Atomistic simulations are widely used in materials modeling, and there is a recognized need to quantify the uncertainty associated with the predictions of these models. A major source of uncertainty originates from the choice of Interatomic Potential (IP). For classical empirical potentials, the largest source of uncertainty is a combination of structural uncertainty, i.e., uncertainty due to the functional form of the IP, and parametric uncertainty, i.e., uncertainty in the fitted parameter values. Quantifying the parametric uncertainty of these models reveals that most classical IPs are sloppy. Sloppiness is a phenomenon common in multiparameter models in which most combinations of parameters have very large uncertainty and only a small number of parameter combinations are tightly constrained by available data. Uncertainty Quantification in sloppy models is challenging for both technical and fundamental reasons. In this presentation, I give a practical introduction to UQ for sloppy atomistic simulations. I discuss the algorithmic challenges associated UQ in sloppy models and explain the relationship between parametric and structural uncertainty in sloppy models. Finally, I discuss how these issues carry over to machine learning models, such as neural network potentials, and give a general outlook for uncertainty quantification in atomistic modeling.