We tried them as well and they did worse than a foundational LLM model which is saying something.