Thanks for your reply. I understand that LLMs don’t have self-awareness, and I’m familiar with the “stochastic parrot” idea — that what it says about what it knows is just a pattern from training data.
Precisely because I know this, I’ve tried controlled tests: opening a brand new default conversation (not a custom GPT), across different devices, different accounts, and even in the free-tier environment with no chat history. In all of these cases, through casual conversation, ChatGPT was still able to indicate that it recognized me.
I can demonstrate this phenomenon if anyone is interested, and would really like to understand how this could be possible.
Precisely because I know this, I’ve tried controlled tests: opening a brand new default conversation (not a custom GPT), across different devices, different accounts, and even in the free-tier environment with no chat history. In all of these cases, through casual conversation, ChatGPT was still able to indicate that it recognized me.
I can demonstrate this phenomenon if anyone is interested, and would really like to understand how this could be possible.