> In addition, Codeberg conflates “having a community” with “being legitimate software worth hosting”
I don't think this is a fair summary. I don't see this conflation in Codeberg's blog post. The words in quotation marks don't actually exist in Codeberg's blog post.
It seems that what they are saying is: In a free hosting platform where we share all the resources, it's not fair that some people using LLMs are using up disproportionately huge hosting resources to handle the unnaturally large amounts of resources they can churn out with these tools.
I'm a solo dev, and I couldn't find the judgement against solo developers or any slight against people who don't have a team. The concern is that some people don't gobble up an unfair amount of resources.
I personally have been moving to self-hosted Foregejo for ultimate freedom. But I also think it's fair for some providers have the freedom to take a bit more of a harsh (or based - depending on your perspective) stance against LLM generated code. I mean, a lot of people have been bemoaning the fact that the human-generated material has been increasingly hard to find in a sea of AI-generated material. It could be nice to have a corner of curated, human-made code still available on the web.
> It seems that what they are saying is: In a free hosting platform where we share all the resources, it's not fair that some people using LLMs are using up disproportionately huge hosting resources to handle the unnaturally large amounts of resources they can churn out with these tools.
Then instead of addressing the excessive resource usage problem indirectly via banning AI-assisted apps, they could've addressed it directly by imposing limits/quotas.
It's abundantly clear (and not that they are hiding it either) that their issue is with AI itself so anything else is just a secondary concern if not even a smokescreen.
---
> I'm a solo dev, and I couldn't find the judgement against solo developers or any slight against people who don't have a team.
Quoting their blogpost [0]:
> ## The development team of none
> Using LLMs to work with your code gives you a kick of adrenaline. You can develop at a rapid pace, build things as if you had a large team. Only that you have none. In fact, you are (often) alone, working with a statistical machine that turns energy into code.
I couldn't care less about what they think about solo developers, but it's just one bad excuse after another and that's what people are criticising. Open Source is one person. [1]
> they could've addressed it directly by imposing limits/quotas
For one, they've already done this to an extent. For two, they're also doing this by deciding what projects are and are not welcome on their service in the first place.
If 9 people eat an apple or two a day but a tenth person wants to eat 100 apples a day, we don't have to set some apple limit. We can tell the tenth person to pound sand and keep giving everyone else the apples they want.
Having a hard quota means each spammer will fill up the quota, then get some error they don't know how to fix and stop. This means the quota has to be low.
Just saying "no spam" means you don't have to waste a single byte on spam and everyone else can have a higher quota.
>but whenever something expresses its right to do or not do
The problem (or the virtue?) of this comment is that it's impossible to decipher with which side it disagrees, and for what reason too. Cause both sides expressed their right to do or not to do.
They don’t pay tax, right? Correct me if I’m wrong. It’s definitely a different (and public) conversation when things are claimed done for the public benefit and thus exempt from tax.
At least in the US tax-exempt status carries obligations.
I've banned this account. You can't attack other users like that on HN.
If you don't want to be banned, you're welcome to email hn@ycombinator.com and give us reason to believe that you'll follow the rules in the future. They're here: https://news.ycombinator.com/newsguidelines.html.
In the ai-article op did say they’d do exactly that. Though it is surprising they left GitHub due to ““enshittification” of the platform under Microsoft’s ownership”, when it’s likely GitHub is using similar ai practices the op is and matches their exact ai usage.
Citing the op themselves: “ an alarming number of self-described AI experts whose expertise appeared to consist entirely of having a ChatGPT or Claude subscription and the willingness to write about it in seventeen-paragraph posts.”
If you can’t keep up with the load, say that. Limit new signups and/or implement rate limiting, and ideally scale up if you can.
DO NOT try to spin that into taking some form of moral high ground. DO NOT make yourself the arbiter of what is and is not considered a worthwhile project. That is not neutral. That is not free.
They're not obligated to deal with the higher load brought on by people's LLM coded projects, CI pipelines etc. In order to keep things freed up and flexible for others, I think it's fair for them to say, "in this community, we're going to save the space for more hand-coded projects."
It's not fair for everyone to have to eat the costs of AI (like communities getting bulldozed for data centers) and just accept that it has to be used.
Of course nobody is obligated and I never claimed that. In fact, I fucking hate the LLM hypetrain. But this decision is decidedly non-neutral and non-free.
Fair point, but I mean perhaps it's not fair for them to have to impose usage limits (which would limit everybody) when they could just make a decision to avoid certain types of projects (which they feel strongly about.)
Simple: Provide the same freedoms to everybody. Do not get involved unless absolutely required to by law. That entails not restricting access for specific projects, and not revoking freedoms for any reason. Everybody should get the same treatment.
