370
AI Completely Failing to Boost Productivity, Says Top Analyst (futurism.com)
submitted 9 months ago by to c/fuck_ai@lemmy.world
27 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[–] 40 points 9 months ago (2 children)

Solow Paradox

The main difference is that PCs actually worked as advertised, back in the day and the reason for this productivity dent wasn't a false promise from the start. Before AI the main use of computers was of a deterministic nature, meaning you get a directly reproducible outcome depending on the input. AIs (especially: LLMs) are probabilistic in nature, the output cannot be guaranteed to be correct, and it turns out just bolting on guardrails on top of the system is a band-aid. In practice, instead of getting a general-purpose intelligent machine which is capable of making autonomous decisions, you get a word predictor with an unlimited amount of possible failure modes.

  • permalink
  • fedilink
  • source
  • hideshow 2 child comments
    •
  • [–] 3 points 9 months ago (1 child)

    Good summary.

    I also find that LLMs are mostly good at the parts of my job that i enjoy or that are too important to entrust to something that is right 80% of the time.

    Tech boosters suggest using LLMs to draft emails. My job involves explaining mathematical ideas to (well educated) people who have less math training than i have. That part of the job is fun and challenging - why would i outsource a task that i like doing and am good at? “Everyday” emails are not important enough to stress about - i can dash them off as fast as i can write a prompt.

    The first time i used a computer instead of a typewriter to compose something, i knew that the world had changed. all that AI has changed is that i now doubt every article and picture i see.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 3 points 9 months ago

    Its similar for me as a software developer. I'm sometimes using LLMs to get alternative concepts or implementations to the task at hand or simply to refresh my memory about something with an example. This works well, because I'm already well aware of what I am trying to accomplish and so my prompts are precise enough to get a decent result. However generating directly usable code that meets my expectations just with prompts is really hard to do. There is so much fine tuning necessary that I'm faster just doing it myself.

    I don't see the technology itself as evil, there are some good uses if you know about the capabilities and limitations. What's evil is big corporations selling this technology to people who are not prepared to understand or handle the limitations.

    But personally I'm using it less and less these days. I don't want to take part in the massive environmental damage AI is causing and the thought of contributing to this mess makes me feel icky every time I think of using it.

  • permalink
  • fedilink
  • source
  • parent