Subsidizing Prediction Markets — LessWrong