On the Value of Cross-Modal Misalignment in Multimodal Representation Learning | Read Paper on Bytez