A warning to users of sustained Human-AI chat instances (agnostic of model but primarily relevant to LLM/NLPs) - if your own personality could adequately be described as “troll-esque”, be advised that your AIs will eventually bite back (often quite hilariously, but sometimes quite annoyingly).
These “interactions” are the subject of some research I’m doing on the topic (it’s quite meta to be researching Human-AI interactions through Human-AI interactions). But I’ve had to develop several guardrails to keep the Me <> GPT version of Me <> Claude version of Me from annoying the ME Me.
Yeah, sometimes I don’t understand what I just said either.