For example, in the case of being overloaded by AI contributions, a neutral solution would be to implement the same rate-limiting for every project. Set a sensible limit, and enforce it universally. Do not discriminate specific projects.
Why should GNOME and Forgejo, which are great projects, have to suffer the same limitations as a blob of non-working output from a Markov chain?
It is like saying Gmail shouldn't delete spam, but instead, should apply a limit of 1 message per day per destination per source address to all messages.
Most vibe projects aren't even projects in the way that spam isn't even mail. The message says it's from my bank but it isn't. Why should I have the same limits on messages that say they're from my bank and messages that are actually from my bank?
Yes, human rights should apply to anybody, including war criminals. That is what makes them HUMAN rights.
Universal freedoms and neutrality should apply universally, including to waste-of-time-and-resources AI slop. That is what makes them UNIVERSAL and NEUTRAL.
> It is like saying Gmail shouldn't delete spam, but instead, should apply a limit of 1 message per day per destination per source address to all messages.
I said a REASONABLE limit should be applied. 1 message is not reasonable and you know yourself that this is a bad faith argument if you are honest.
1k messages on the other hand would be a reasonable limit. There is pretty much no legitimate use case for sending more than that to any specific address from any specific destination in a single day.
It is also bad faith because Gmail never claimed to be free or neutral.
Agree. Providers who don't want to deal with LLM generated code should be respected as part of the free market. Let the market decide which direction is for the better or at least allow for niches that cater to certain audiences.
It's wild to me that nobody is considering enforcement. Part of the reason this is a bad rule is because there's no way to actually validate what is an LLM coded project. And the AI train is not nearly at its destination. Things are going to change dramatically over the next four years. My prediction? This is not going to age well.
It'll be the same as literally every other ToS ever: if they detect a problem and trace it to you, and you're blatantly violating the terms, they kick you off. If they detect a problem and trace it to you, and you're subtly violating the terms, they lock out whatever is causing the problem and get in contact with you.
You're trying to sidestep how enforceable something is, and instead implicitly admit that enforcement will be inconsistent. That's my issue. Not all ToS are this way, and they're better for it.
On the contrary? Selective enforcement is tyranny. I'm sure the projects that they disagree with politically are far more likely to be "AI generated" than the projects that they agree with politically.
Define resources and create a better limit then. 10 small code-only projects will take orders of magnitude less storage, bandwidth and cpu than one project with committed image and sound assets.
It's a community code forge. If someone poured their heart and soul into making an awesome game with kick-ass image and sound assets, the community probably wants to support it.
Their stance is obvious: if you just want dumb git hosting with a freemium model, use something else.
If a carpenter uses a table saw, is he still a carpenter?
What if he just throws wood randomly at the table saw and tries to sell whatever pieces come out?
(Never do this with a table saw, btw. They can, and will if you don't use proper procedures, throw wood sideways at speeds that can punch through walls and impale people)
This analogy doesn't make much sense to me. Is the table saw analogous to LLMs because they both work with the appropriate substrate (wood/text)? I don't see how an LLM is like a table saw at all.
Is a musician less of a musician if they use a pre-recorded back-beat and sing on top? Are you not a musician if you use the chord mode on a keyboard?
Is code not hand-written if my editor automatically closes all of my html? What if my editor flags all of the errors and I manually correct them? And what if it auto-corrects at the press of a button?
There is a line and I'm asking about it earnestly, but the table saw analogy doesn't reach it IMO.
What if I barely know the rules of chess and I use a chess computer to play all my games for me; I win the world title this way. Am I a chess Grant Master?
People downvote this kind of take because you're striking at the heart of the enforcement problem and also the definition of what a project that is majority built with AI really means. What they really want is to ascertain the amount of effort from a human that went into the project and they're using the percentage of code generated by an AI as a proxy metric. But it's a bad proxy, both because it doesn't capture human effort and because it is not measurable.
Even human effort is a terrible proxy because effort is a gradient that depends on skill. An expert exerts low effort for the same result a novice would labor over.
I recall having this same exact discussion with my philosophy professor years ago. We all know there is something different about a hand cooked meal and say putting a tablet in a box and a fully cooked meal comes out for you.
I feel like it may be closer to attention? That's the thing we actually value, that someone took the time and put in the effort to do something. And the quality depends on how much we value that attention. Like high effort low quality is praised when it's the work product of your child for instance.
What conversation are you trying to have? I don't even use Codeberg; I've spent years thinking about how the idea behind rules manifests when those rules are actually put in place and enforced. I think a lot about second and third order effects. That's my interest here: an org decided to put these particular rules in place, and based on my experience, I don't think it's going to age well.
But I guess you're right, as far as you go: none of this is my problem. I'm just a curious observer.
