AI may attain human-level soon — LessWrong