Can you send a paper regarding that LLMs can build "reliable agents"?