Exploring Lily’s world with ChatGPT [things an AI won’t do] — LessWrong