Ideas for AI labs: Reading list — LessWrong