The AFP’s use of AI has been limited so far but the agency hopes the technology will help police identify money laundering and potential fraud

  • med@sh.itjust.works
    link
    fedilink
    arrow-up
    6
    arrow-down
    2
    ·
    1 year ago

    “It [AFP] noted that AI tools – including large language models (LLMs) – gave the AFP an opportunity to find useful information in large, lawfully collected datasets.”

    Literally everyone knows you can’t trust the output of these things. You can’t even get it to solve math problems with any degree of consistency without specifically training a math module. Who trained the AI module that finds connections in homocide cases?

    • SheeEttin@lemmy.world
      link
      fedilink
      English
      arrow-up
      3
      arrow-down
      2
      ·
      1 year ago

      And since money laundering is a math problem, when they train it on datasets with known money laundering being labeled, you’re saying it should solve with consistency?

      • med@sh.itjust.works
        link
        fedilink
        arrow-up
        2
        ·
        1 year ago

        No, I’m saying that the LLMs are good for one thing - finding the next most likely word. They can’t generalise, they can’t pick up special cases, and they really, really struggle with logically corollary. There’s no brain in there.

        Teaching a model that x = y won’t teach it the y = x. Even for discovery, that’s going to miss a lot.