x
Defining AI wireheading — LessWrong