Why keep a diary, and why wish for large language models — LessWrong