Values in the wild: Discovering values in real-world language model interactions