Estimating the cost-effectiveness of research — LessWrong