Progress and Prizes in AI Alignment — LessWrong