If CLIP Could Talk: Understanding Vision-Language Model Representations Through Their Preferred Concept Descriptions - View it on GitHub