• nymnympseudonym@piefed.social
    link
    fedilink
    English
    arrow-up
    4
    arrow-down
    12
    ·
    12 hours ago

    Reminds me of people who proclaim AI can’t code for shit.

    I wonder if they’ve used Claude or Codex in the last ~3 months.

    • d15d@feddit.org
      link
      fedilink
      English
      arrow-up
      7
      ·
      9 hours ago

      I have and while it seems to be doing a decent job at things I’m bad at it is far from usable for the things i’m actually good at

      • flying_sheep@lemmy.ml
        link
        fedilink
        arrow-up
        6
        arrow-down
        1
        ·
        9 hours ago

        Which is often a sign for something not being good at anything, only you not being good at assessing something you don’t have expertise in.

        See also: journalism about events you’ve actually participated in.

      • nymnympseudonym@piefed.social
        link
        fedilink
        English
        arrow-up
        1
        arrow-down
        8
        ·
        7 hours ago

        Yes exactly. At least for now, use the AI for things you are not expert at.

        I suck at graphic design so I get nanobanana to do that for me.

        For simple bugs or creating testcases or easy to describe new features, Codex does it because it does a good job 85%+ of the time saving me a lot.

        For really complex race conditions or vast complex enterprise applications… I’m literally an expert at debugging, a minority of human practitioners are better than me. Ofc I do that myself.

    • Flamekebab@piefed.social
      link
      fedilink
      English
      arrow-up
      4
      arrow-down
      2
      ·
      7 hours ago

      Whilst I’ve not used those tools, I’ve used their competitors and they constantly hallucinate methods, parameters, and write pointless unit tests (as in mock a return value and then test for that return value).