Hot take: "thinking" isn't the right word for what LLMs do. Humans think. Models reason. He feels important that we keep these two things separate until machines do actually think.