Different time horizons for forecasting — LessWrong