A framework for thinking about wireheading — LessWrong