• Sludgehammer@lemmy.world
    link
    fedilink
    English
    arrow-up
    0
    ·
    3 months ago

    I can’t remember where I read it but someone said “LLM’s provide three types of answer: so vague as to be useless, directly plagiarized from a source and reworded, or flat out wrong but confidently stated as the truth.” I’m probably butchering the quote, but that was the gist of it.

    • Ranvier@sopuli.xyz
      link
      fedilink
      English
      arrow-up
      0
      ·
      3 months ago

      Hold on let me have chat gpt rephrase that for you.

      I’m not exactly sure of the source, but there was a statement suggesting that language models offer three kinds of responses: ones that are too general to be of any value, those that essentially mimic existing content in a slightly altered form, and assertions that are completely incorrect yet presented with unwavering certainty. I might be paraphrasing inaccurately, but that was the essence.

  • RubberDuck@lemmy.world
    link
    fedilink
    English
    arrow-up
    0
    ·
    edit-2
    3 months ago

    LoL… if you hear all the real estate moguls defending trump that “everyone does it, and if this is prosecuted noone in real estate will be able to make a dime” it is clear that the bot is trained on real life data.

    Maybe they can train a bot to find the actual fraud and hammer down on it. This is a place where the addition of AI to workers is able to massively increase the ability to audit complex tax schemes.

    Taxes for normal people should be pre filled by the IRS and just require a person to log in, verify and commit.