If you are a member of Codeberg e.V. please take the time to participate in the poll that was just sent out about banning vibe-coded projects on Codeberg.
Please agree to the proposal. Slop can live on GitHub.
The biggest question is, how will it be enforced, and how strictly? Will they rely on common LLM files, author signatures, or will they use LLMs themselves to judge whether a code base is probably AI-generated? What resort will there be to appeal false positives?
@csolisr it turns out to be super easy to detect LLM code, because the AI bros never STFU about it
@davidgerard @csolisr also in the rare cases when they try to hide it, a few quick questions about the code and either you get obvious LLM replies or they completely lock up about it
@csolisr How are the rules in most (all?) F/OSS projects against the use of proprietary code enforced, and how strictly?
Same here.
So it will depend on reports then - at least regarding proprietary software, it's typically 100% clear when a source code file is under full copyright, but with LLM-generated software it's easier to mask it, hence my worry about the enforcement (both in effectiveness, and in the overhead of applying it in the first place)
And how does one propose to tell if it's LLM-assisted code?
A code scanner? Like this Mary Shelly Frankenstein that's also 100% AI?
Good luck.
@crankylinuxuser @tante It is amazing to me how loudly the slop slingers tell on themselves while very clearly thinking that they're being subtle.
@crankylinuxuser @tante this gets brought up a lot, but my perspective is:
A) a lot of AI projects are proud of it and will disclose it somewhere.
B) the collaborative nature of Open Source makes it difficult for larger projects to hide process details indefinitely.
C) if the outcome of this is that projects using AI hide their usage and don't evangelize it, that is a significant win that we should take.
AI "detection" is unnecessary for a rule like this.
@crankylinuxuser @tante programmers are some of the biggest AI evangelists right now, and if a bunch of them start hiding their usage, that is a serious blow to companies like Microsoft and Anthropic who are counting on them to help drive a narrative about demand.
We take those wins; those are real wins. I want people using AI to hide their usage rather than to be public about it, I want it to be something that's culturally shameful to admit.
Are you shameful of spellcheck?
How about Markov chains and spam detection?
Are IDEs "AI"? Like function completion, or auto completing templating?
Or if its about training and recitation, are you against PID control loops? 3 neurons.
Or STT/TTS assistive technologies offensive to those disabled people?
Or is this how much data is in the statistical array? Is 1B OK, but 35B bad?
Cause this nebulous term " AI" doesn't actually mean anything. Has no scientific definition, and has changed hundreds of times even in my professional life.
Just how far up the statistical computing ladder do we go before its "evil"?
And LLMs are just tools. I am a tool user. Just so happens that I run my own #LocalLLM .
@crankylinuxuser @tante See, it's exactly like I said: detection isn't a problem because LLM users will tell you :3
No, I'm asking on the spectrum of statistical programming where is 'evil'? How many neurons? What kind of training functions? How much training data?
My 8bit Marlin 3d printer runs 'AI'. Sure, its when I run a PID training and then using to calibrate the hotend and bed, but it absolutely is learning software and AI.
@crankylinuxuser @foxyoreos @tante
If you read the actually proposed change
7. You must not share projects that mostly consist of code written by "generative AI"-tools (including services such as *Claude*, *OpenAI Codex*). Such projects having an unclear copyright status (see requirements § 2 (1) 1 and § 2 (1) 3) and furthermore have little safeguards to ensure that they do not include harmful code (c.f. § 2 (1) 5).
https://codeberg.org/Codeberg/org/pulls/1253/files
You will notice that it is quite clear what the ruling means and does not mean your PID controller or text to speach. The problem for Codeberg is specifically Copyright, although there are a lot more things to be concerned about when using even locally running LLMs or other such models.
@crankylinuxuser @tante Right, and what I'm saying is that when a project author shows up demanding ontological proof that AI is evil and for you to rigorously define the exact syntactic line between modern LLMs and other forms of AI - that's your detection. You don't need to run an AI detection service on the code.
You wait for the project author to completely voluntarily, unprompted (pun intended) volunteer that they use specifically LLMs, and then you ban those projects.
> And how does one propose to tell if it's LLM-assisted code?
You post that it's good to hide LLM usage and that people should be embarrassed about using them, and then your LLM-using projects will tell you that they're using LLMs during the resulting argument.
It's a surprisingly reliable detection method.
@tante ah yea, SlopHub, not Slopberg, thank you!
@tante I could care less if people host their own vibe coded projects. I programmed for years, but who am I to judge if their project is actually useful for someone?
As long as there's no AI PR spam I'm find with it. In fact, I really mostly object to AI PR spam. Add a filter or something to let people act responsible around this.
Outright bans? nah. live and let live. If you vibe code, it's your own loss imho.
@tante
Not sure. I am afraid that the project might head into a dogmatic direction.
I would want Codeberg to be a place also for art projects and crazy side activities of people who are not experienced coders. I think, it can be fine to „vibe code“.
Maybe self-assess the handwritten vs. AI delivered code via a filterable icon? An approach simulate to the LICENSE as a middle ground?
@anokasion müsstest email bekommen
@tante how do you prove something is slop of it doesn't have it written all over the code with em dashes and coauthored by commit signatures
@tante I will vote in favor of it. I just wish the reasoning for banning such projects would go beyond an unclear copyright situation. there are so many more good reasons to not want such projects on a platform like Codeberg.
@tante somebody buy slophub.com and redirect it to github
@tante Time to become a member, I'd been meaning to anyway. Thanks.
@tante haven't received it yet, it seems.
