Humans vs LLM, memes as theorems — LessWrong