• Yaky@slrpnk.net
    link
    fedilink
    arrow-up
    5
    ·
    13 hours ago

    Just yesterday, saw a fairly long article on how to properly train an LLM to send short email replies, and the time the guy wasted on prompts and writing the article would have been enough for a hundred responses.

    I still read HN, but I swear, they are in their own world at times.

  • drspod@lemmy.ml
    link
    fedilink
    arrow-up
    15
    ·
    2 days ago

    I stopped reading HN a long time ago but I was thinking recently that for all the hype about coding with AI, I’m not aware of a single worthwhile project that has come out of AI assisted coding. If AI was really that good at coding assistance you would think we would be flooded with new tools and solutions and hearing about them all the time.

    Does anyone know about even a single decent project that claims to be AI coded?

    • addie@feddit.uk
      link
      fedilink
      arrow-up
      4
      ·
      1 day ago

      Google did claim “half their new code” was AI-generated; obviously, take that with a pinch of salt, since they’ve a vested interest in promoting LLM.

      Speaking as a professional dev, about half of my lines-of-code consists of whitespace, opening-and-closing marks for the javadocs, and such matters as function, method and class definitions and their matching curly-close-brackets. My IDE generates all of that for me, but I dare say that I could use an LLM to do it as well, and then “half my code” would be AI-generated as well.

      My colleagues who are most enthusiastic about AI do turn in some right shit for code review; I suppose the best of it is over-complex and has confused error handling. They also tend to have about a hundred lines of what they’ve changed in the pull request description, and little or nothing about why. Github shows me what you’ve changed, I’m only interested in why you’ve done it, so that’s actually providing negative value by wasting my time having to read it.