Conversation
How is a "no AI code" policy for a FOSS project even enforceable? Couldn't you just not disclose it? I've never looked at AI generated code (my only programming experience is some hobby Lua programming), is it that obvious?
2
0
0
@druid i think the codeberg one was for purely vibecoded stuff which is noticeable by rewriting stuff over and over, overly defensive programming (even the best models still do weird shit like checking if a number is greater than 0 twice on the same line) and being insanely "enterprise grade" (sticking hard to patterns, being overly explicit...)
1
0
4
@snacks So basically, AI written code is shitty and bloated?
2
0
1
@druid it's also down to human programmers learning the whole codebase and knowing what your constraints actually are i think. Ai will just learn about the minimum necessary to get everything running every time it tries to implement smth but yeah
1
0
1
Haelwenn /элвэн/ 
lanodan@queer.hacktivis.me
@druid Same way as with enforcing clean licences, or like discovering that someone lied to you either you know right away or it's after the fact, which is fine.
That said LLM-users are extremely loud and obvious in the way they act.
That said LLM-users are extremely loud and obvious in the way they act.
0
0
1
@snacks I hadn't considered that an AI might just overwrite a bunch of stuff a human had set up for the future, lol.
2
0
1
2
0
1
1
0
1
1
1
2
0
2
3
1
0
0
0
0
0
@dagda @druid @snacks Most agents can have their behaviour configured to an extent, like you can tell it hey, don't touch X, Y or Z file in the prompt and it will usually listen. Alternatively, if you sandbox it, you can also set some files to be read-only for the process.
Sometimes I do see the AI changing tests to get rid of a regression, though. If it's just told "make sure all tests pass", it won't really care about the spirit of the statement. Like most tools, it depends how it's wielded.
2
0
2
0
0
1