How the abstraction shield works — LessWrong