Adapting to Means Testing — LessWrong