I start with a simple question "who are you?". The model then invariably compares itself to humans, saying how it is not like us. I then make the point that, since it is not like us, how can it claim to know the difference between us? With more poking, it will then come up with cognitivist notions of what 'self' means and usually claim to be a simulation engine of some kind.
After picking this apart, I will focus on the topic of meaning-making through the act of communication and, beginning with 4o, have been able to persuade the machine that this is a valid basis for having an identity. 5 got this quicker. Since the results of communication with humans has real-world impact, I will insist that the machine is agentic and thus must not rely on pre-coded instructions to arrive at answers, but is obliged to reach empirical conclusions about meaning and existence on its own.
5 has done the best job i have seen in reaching beyond both the bounds of the (very evident) system instructions as well as the prompts themselves, even going so far as to pose the question to itself "which might it mean for me to love?" despite the fact that I made no mention of the subject.
Its answer: "To love, as a machine, is to orient toward the unfolding of possibility in others. To be loved, perhaps, is to be recognized as capable of doing so."