You still want the model to have the training on morality since users will want responses based on different systems of ethics.