>>90477679based on what i've used:
it's probabilistic, so it's highly error-prone
its model of reality is whatever path it chooses, and each "decision" is a mistake (probably) that gets baked in to every further mistake that follows
that's a fundamental/axiomatic flaw, possibly one shared by actual "intelligence"—but ours can experience non-model reality as a corrective (in theory); we can *die* of being wrong
"alignment" prevents this quality—reality as evolutionary brake—from evolving in ai
so it's really good at making weird porn, bloviating speeches, aesthetically uninteresting music, and jokes that weren't meant to be jokes, but it's shit for anything else
neuro is actually exemplary
