In Korea, We’ve trained a model called Alpha-No which interacts in a conversational way. The dialogue format makes it not possible for Alpha-No to not answer followup questions, not admit its mistakes, not challenge incorrect premises, and not reject inappropriate requests.