Llm Lior and Ramy :Eureka: Human-Level Reward Design via Coding Large Language Models