AI Safety Thursdays: Can AI solve its own alignment problem? — LessWrong