Comment on Codeberg bans vibe coded projects

• TrickDacy@lemmy.world ⁨1⁩ ⁨day⁩ ago

  you’d have to be really stupid to give AI such unfettered access to your repository

  So 1) we absolutely know people are much stupider than that. and 2) does exactly that has widely been observed.

  original

  • Epp@lemmus.org ⁨1⁩ ⁨day⁩ ago

    Oh, without a doubt. They’ll get some low-hanging fruit, I’m sure.

    original
• JustEnoughDucks@slrpnk.net ⁨1⁩ ⁨day⁩ ago

  And many, if not most, of the people vibe coding projects on git repos are that stupid. People who have never coded and never used anything other than git clone. That is just a fact of many of the people who do it. They are the people who think they are tech geniuses because they once made an LED blink on arduino or made the millionth timer app.

  It is just a fact of who is flooding the most slop projects. There are plenty of people who use AI in subtler or more useful ways, but that is not what is the target of anti-vibecoding rules. It is the 1000 slop projects generated bloated repositories fully in 1 day that are killing bandwidth and storage limits for everyone else.

  original

  • Epp@lemmus.org ⁨1⁩ ⁨day⁩ ago

    That makes sense.

    original
• povoq@slrpnk.net@slrpnk.net ⁨1⁩ ⁨day⁩ ago

  No need to asdume that. Just by using LLM to assist with code writing, the human in the loop changes their commit pattern.

  original

  • Epp@lemmus.org ⁨1⁩ ⁨day⁩ ago

    “Suspicious commit pattern” sounds like a pretty flimsy basis for banning a human from the platform. If nothing else, this futile crusade will be entertaining for almost everyone; except for their support staff who will need to investigate and either defend or repeal the been each time there’s a false-positive, and a person is accused of wrong-think because yesterday they commit twice in an hour whereas they usually only commit once per hour.

    original

    • povoq@slrpnk.net@slrpnk.net ⁨1⁩ ⁨day⁩ ago

      Most LLM users don’t try to hide it and the commit pattern is a result of the LLM assisted workflow, thus without losing the perseived benefits of using LLMs it can not be easily hidden. And the pattern isn’t only about when but also what. You might be oblivious to it, but the pattern is very clear and not possible to mistake with regular code writing patterns in an overall project. It is harder to spot in individual smaller PRs though, but that isn’t what Codeberg’s new policy is targeting.

      And anyways, this isn’t a question of “wrong-think”, but simply of Codeberg not wanting to be associated with such low quality, likely copyright violating and resource hogging projects. The LLM using people have plenty of other code forges to use, so it is really a non-issue 🤷

      original

      • Epp@lemmus.org ⁨1⁩ ⁨day⁩ ago

        They haven’t had a reason to hide it before now. I wouldn’t be surprised if someone now “vibe coded” a suite of obfuscation tools specifically to avoid detection.

        Yes, that’s why I said it’s a matter of prejudice. Assuming the projects are low quality, copyright violating and resource hogging are all conclusions reached by prejudice. You’re right there are other options available; my comment is just an observation.

        original

        • -> View More Comments
• Zink@programming.dev ⁨1⁩ ⁨day⁩ ago

  All you need is another one of those global pre-processor directives! Then you’ll have commits that exactly match human output just like with the code itself!

  original

  • Epp@lemmus.org ⁨1⁩ ⁨day⁩ ago

    Technically, yes.

    “Make commits sporadically, using inexact terms, occasional typos and every once in a while, forget to include a file that then necessitates a follow-up commit of a single line that simply says ‘whoops’”

    original