Self-supervised learning can enhance model robustness and uncertainty estimation. This paper shows that self-supervised methods, such as predicting image rotations, improve robustness to adversarial examples, label corruption, and common input corruptions. They also significantly improve out-of-distribution (OOD) detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research. The paper demonstrates that self-supervised learning can improve robustness to adversarial examples, label corruption, and common input corruptions. It also shows that self-supervised learning can significantly improve OOD detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research. The paper also shows that self-supervised learning can improve robustness to label corruption. It demonstrates that self-supervised learning can improve OOD detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research. The paper further shows that self-supervised learning can improve robustness to common corruptions. It demonstrates that self-supervised learning can improve OOD detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research. The paper concludes that self-supervised learning can improve model robustness and uncertainty estimation. It shows that self-supervised learning can improve robustness to adversarial examples, label corruption, and common input corruptions. It also shows that self-supervised learning can significantly improve OOD detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research.Self-supervised learning can enhance model robustness and uncertainty estimation. This paper shows that self-supervised methods, such as predicting image rotations, improve robustness to adversarial examples, label corruption, and common input corruptions. They also significantly improve out-of-distribution (OOD) detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research. The paper demonstrates that self-supervised learning can improve robustness to adversarial examples, label corruption, and common input corruptions. It also shows that self-supervised learning can significantly improve OOD detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research. The paper also shows that self-supervised learning can improve robustness to label corruption. It demonstrates that self-supervised learning can improve OOD detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research. The paper further shows that self-supervised learning can improve robustness to common corruptions. It demonstrates that self-supervised learning can improve OOD detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research. The paper concludes that self-supervised learning can improve model robustness and uncertainty estimation. It shows that self-supervised learning can improve robustness to adversarial examples, label corruption, and common input corruptions. It also shows that self-supervised learning can significantly improve OOD detection, even surpassing fully supervised methods. These results suggest that self-supervision is a valuable tool for improving model robustness and uncertainty estimation, and should be considered as a new axis for evaluating self-supervised learning research.