The authors propose a new differentially private mechanism for a large class of statistical tasks. The reviewers thought the paper was interesting, and the fact that the new mechanism gets around deficiencies of directly applying the exponential mechanism in certain settings is significant.