Code for the ICLR'24 paper: "Visual Data-Type Understanding does not emerge from Scaling Vision-Language Models" - View it on GitHub