When large language models, or LLMs for short, produce outputs, several criteria are at stake, including not only overall response relevance but also coherence and creativity.