On the Loss and Preservation of Knowledge — LessWrong