Soft capabilities such as conversational naturalness, knowledge accuracy, and genuine helpfulness are difficult to reflect in evaluation leaderboards but are most strongly felt in everyday use.