What the skeptics seem to get hung up on is the lack of common sense. These LLM’s are trained on language describing the world, not on the world itself. They lack the lived experience that we take for granted. So it is trivial to ask them common sense questions and see them fail amusingly, but that is unfair because they haven’t been taught the world, only a linguistic map of the world, and the map is not the territory. Also, we don’t even know what map they learned, because the only way to know is to ask a million questions, like an early seafarer discovering the shape of a foreign continent beach by beach.

LLM’s are not people, they lack common sense, but they understand and can reason about what they are trained on. That is exceedingly powerful and very useful even at today’s level of ability, so products built on top of this technology are going to transform everything. The trick is boxing it in and only making it do things it can, so the art of LLM product development will have to become a whole subfield of software engineering until the LLM’s develop to the point where their map of the world is close enough to the world itself.