How well do LLMs detect anomalies in your data?