In the land of LLMs, can we do better mock data generation?