The company warns against applying strong supervision to chatbots, as they will continue lying and just not admit it.