I believe you four are talking about different things; the models are executed on very good "calculators" (if you want to call the GPUs that), but themselves are not very good at being used as calculators.

LLMs are sufficiently good hammers that people see everything as a nail, then talk about how bad they are at driving screws.