Humans are born with very high acuity (we can learn pretty much anything) but a tiny knowledge base. These two mental factors follow inverse curves over most of our lives, so over time we have reduced acuity but much more knowledge. *LLMs are "born" (current state) with fairly low mental acuity but with a very large knowledge base. In other words they have a ton of knowledge about the world (more than any individual human) but aren't yet very good at cogitating upon it. *The smart/stupid conundrum comes in because humans expect those with a large knowledge base to have a higher degree of mental acuity: experience has taught us that those who "know big words" (more generally know a lot about the world or have a large knowledge base) are also those who still retain a good deal of mental acuity. Thus when a LLM uses big words or knows esoteric concepts our experiences with humans inform us that these models will have a high degree of acuity--but that's simply not true at present. As LLMs gain more and more effective mental acuity this disjunction will reduce in scope but for now we're stuck with these apparently oxymoronically smart/dumb models due to their fundamentally different evolution.
