Abstract
Large Language Models (LLMs) seem capable of sustaining discussion on all kinds of topics. There are three possible reactions to this observation. We can attribute to LLMs a form of intelligence, which includes a certain understanding of the content of the discussion. We can deny them any intelligence or understanding. Finally, we can question the use of the term "discussion" to designate the process in question. This last point will be explored: what is a discussion between human interlocutors, and what changes are brought about when one of the interlocutors is replaced by an LLM?
