secure in the long run. I know of a few self hosted projects that has been used to some extent with AI and has gotten reaaaalllyyy bad security patches.
I know of a few self hosted projects that has been used to some extent with AI and has gotten reaaaalllyyy bad security patches.
Yeah, I noticed similar too. I’ve been using a game mod for many years that is in active development, and its devs switched to very AI-heavy development earlier this year. They now add new features with lightning speed, but they are noticeably buggier and unreliable.
Never in my life have I wanted to read about or use an AI slop project.
Since speaking directly to the proposed change didn’t do much, here’s an alternative to “banning ai slop”:
Ban the thing that’s happening. If there are too many “low effort” posts, ban low effort posting. If there are too many posts about non open source packages, ban posting about non open source software. If you really don’t want people self promoting with open source (a thing they’ve been doing for decades!) then ban self promotion.
The problem with all the things I just suggested is that they require serious consideration on the part of a moderator. “Is this post too low effort?”, “is this person promoting themselves in a way the community doesnt want?” and “is this post even about open source?” Are questions that require moderator effort. “Was the repo signed by Claude?” Takes no effort to answer.
Of course kicking kids out of the sandbox for not playing right kind of defeats the purpose of a sandbox and placing unnecessarily high barriers to entry on the commons defeats the purpose of a commons but what the hell do I know…
The biggest problem I’ve seen with the “use AI help but you’re ultimately responsible for what you submit” approach is that it’s damn near impossible to usefully define or enforce what “responsible” actually entails.
Yeah that’s the problem: accepting that there is some other failure happening, broadly defined as a lack of “q” factor, forces people to grapple with the fact that open source has historically been the purview of the bad programmers self promoting their learning projects.
There are cases to the contrary of course and that’s not the rule, but no one cared about these behaviors until now and their care is all focused on ai use as opposed to the behaviors of the programmer in question.
Your response is a perfect example: even though you’re declaring what the person is responsible for you can’t articulate what the consequences of that responsibility are.
I’m not picking on you either, this has been a problem for thirty years, oh you just graduated? Grind away at the gnu tools or some open source stuff so you have something to point at in interviews. Youre reimplementing an x feature in windows having finished a book on c++? Throw it up on sourceforge.
Just ban low effort content, AI or human made slop both count and have no value.
What about high effort slop? Seems like a new category that’s growing unfortunately…
