Advancing Interpretability of CLIP Representations with Concept Surrogate Model | Read Paper on Bytez