An International Manhattan Project for Artificial Intelligence — LessWrong