calling the llm model random is inaccurate