Modern day "AI" Large Language Models are best thought of, in my opinion, as Jellyfish AI. Or, if you prefer, cockroach AI. Meaning, dumb as shit. Not much more than the simplest of perceptual capabilities and not much more than the simplest of processing capacities. Entirely geared to a single type of sensory input and without the capability to adopt higher-order reasoning. This is why they are entirely black-boxes. Sometimes they get lucky and find a globally (or locally) optimal solution to a problem, but have no means by which to explain the reasoning.
