5 comments

  • fedpost 9 hours ago ago

    I think you hit the nail on the head.

    In my experience you can maintain a large codebase if you are exceedingly disciplined and are actually reviewing the lines that the LLM is trying to edit but beyond that you're just pulling a slot machine handle and hoping for good results. That works well with small programs and sometimes people get lucky with the repeated coin flips on larger ones but eventually it breaks down.

  • proc0 9 hours ago ago

    I've only been using some of the smaller but still capable local AI. I'm still skeptic on investing too much effort in the skills (my skills) needed for an AI-first workflow.

    I think one of the main pillars of engineering is being ignored, which is precision and predictability. Now I'm sure there is an engineering solution to this problem itself, meaning you can take the chaotic nature of AI workflows and build things around it to manage it, however this is not a trivial amount of overhead. I'm not convinced it is worth it, and in a way it defeats the purpose of AI to begin with, which is intelligent systems that would at least know when they are wrong and what their own limitations are.

    That said, AI already has a place in development in general just not as a main driver. If progress stays at this level, it's a matter of making it smaller, cheaper and local first, and then it will be an integral part of build tools and the like. So far I think the best use case is super-charging human driven development.

  • uain 30 minutes ago ago

    [flagged]

  • officially_mike 8 hours ago ago

    [flagged]

  • adam-margolin 8 hours ago ago

    [flagged]