|
|
|
|
|
|
by rvz
5 days ago
|
|
|
> Codeberg banned projects that « mostly consist of code written by "generative AI"-tools ». Since you studied the entire policy, what is the methodology in determining a project that uses generative AI-tools or not that covers the ban? What does "mostly" mean here? Is there a percentage defined? Or did they look at sourcehut's Anti-AI policy and start from there? This is actually a good example since people can just fork existing "human made" software, (Free software permits you to do that) and run agents on it and hide the "slop" commits to circumvent that policy. Maybe someone would like in Rust and use an agent to do that. Or continue it as is but just adding new features quickly. > subsurface is not slop and is therefore not targeted by the codeberg ToS update. ...for now and it will eventually have more gen AI code anyway. subsurface already has generative AI tools committing to the code base and the presence of the CLAUDE.md guarantees that. The same can be said with Vim which has an AGENTS.md allowing AI contributions. |
|
|
Overall: there are humans performing the moderations. I trust them to have good judgement. This is not a witch hunt proposing to delete every repository where Claude has ever committed.
Personal interpretation: i think, from a very outside perspective (but that's what i'd do if i were in their place), that they'll inspect projects sorted by CI minutes consumed, then individually seeing if that's useful CI for useful projects (eg. fuzzing for tor or whatever) or if that's just slop CI for slop projects running countless steps on 100 commits an hour for software that's used by exactly 1 person (or even 0, who knows). From there, i'd start by failing the CI (and pushes) with a message inviting the user to read their mailbox, and send an email saying the user has been detected as malicious and should move away from codeberg, that existing code will remain available for a year, but no push/CI/issues will be permitted.
This will not remove 100% LLM-assisted project (which then again, was not the case), but might reduce codeberg resource usage by 10-99% depending on how bad the situation already is.
> run agents on it and hide the "slop"
I'm not sure how an agent would be able to hide that it's committing 100 commits an hour with 10k lines each, adding and removing tests as a mere inconvenience. But sure, let's assume so. That's just going against the rules, so the user may get banned.