“Thinking” by guessing what words to say next based on words we’ve previously heard might actually help find a good idea — and it’s also how know-nothings get through work meetings, and how people come to think they know stuff they really don’t, and how they internalize the stupidest notions. I am starting to think that in today’s environment, high cognitive skills are an actual risk factor for stupidity, and that learning words without learning a model of what they refer to is one big part of the problem.

Source: LLMs aren’t world models

Well worth the read.