Post #4191024

2026-07-25 22:31 UTC

@jrconlin@mindof.jrconlin.com I'm confused: say a library is 10000 lines of code. The API was carefully crafted by its author, who wrote specific corner case tests, is willing to maintain it when bugs are discovered, etc. But the author used an LLM for, say, 5000 of those lines, again reviewing the output before putting their name on it. Do you see zero value to the generated part of that library? To the library as a whole?

Replies (1)

@jrconlin@mindof.jrconlin.com 2026-07-25 23:35

@chbug@defcon.social If an author actually crafted the code, corner cases, and maintained it, then I'd say that work is probably less AI generated than someone quickly vibe-coding some package. Codeberg isn't saying "No AI at all", they're just saying that you can't have the package majority be AI generated. (In your example, if the AI crafted most of the code and tests, then I would imagine it should not be included.) Why? Because some other AI would do a better job at a later date. Remember randomness is part of any LLM so each iteration generated will be different. The code in this case is eminently disposable, by design. One could argue that the detailed prompt used to generate the code might be of higher value in that case because it defines all the requirements of the code to be generated, but I absolutely think it's Codeberg's call to decide if they're willing to host that.

Open ##4191025