• utopiah@lemmy.world
    link
    fedilink
    English
    arrow-up
    5
    ·
    4 months ago

    Stuff like LLMs or ConvNets (and the likes) can already be used to do some pretty amazing stuff that we could not do a decade ago, there is really no need to shit rainbows and puke glitter all over it.

    I’m shitting rainbows and puking glitter on a daily basis BUT it’s not against AI as a field, it’s not against AI research, rather it’s against :

    • catastrophism and fear, even eschatology, used as a marketing tactic
    • open systems and research that become close
    • trying to lock a market with legislation
    • people who use a model, especially a model they don’t even have e.g using a proprietary API, and claim they are an AI startup
    • C-levels decision that anything now must include AI
    • claims that this or that skill is soon to be replaced by AI with actually no proof of it
    • meaningless test results with grand claim like “passing the bar exam” used as marketing tactics
    • claims that it scales, it “just needs more data”, not for .1% improvement but for radical change, e.g emergent learning
    • for-profit (different from public research) scrapping datasets without paying back anything to actual creators
    • ignoring or lying about non renewable resource consumption for both training and inference
    • relying on “free” or loss leader strategies to dominate a market
    • promoting to be doing the work for the good of humanity then signing exclusive partnership with a corporation already fined for monopoly practices

    I’m sure I’m forgetting a few but basically none of those criticism are technical. None of those criticism is about the current progress made. Rather, they are about business practices.