• taladar@sh.itjust.works
    link
    fedilink
    English
    arrow-up
    25
    arrow-down
    3
    ·
    4 days ago

    I can’t speak for others but I simply hate that people keep telling us how amazing AI is yet not a single one of them can ever point to a single task completed by AI on its own that is actually of decent quality, never mind enough tasks that I would trust AI to do anything without supervision. I mean actual tasks, e.g. PRs on an open source repository or a video showing some realistic every-day task done from start to finish by AI alone, not hand-wavy “I use it every day” abstract claims.

    People like OP seem to be completely oblivious to the fact that reading code takes a lot of time and effort, even when there was an actual human thought process behind it, never mind when it might be totally random garbage. Writing code is also not nearly as much of a bottleneck as AI proponents seem to think it is. Reading code to verify it is not total garbage is actually much more effort than writing the same code yourself. It might not appear like that if you are writing in a low expressiveness language like Go or Java because you are reading or writing a lot of lines for every actual high level action the code takes that you need to think about but it becomes more obvious in more expressive languages where the same action can be expressed closer to 1:1 in terms of lines per high level action.

    • auraithx@lemmy.dbzer0.com
      link
      fedilink
      English
      arrow-up
      2
      arrow-down
      3
      ·
      edit-2
      4 days ago

      Why does it need to complete it on its own?

      With a human reviewer you can still do things a lot quicker. Code is complex so more the exception to the rule.

      Next time your stuck on an issue for hours stick it into deep research and go for walk

      • BombOmOm@lemmy.world
        link
        fedilink
        English
        arrow-up
        16
        ·
        edit-2
        4 days ago

        Any code reviewer will tell you code review is harder than writing code. And it gets harder and harder the lower the quality the code is; the more revisions and research the code reviewer needs to do to get the final product to a high quality.

        One must consider how humans will interact with this part of the program (often this throws all kinds of spanners in the works), what happens when data comes in differently than expected, how other parts of the system work with this one, etc, etc, etc. Code that merely achieves the stated goals of a ticket can easily produce a dozen tickets later if not done right.