Supplementary Materials 575 A ViT-3B model details 576 The ViT model we use in this work is based on a standard Vision Transformer [ 7 ] model scaled to 577

We include screenshots of the reviewing tools we built to analyze model mistakes. Figure 3: A screenshot of the UI we built to review model predictions. We also flagged images as problematic if the ground truth label for the image was incorrect. 'siberian husky' label would be considered correct, whereas a prediction of'siberian husky' for an All siberian huskies and malamutes are also eskimo dogs. Sunglass and sunglasses are the same class (bidirectional).