I'm just commenting, and have some basic understanding. It's good to start thinking about it but I seriously doubt GPT or anything that can be built with it can have a direct existential risk, meaning it would somehow free itself and go rogue. Otherwise I don't know how it would be an existential risk (outside of massive impact in society, but that's technically a different problem). My argument hinges on not being able to accidentally stumble upon AGI without a theory first. At least not an AGI that thinks and has agency like a human, with selfish goals.