Ok, bye bye [LWN.net]

Ok, bye bye

Posted Jul 24, 2026 1:00 UTC (Fri) by dskoll (subscriber, #1630)
In reply to: Ok, bye bye by mjg59
Parent article: Codeberg: Protecting our FLOSS commons from LLMs

Isn't relying on vendors who host models a little dangerous, though? That's like using a build machine that anyone can change whenever they want. They can also change the pricing on a whim. So something that works pretty well for you one day might be much worse the next, or much more expensive.

Yeah, I was somewhat trolling by saying that all you need to version-control is the prompt, but that was in response to OP's claim of a "100% AI" project.

Anyway. This is a pretty divisive issue, and IMO if Codeberg wants to refuse to host mostly-AI projects, that's its right. I happen to agree with Codeberg, but those who don't agree have plenty of other free options.

Ok, bye bye

Posted Jul 24, 2026 6:10 UTC (Fri) by Cyberax (✭ supporter ✭, #52523) [Link]

The world of AI development is in a weird state right now. There were attempts to just use prompts and keep the source code as an ephemeral artifact, but they are too impractical for large projects. It's like if every compiler run cost you $100-$10000.

And committing the entire dialog with the LLM is kinda like committing a recording of a GDB debugging session.

Ok, bye bye

Posted Jul 24, 2026 6:23 UTC (Fri) by mb (subscriber, #50428) [Link] (1 responses)

See, we can't even agree on what "100% AI" means, right?
For me: Thing comes out of AI agent -> It's 100% AI.
For you: Thing that comes out of an AI agent is reviewed without a change -> It's not 100% AI anymore.

The repository state is unknown (how much AI is in there?).
The platform thresholds for permissible amounts of AI is unknown or secret.

And under these circumstances I am supposed to agree to Codeberg's ToS?
Sorry, thanks but no.

>Isn't relying on vendors who host models a little dangerous, though? That's like using a build machine that anyone can change whenever they want. They can also change the pricing on a whim. So something that works pretty well for you one day might be much worse the next, or much more expensive.

Sure. But why is that your problem, the forge's problem or anybody else's problem? It's my problem.

Ok, bye bye

Posted Jul 24, 2026 12:33 UTC (Fri) by antacon (subscriber, #138885) [Link]

In a practical sense it's Codeberg's problem because while you can afford to generate code you'd otherwise not be producing you are flooding Codeberg with hosting costs. Multiply your impact by however many other users that do the same thing and those demands go higher, faster.

That in turn becomes every user's problem since they all enjoy free hosting currently but costs may eventually require paid subscriptions. The forge may shutdown if costs can't be kept manageable.

Here is the Codeberg 2026 budget - https://blog.codeberg.org/codebergs-budget-of-2026.html

I'd advise you to find LLM friendly forges in the meantime and set up remotes with them for your projects. At least this will be a good exercise for you in redundancies.

You can see the bigger brief of the decision here - https://blog.codeberg.org/protecting-our-floss-commons-fr...

You can also try joining the org to be more involved in these decisions - https://codeberg.org/Codeberg-e.V.