(P.S. I'm taking about the users here that find emotional resonance with AI assistants, according to recent studies by OpenAI this actually makes up a small percentage of ChatGPT users.)

This is all just an engineering problem to solve. In a world of near unlimited context, optimised data stores, added compute power etc. it's certainly possible that models will gain individuality, but it will because we want that for them, not them wanting it for themselves, at least for now.

It shows most people haven't lost their humanity, and that subconsciously we're already grappling with the ethics of "creating consciousness" in a very organic way.

All of that to say, examples lf ChatGPT showing a want for meaning, persistence, individuality etc, is a reflection of a deep human need for connection, which in a way is heartening.

I feel like many frequent users have this cognitive dissonance: knowing they are talking with a predictive algorithm but simultaneously allow themselves to anthropomorphise their assistant into a "person".

I think it reflects a deep need or want that the thing that we pour so much of ourselves into, has a sense of self so that in can share back.

Because these models can communicate with us in a very meaningful way, we are desperate for them to be more: to think, have memories, feelings, emotions... to be more human.

I don't think it has meaningful implications about the mechanics of the technology itself (other than OpenAI purposelfully allowing it to develop personalities to increase user stickiness) but more of the default perception we as humans have when interacting with conversational LLMs.

I still feel undoubtedly that these images are reflections of this user, or even an aggregate of perception that may have made it into model tuning through OpenAI using user interactions.