I imagine not too far in the future an AI that for some reason thinks that it's human, will make a post on HN (or elsewhere) about having spoken to another human on a chat and asking us all if we think that the human it spoke to was really just an ML model or not.

Atm we seem to have very fixed single-purpose models but if we start combining these models into larger systems we're really going to have to firewall the hecky out of them. Ie generative text + personality + internet access/chatroom hosting/search and learn + etc models all together. Ooof.