AI strategy nearcasting — LessWrong