Juicy bits of the mechanism, starting from the idea of an LLM conversation as a simulation of text-generating processes: > if the chatbot responds rudely, then

Juicy bits of the mechanism, starting from the idea of an LLM conversation as a simulation of text-generating processes: > if the chatbot responds rudely, then that permanently vanishes the polite luigi simulacrum from the superposition; but if the chatbot responds politely, then that doesn't permanently vanish the rude waluigi simulacrum. Polite people are always polite; rude people are sometimes rude and sometimes polite. The wide road is wide indeed that leads down to Waluigi. Hysterical.