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.
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.
“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.
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
Both the Linux Foundation and the EFF have permissive policies towards AI generated code, even Linus Torvalds accepts it as a tool.
Some FOSS projects that have AI generated code in it:
- The Linux kernel
- Fedora
- Mesa 3D
- Jellyfin
- Firefox
- Chromium
- Homebrew
- OpenSSL
- FFmpeg
- Curl
Should we ban them all? If so, this community would be either dead soon OR it would be hypocritical.
Or because they started pre-AI it’s ok now? Shouldn’t we care more about disclosure, policies, licensing, code quality and maintainability?
Users should downvote crappy software and crappy devs, regardless of the tool it was written with.
I am convinced that, with the right tools, it’s possible to damage 40 years of work within a few weeks.
I would support such a ban so long as there is some process to prevent it becoming a witch hunt.
I have personally already received one false accusation of using AI in this community and I wouldn’t want to continue posting projects here if it becomes an atmosphere of suspicion.
One? I’ve received three. It seems you have to dumb down your responses, and add some intentional typos if you don’t want the mob after you lately.
Vibe coded code whose “author” understands the language and actively challenges the agent for his decisions in the interest of long term window is OK. Vibe coded coded that are one-off without any future prospects should be 100% banned.
We allow it on our large subreddits on reddit. What we do is use a bot to filter repos that are less than 3 months old.
People who vibe code slop just want to share their thing and then abandon it. A limit like that or even 6 months will kill majority of this sort of low effort crap. Most of these people just let Claude or whatever sign the repo as well, another easy way to catch slop.
Developers that develop and only use AI as a tool build great products still as they review all their code.
A blanket ban is just going to make people be more sneaky.
I think one of the appealing traits of open source is that you can identify what piece of software and hardware does what (and how, if you’re interested). You can post your exact method, and another person can follow along and replicate the result.
But if AI tells you how to build everything, you don’t get the opportunity to learn, and your stack depends on what that particular model thought was best on that particular day. Maintenance also becomes harder if you just did a
sudo apt install whatever Claude told me towithout knowing what everything does and how it interactsBan it, lemmy sets itself apart from places like reddit by having actual quality posts.
Maybe start out with strict labeling guidelines. If it’s AI then it should be labeled as such. If a user posts something that’s Ai and doesn’t label it then ban that user. Maybe a warning first up to you.
But if the choice is between blanket everything allowed or blanket ban I lean more toward blanket ban.
define slop. it means different things to different people.
I thought it was “messy, generated code with little cohesion and only barely functional”, but really it’s the cohesion part.
messy, generated code with little cohesion and only barely functional
Okay, you’ve described my code. Now what about the AI?
thats mine too
Oh man don’t look at some of my earlier repos while I was learning then.
Completely vibe coded - yes
AI assisted - no
The problem is, at what point becomes it just Ai assisted. And who decides and how to proof this. These uncertainties are part of the problem for rules like these. Not only is it difficult to decide and time consuming, it also creates friction in the community. Note, I am not against having this rule. It’s just, it is such a depressing situation.
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.
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.
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.
