The paper discusses the importance of uncertainty quantification (UQ) in statistical learning algorithms, particularly in the context of atomistic machine learning for chemistry and materials science. The authors propose a practical approach called "direct propagation of shallow ensembles" (DPOSE) to address the challenges of UQ, including computational efficiency and ease of implementation. DPOSE combines ensemble modeling with weight sharing and minimization of the negative log-likelihood (NLL) loss, which reduces the need for post-hoc calibration. The method is evaluated on various regression benchmarks and atomistic datasets, demonstrating its effectiveness in providing accurate and well-calibrated uncertainty estimates. The paper also highlights the importance of proper calibration and the benefits of ensemble propagation for derived quantities, such as forces and total energies. Overall, DPOSE is shown to be a robust and efficient approach for UQ in machine learning models, especially for applications in atomistic simulations.The paper discusses the importance of uncertainty quantification (UQ) in statistical learning algorithms, particularly in the context of atomistic machine learning for chemistry and materials science. The authors propose a practical approach called "direct propagation of shallow ensembles" (DPOSE) to address the challenges of UQ, including computational efficiency and ease of implementation. DPOSE combines ensemble modeling with weight sharing and minimization of the negative log-likelihood (NLL) loss, which reduces the need for post-hoc calibration. The method is evaluated on various regression benchmarks and atomistic datasets, demonstrating its effectiveness in providing accurate and well-calibrated uncertainty estimates. The paper also highlights the importance of proper calibration and the benefits of ensemble propagation for derived quantities, such as forces and total energies. Overall, DPOSE is shown to be a robust and efficient approach for UQ in machine learning models, especially for applications in atomistic simulations.