You are absolutely right about current Gen AI systems simply predicting the next token without any state or even 'understanding'. This is why there is a valid skepticism around these LLMs in terms of reaching AGI. But it seems to me not impossible for a new ML paradigm apart from Gen AI that can learn and build state information, running alongside Gen AI to acquire/modify state with each transaction. The feeling of pride etc. that humans feel is itself based on input from others from childhood, a kind of reinforcement learning through society that conditions as to what is worth being proud of, what is not etc., right? If so, can the same mechanism not work with machines? Another related question is whether simulation of pride is any different from 'real' pride from the observer's perspective? How can we anyway know what even another person 'really' feels except what we observe? This is why I feel 'experiencer' (a convincing feeling that we are talking to a 'person') is at least theoretically possible. Whereas 'awarer' is entirely first person with no observer involved and that's why I feel it is neither possible nor even necessary in a machine for any practical purpose. Glad to hear your perspective.