• wndy@awful.systems
    link
    fedilink
    English
    arrow-up
    4
    ·
    2 days ago

    Like, it’s not even THAT much better. I mean, not so much so that everyone should flood it lmao. The main plus was no restriction on tokens used, but that’s useless when it’s getting overloaded all the time.

    I would say it’s just barely noticeably better than the free tier of GPT. Which makes it a little annoying to go back but w/e.

      • wndy@awful.systems
        link
        fedilink
        English
        arrow-up
        3
        ·
        edit-2
        1 day ago

        Not people who can’t afford 100k to spin up their own servers. It’s going to be a game changer for AI startups and such though who won’t have to spend as much as previously thought.

        edit: Basically, numbers out of my ass, but it’s like they reduced the amount you have to spend to get chatgpt-level output from $500k to $100k. Amazing and all, definitely newsworthy, but uh… not directly relevant for us little folk, more about the ripple effects

      • Architeuthis@awful.systems
        link
        fedilink
        English
        arrow-up
        3
        ·
        1 day ago

        The 671B model although ‘open sourced’ is a 400+GB download and is definitely not runnable on household hardware.