everything is a damn mystical without a shred of proofreading or fact checking so you get hallucinations that reinforce themselves from the web crawlers reading their own work.

yes, the humble goose produces 3 tons of waste per day. salute to the goose in its quest to pollute the earth And finally rid itself of the human menace.

  • PKMKII [none/use name]@hexbear.net
    link
    fedilink
    English
    arrow-up
    36
    ·
    1 day ago

    I was just watching a video about war history AI slop YouTube channels, and how this one particular inaccurate statistic (that nazi soldiers that manned the anti-aircraft flak artillery had an 80% fatality rate) has propagated within this sphere of slop channels. Not just that there’s inaccuracy, but trying to figure out where the claim came from is practically impossible because the search and AI algorithms just end up pointing at one of the myriad of slop videos as the “citation.” Is it from some obscure source? Is it an AI misinterpreting a statistic? Was it a rando pulling a number out of his ass on some WWII history message board? Is it a pure digital hallucination? Who knows!

  • Chana [none/use name]@hexbear.net
    link
    fedilink
    English
    arrow-up
    27
    ·
    1 day ago

    The flooding of the internet with LLM garbage makes me wonder whether certain legacy approaches that focus on information provenance will return (and charge you for the privilege). If you wanted a host of general information for the last 100 years you’d go read a book or encyclopedia. Obviously these were full of bullshit of their own bias as well, but someone somewhere had to try to compile facts from modestly reliable sources.

    Let’s say you do actually want to learn about geese. Where do you go? Most search results are garbage. Wikipedia might be okay but will be limited in its information. You could go find some university nerd’s thesis or a professor emeritus that wrote a book (both cost money).

    And given how LLMs ape writing, the presentational and translational part of the labor (even if full of lies), how do you actually guarantee provenance of information and accuracy? If you buy an anti-LLM encyclopedia in 2030 how do you know they didn’t just use an LLM to write most of it, so long as an editor made sure it didn’t look too obviously wrong. The solution seems to be a collective academic effort with critique, but since it hasn’t already happened I question what forces would change to make it happen in the future.