NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples Paper • 2410.14669 • Published Oct 18 • 36