Catastrophic Risks from AI #3: AI Race — LessWrong