Domains of forecasting — LessWrong