LLMs could consume vast amounts of energy and still fall short of our 20W low-tech brains, so let's start building with what we have rather than be blinded by the distant light of AGI