• Rivalarrival@lemmy.today
      link
      fedilink
      English
      arrow-up
      1
      ·
      12 hours ago

      Define “grasp ideas”.

      They are beginning to be able to correlate images, sounds, text (in multiple languages). If we fitted them with other sensors (chemical receptors, accelerometers, etc) and feed them sufficient training data, they would be able to correlate those as well. I would call this correlative ability the “grasping of ideas”.

      Where they fail is abstraction. But, this is a failing of human intelligence as well. Some fully productive humans never develop more than a rudimentary capacity for abstraction, arguably less than LLMs have demonstrated.

      Don’t get me wrong: They’re at toddler-levels of actual intelligence and only simulate greater capacity by regurgitating what they’ve learned people like to hear. But, a hell of a lot of people are guilty of the same damn thing.