• linearchaos@lemmy.world
    link
    fedilink
    English
    arrow-up
    10
    arrow-down
    1
    ·
    5 months ago

    Yes, and no.

    The tech is absolutely astounding. Somebody posted a random idea on Facebook and it caused me to make it have a conversation between Fred Rogers and Steve Irwin and it absolutely nailed it. I’ve had it look at pictures of memes it got details like one person looking at another person.

    The power wise, it’s rather unsustainable. There’s a real cost associated with each one of these queries were running and the price to train it with all the data. There are many jobs at which it makes financial sense to pay for it is a service but the vast majority of work or sending off to AI is nothing that anyone is willing to pay for.

    We’re in an AI bubble because we can only make queries against AI as long as Microsoft and Google decide that it’s in their plans to allow us to do it for free.

    • postscarce@lemmy.world
      link
      fedilink
      English
      arrow-up
      5
      ·
      edit-2
      5 months ago

      You’re forgetting about models that are open source and able to run locally. Llama3 is not the best model, but it’s still very useful and will continue to get better along with the top closed source models.

      • aesthelete@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        5 months ago

        This round of technology reminds me of when Google was the shiniest thing on the block, and everyone was trying to cram a site specific search engine into their website.

        This resulted in open source projects such as Lucene, which is incredibly useful, but is not in reality anything like Google. Over time interest in these projects faded and now they’re just another pretty optional component of a website (many sites just use SQL queries rather than a search engine).

        I think chatbots are pretty similar. The premier versions of these things cost way too much to run to be practical for most sites, so they’ll play with the scaled down, easier versions for a time before abandoning the functionality entirely over time unless it’s found to be actually useful.