Is that not literally the exect opposite of the direction asics for LLM inference is going?