Overcoming Clinginess in Impact Measures — LessWrong