Learning Reward for Physical Skills using Large Language Model