Most of them. Given a set of incentives to publish as much as possible, one would expect the quality to decline at least linearly (assuming that the rate if actual discovery is constant).