Using Information Theory to tackle AI Alignment: A Practical Approach — LessWrong