You know how chatbots can do fine in short bursts, but then you ask them how many ‘R’s there are in “strawberry” and they act like they’ve got a concussion? For the British Medical Journal’s Christ…
  • swlabr@awful.systemsEnglish
    13·
    3 days ago

    Feh. True AI adherents know that it is simply time to claim that AI has outgrown measurement designed for human minds (e.g. mental impairment tests) and require new tests for machine intelligence (trenchcoat volume tests). Of course these tests are yet to be invented and will obviously require billions of your human dollars for development.

    • David Gerard@awful.systemsOPMEnglish
      10·
      3 days ago

      there were counterarguments that the authors anthropomorphised the LLMs inappropriately - which is a valid objection - and therefore these tests shouldn’t have been run at all. The answer to that is that the companies are marketing this shit by anthropomorphising the hell out of it and literally claiming these spicy autocompletes are on the path to artificial super intelligence. So of course cognitive tests are gonna be appropriate.

  • luciole (he/him)@beehaw.orgEnglish
    10·
    3 days ago

    “Slightly senile robot” is giving LLMs too much credit. My slightly senile homies rock in ways LLMs cannot fathom.