While I'm definitely not in the "let's assign the concept of sentience to robots" camp, your argument is a bit disingenuous. Most modern LLM systems apply some sort of loop over previously generated text, so they do, in fact, have state.