• kibiz0r@midwest.social
    link
    fedilink
    English
    arrow-up
    40
    arrow-down
    1
    ·
    1 day ago

    Tech bosses have well and truly lost it.

    Consider that:

    • Code is primarily to communicate from human-to-human, and only incidentally for computers to execute
    • A codebase that is 30+ years old has an absolute shitload of learnings incorporated into it, much of it very subtle
    • Languages are, in fact, different. So some things cannot be directly translated with exactly the same semantics, so devs will need to fully understand the intent and resolve ambiguities
    • A million lines per month is a lot of text for someone to successfully interpret and translate without losing any subtleties
    • FizzyOrange@programming.dev
      link
      fedilink
      arrow-up
      1
      arrow-down
      5
      ·
      5 hours ago

      Code is primarily to communicate from human-to-human, and only incidentally for computers to execute

      Uhm what? No. That is a stupid thing to say. It is primarily intended for computers to execute, but in a way that humans can understand.

      • Mr. Satan@lemmy.zip
        link
        fedilink
        arrow-up
        4
        ·
        4 hours ago

        It’s definitely for humans first and computers second. Compiled, machine code is for computers, everything else are tools so that humans don’t have to deal with machine code. An abstraction made by humans for humans to use.

        This is one of the issues I see with LLMs for code: instead of engineering and leveraging machine learning for optimizing specific problems, we’re now forcing text prediction engines to write human oriented text that happens to be a programming language.

    • SpicyLizards@reddthat.com
      link
      fedilink
      arrow-up
      26
      ·
      edit-2
      1 day ago

      And 1 million lines to truly review. Reviewers are used to their own heuristics based on common and critical mistakes to find errors. I reckon AI errors won’t follow familair patterns, making reviews even more tedious.

        • SpicyLizards@reddthat.com
          link
          fedilink
          arrow-up
          2
          ·
          5 hours ago

          Well, being the job and all, yes. But also set expectations, which there cannot be any based on the nothing job ad. Woudlt touch that with a 10ft pole.

      • kibiz0r@midwest.social
        link
        fedilink
        English
        arrow-up
        23
        ·
        1 day ago

        Yeah…

        Human mistakes tend to 1) look like mistakes, and 2) are surrounded by lots of hints that the author had trouble with that section of code.

        AI mistakes tend to 1) look like regular code, and 2) look just as confident and effort-ful as the rest of the code.