Towards AI-assisted cardiology: a reflection on the performance and limitations of using large language models in clinical decision-making