My speculation was not about LLM being an abstraction around CPU instructions. I don't think such a jump would produce a reliable code, so I hands down agree wi

My speculation was not about LLM being an abstraction around CPU instructions. I don't think such a jump would produce a reliable code, so I hands down agree with you. In fact, your proposal is on par with what I had in mind.