Iterated Learning Improves Compositionality in Large Vision-Language Models | Read Paper on Bytez