Predicting for charity — LessWrong