What's a Decomposable Alignment Topic? — LessWrong