• FizzyOrange@programming.dev
    link
    fedilink
    arrow-up
    8
    ·
    edit-2
    3 days ago

    is having to pick apart someone else broken code.

    I agree, but also I do find that AI’s broken code is generally waaay less annoying to pick apart than my colleagues’ code. I’m not sure exactly why. Probably partly because it’s better at commenting code and naming variables so it’s easier to follow?

    I think also partly it’s because reviewing other people’s code is usually done during code review, where you can’t just directly edit the code to fix it - you have to start a conversation convincing them to do it differently. That’s quite annoying and doesn’t happen with AI generated code.

    • sturger@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      11
      ·
      3 days ago

      I still don’t understand why we’re using humans to review AI code. Shouldn’t AIs be reviewing the code?

      We’re letting AIs do the fun part (coding) and forcing humans to review (the worst part) reams more janky code.

      • greenskye@lemm.ee
        link
        fedilink
        English
        arrow-up
        7
        ·
        3 days ago

        AI’s get the fun part of everything right now. AI gets writing, humans get editing. AIs get drawing, humans get fixing hands and details, etc etc

        • Beacon@fedia.io
          link
          fedilink
          arrow-up
          2
          arrow-down
          1
          ·
          3 days ago

          What are the results if you take code written by one brand of ai and then have another brand of ai review it? Like use chatgpt to write code, and then ask copilot if the generated code has any errors and will work as intended?

          • NoForwardslashS@sopuli.xyz
            link
            fedilink
            arrow-up
            4
            ·
            3 days ago

            I don’t know, if I put my hand in a fan and then put that mutilated hand into another brand of fan, do you think that might fix it?

          • FizzyOrange@programming.dev
            link
            fedilink
            arrow-up
            2
            ·
            3 days ago

            Interesting idea, I’ve never tried that. I feel like it wouldn’t be a silver bullet but you might get slightly better results I guess.

    • Ephera@lemmy.ml
      link
      fedilink
      English
      arrow-up
      3
      ·
      3 days ago

      Huh, I feel the complete opposite. Human-written code follows some sort of causality, for example when you see complex code or a strange detail change, they had a reason to write that and they likely tried other solutions first. With AI-generated code it feels a lot more like I have to rate each changed line in isolation, which is exhausting.

      But yeah, I don’t know, we don’t typically do code reviews. I’ve only been in that situation so far, when I had significantly more knowledge of the project and language, so there were rarely discussions beyond trying to teach them.

    • Beacon@fedia.io
      link
      fedilink
      arrow-up
      2
      arrow-down
      1
      ·
      3 days ago

      I’m not a programmer so i don’t know if this makes sense, but I wonder if it’s easier to retool ai code because ai code is janky in a similar-ish way most of the time, while human code is janky in different ways all the time? Whadda ya think?

      • Windex007@lemmy.world
        link
        fedilink
        arrow-up
        5
        ·
        3 days ago

        I disagree with the premise.

        AI is good at making things that LOOK right. Pictures. Words. Whatever. Actually makes errors harder to find IMO.

      • FizzyOrange@programming.dev
        link
        fedilink
        arrow-up
        2
        ·
        3 days ago

        Yeah definitely could be. I also think when AI gets things wrong it gets it so obviously wrong you have to delete it and do it yourself (and not worry about offending someone). It rarely seems to make the same kinds of trivial mistakes humans do (like copy/paste errors for example). It either does a pretty decent job that’s easy to fix up, or it totally fails and you do it yourself.