This is a linkpost to a summary of a 2-hour discussion I had with Max Heitmann, a co-founder of Aether, and Miles Kodama about whether large language models (LLMs) or LLM agents have beliefs and the relevance of this to AI safety.