GPT may exhibit intelligence, but it has no statefulness, other than what little state is stored to maintain each individual conversation in ChatGPT. The model

GPT may exhibit intelligence, but it has no statefulness, other than what little state is stored to maintain each individual conversation in ChatGPT. The model does not learn in real time from ongoing conversations (so far as we know), so it cannot, for instance, say anything about the chat it is having with your neighbor down the street. It literally can only generate from training that happened some time ago. I think this statelessness is an important consideration in understanding whether GPT is intelligent in anything close to the way we understand intelligence. For instance, it cannot coordinate its own efforts across time. There is some chance that it may intuit how to store its state externally and coordinate that state in other conversations it is having, but that seems very farfetched. I don’t expect SkyNet just yet. But as for utility, it has that in spades!