closest analogy I can see is a RISC chip -- it can't do fancy math directly, but it can get there by doing simple math

LLMs can't do real AGI but can get pretty close