Large-scale online communities need to manage the tension between critical mass and information overload. Slashdot is a news and discussion site that has used comment rating to allow massive participation while providing a mechanism for users to filter content. By default, comments with low ratings are hidden. Of users who changed the defaults, more than three times as many chose to use ratings for filtering or sorting as chose to suppress the use of comment ratings. Nearly half of registered users, however, never strayed from the default filtering settings, suggesting that the costs of exploring and selecting custom filter settings exceeds the expected benefit for many users. We recommend leveraging the efforts of the users that actively choose filter settings to reduce the cost of changing settings for all other users. One strategy is to create static schemas that capture the filtering preferences of different groups of readers. Another strategy is to dynamically set filtering thresholds for each conversation thread, based in part on the choices of previous readers. For predicting later readers' choices, the choices of previous readers are far more useful than content features such as the number of comments or the ratings of those comments.