A graphing calculator for all knowledge work is just a(n artificial) brain. It’s not hard to see how offloading brain work to another brain makes your brain less powerful.

I’m not saying you can’t be intelligent in a world of LLMs but people who use the calculator analogy underscore the fact that the LLM isn’t just eliminating the drudgery of one subdomain of mechanical problem solving — every problem is one an LLM can provide and answer for, including what you should learn, what problems you should solve, what your next conversation with the LLM should be about. So they do pose at least a threat of people truly offloading indefinite amounts of their own critical thinking to models, and there’s also danger that such people will in general still be competent enough often enough (due to LLM competence) that they are hard to screen out of key decision making roles or even from highly profitable roles.