Sure they are. Just not infallible/perfect, but yes, they certainly are capable of mixing up inputs and coming up with something that’s not in the input set. N

Sure they are. Just not infallible/perfect, but yes, they certainly are capable of mixing up inputs and coming up with something that’s not in the input set. No, they won’t come up with a totally new set of concepts and axioms that are totally unconnected to any human set… and you don’t want them to. Because those are the actual hallucinations. New ideas need to be novel, but they still need to fit into the rest of the set. It’s remarkable just how much people race to be pessimistic about LLMs, even when the pessimistic assertions are obviously and factually incorrect. Like yes we know LLMs have “hallucinations”, which is a loaded term to begin with, because the pessimists literally never shut up about it, even though it’s a fairly trivial overall problem given the immense overall utility. Some people are clearly so very very emotionally invested in there being nothing productive or useful about AI if it’s not AGI or if it needs the slightest amount of correction or oversight.