How would you develop novel evaluation methodologies and metrics for large language models at Character.AI?