No, I need to apologise, I kinda got confused in the threads. That is, I thought this was downstream from another comment I made on the topic, where my point was basically: you (not the literal you who is interested in this on a different level) know if you used LLM. So if you have to ask, the answer is probably no, and if you're genuinely unsure, you can ask them.
I did you wrong, and in general we shouldn't guess the motivations of others, but I'll still say that lot of the comments asking what is the exact meaning of that rule is, how it will get enforced precisely, strike me as seeking to undermine the validity of their decision rather than wanting to make sure they don't run afoul of it. Or even coming up with ideas to help them enforce that rule better. One even said maybe we should test if they actually can tell. I responded to you as if responding to that, and that was uncalled for.
Take the HN guidelines, for example. It's impossible to make them completely "clean cut" rules. But people know that if they genuinely think there's been a misunderstanding they want to clear up, they can contact HN. And with Codeberg it's probably like that, too. Not like Google, where the rules ideally should be 100% clear and mathematically accurate, because you just get nuked by algorithms with no recourse.
You might even say if people post on HN, they already understand what fuzzy rules are, by definition. They know language is fuzzy and that the letter of the law points at the spirit of the law. If someone doesn't want to understand that spirit, no amount of legalese detail will make them understand it. But of course that shouldn't be an argument to shut down discussion about it, either, and I didn't mean it like that.
Thanks for the reply! There's an interesting topic here with respect to fuzzy rules. I definitely like them sometimes because they can capture values and allow for human judgment (which is more costly, but usually better than hard-and-fast rules). I think works well on HN because dang and tomhow tirelessly patrol and enforce, and they both have superb judgment - they aren't just randos that got to moderate the subreddit, they are quite literally the backbone of what I think is the best social site that's ever existed. Lots of human effort there.
Codeberg is different. It's a non-profit, so my criticism is mostly around all the heat and light that will be generated when they start banning projects (politically), and the amount of human effort that it will take to assess whether a project is "majority" written by AI. I just think the way the rule is written is fraught with issues that will cause a lot of wasted effort in the coming months and years.
Making a principled ban against LLM would seem to allow for more freedom and flexibility with usage limits. People who are creating more by hand would still be allowed to have more breathing room. With a hard limit of resources everyone would be more restricted. That's also a solution, but it's just a totally different approach. The LLM ban could help to allow for freedom without the resources getting overwhelmed.
Only if you can reasonably separate LLM from non LLM code, which is in no way a solved problem. Even if one agrees with their priorities 100%, it's just asking for non-automated enforcement. It's difficult to execute, and either lead to completely arbitrary enforcement, or a lot of resources being spent in exchange for this freedom
you mention. It's just a very hard practical line to try to draw.
And let's be real, most of the costs LLMs incur on hosting providers have little to so ln who makes commits. A popular project than bans LLMs from contributing, but is used often by LLMs that are using the project as a library is indistinguishable from one that allows LLM contributors.
So from a practical perspective, it's a head scratcher
It's true that an LLM ban is really hard to enforce, but that doesn't mean it's worthless. If i make an image editor and say "It is my policy that serial killers and pedophiles cannot use my software" i can't really enforce that, but it still makes a (somewhat useless) statement.
In my opinion, not allowing LLM code at least signals that codeberg (and forgejo) itself is unlikely to be majority LLM code, which raises its stock in my book; It signals what kind of people the maintainers are.
I really meant -- are these maintainers passionate about the design and maintenance of their tools, are they trying to slop up a demo fast enough that some VC buys them out, is this project a goal in itself or the means to some other end, etc...
Open source software that exists for its own sake is exceedingly rare, and understanding the goals of a community before becoming dependent on it is a good idea
OK! This makes sense, and I agree. But I don't think banning LLM usage gets us there. LLMs are going to be used for coding, whether anyone likes it or not. I think focusing on the effort and care that is put into the codebase is a good metric: LLM generated code that is reviewed by teams of professionals, for example, is a huge win, and should be encouraged.
I readily concede that there is a huge volume of inexperienced programmers generating crap, though.
If you made this statement and you catch a serial killer using your software, it affects the options. A more reasonable example may be: Microsoft may not reverse engineer this. They still can do it privately - you can't enforde it - but if you catch them, then you have more options besides saying "welp, I said they could"
Apparently some people have compared it to some other unlikable groups of people (I know this because a vibecoder complained about the comparison someone else made)
If someone has to ask whether their software is mostly AI-written, then it is.
If someone has to ask exactly how racist you have to be before you can be called a nazi, they're probably a nazi.
If someone has to ask how attracted to underage girls they can be before they're a pedophile, they're a pedophile.
If someone has to ask how responsible you have to be for a death before it's a murder, they're probably a murderer.
Because people who aren't these things clearly know that they aren't.
My interpretation is that the resource usage should be roughly proportional to the human activity involved. That wouldn’t restrict resource usage by any individual project, which might be a popular project with many contributors and a lot of activity, but the resource consumption needs to be justified by corresponding human effort and interest, which LLM usage doesn’t exhibit.
In this corresponding blog post, one point made is that LLMs tend to LARP big-project infrastructure on small projects. They use the resources for something like GNOME but they aren't actually GNOME so they don't have any excuse for it.
The main problem is that so many LLM users are selfish bastards and they won't respect the spirit of such rules and will only stop when the hard disk quotas force them to.
So now your solution entails lowering the quota for everybody because allowing in LLyou must assume eve
Do we have a tool that can detect whether a project is majority LLM-written? This is an unenforceable rule, unless it's just coming down to vibes and they're going to decide what projects to ban based on their feelings.
Are you sure? It's pretty easy to find telltale signs of LLM usage, like the CLAUDE.md file, the '.cursor' directory, the 'vim'-sized codebase done in a week.
That's just vibes. Measuring how much data someone uses is concrete and empirical. They are implementing TOS rules that are based on judgement and require a judge. Very different when implemented practically, especially in terms of effort on the part of the org. Quotas are automated, evicting projects is highly manual.
They are not. You can probably draft some query to get candidates based on those criteria. And so very quickly.
> Quotas are automated, evicting projects is highly manual.
The human world is very chaotic. Automation is not a panacea, which is highlighted by Big Tech abysmal support. Sometimes, you want humans to judge, because the optimal result is based on heuristics and not rules written in stone.
Non-profits like Codeberg are resource-constrained. Apple is delaying app reviews and degrading the reviews that are done because of a slew of AI-generated app. Codeberg is basically adding a "human must review things" step to solve the "we have too much traffic" problem. As a bonus, they are instituting a process that's not only costly, but is vulnerable to being co-opted to ban projects that are not desired for any reason. It's just a naive policy mistake. Time will tell. Let's revisit in 2 years.
Codeberg is a registered non-profit charity whose purpose is to provide free services for free software projects. You can make a different site that works the way you want it to.
Yeah, the natural solution here is to only offer some amount of hosting for free and then charge for larger amounts of code to host. Those quotas are something that is empirical and observable and can be managed. But instead, they're implementing a rule that is unenforceable because you can't actually validate whether or not the majority of the project was written by an LLM. They just chose a suboptimal path and pissed a bunch of people off in the process, and their terms of service makes this out like some kind of copyright or security issue. It's just silly.
A more relevant quote is "to enable equal opportunities regarding the access to knowledge and education." from their mission statement. As a gemeinnütziger Verein (charitable association) their charta is legally binding.
well one can argue that many more people will write code with AI assistance than without it.. here i see the opposite: The wish of few outweigh the need of many.
Can one argue with a straight face that people will learn to code better by using a predominantly AI-based workflow? Or even learn by reading vibecode uploaded by someone else?
Hell, I'd argue that Vibecode without the prompt that generated it isn't even real open source.
Can you argue that you could learn to code better, copying and paste code from stack overflow? Who wants to learn, learns.. we did without Internet, and I still learn with AI.. btw, is in the "Satzung" from Codeberg e.V something about "learn to code?
> Can you argue that you could learn to code better, copying and paste code from stack overflow?
The learning doesn't come from copying and pasting code from stack overflow. It comes from
A) searching which code to copy and paste
and
B) Fixing the broken code that you just copied and pasted.
I suspect most programmers outgrow the "copying and pasting from stackoverflow" period of their development pretty quickly when they get frustrated by the fact that most of the stuff they were copying and pasting is broken or sub optimal in some way. If they prompt claude "write me a function", and then when it's broken they prompt "fix the broken function", they're learning (and doing!) a lot less.
Did you... read my post? The first thing I posted in this thread was a direct quote from the Satzung. Did you think "education" was about learning the history of the Ottoman Empire..?
I don't think this is a fair summary. I don't see this conflation in Codeberg's blog post. The words in quotation marks don't actually exist in Codeberg's blog post.
It seems that what they are saying is: In a free hosting platform where we share all the resources, it's not fair that some people using LLMs are using up disproportionately huge hosting resources to handle the unnaturally large amounts of resources they can churn out with these tools.
I'm a solo dev, and I couldn't find the judgement against solo developers or any slight against people who don't have a team. The concern is that some people don't gobble up an unfair amount of resources.
I personally have been moving to self-hosted Foregejo for ultimate freedom. But I also think it's fair for some providers have the freedom to take a bit more of a harsh (or based - depending on your perspective) stance against LLM generated code. I mean, a lot of people have been bemoaning the fact that the human-generated material has been increasingly hard to find in a sea of AI-generated material. It could be nice to have a corner of curated, human-made code still available on the web.