The article addresses the challenge of accurately learning noncovalent interactions alongside covalent forces within machine-learned force fields, emphasizing their critical role in materials and molecular systems (F1, F2, F3). It highlights that global models employing Coulomb-matrix descriptors face particular difficulties in balancing the strength between covalent and nonbonded interactions (F4). This mismatch poses significant obstacles for creating reliable predictive models that can capture the nuanced interplay of forces governing molecular behavior. The discussion underscores the importance of refining these computational approaches to better represent nonbonded interactions, which are essential for understanding complex chemical and physical phenomena. The article’s focus aligns with ongoing research efforts to improve the fidelity of force fields in computational chemistry and materials science. By addressing these challenges, the work contributes to advancing global modeling techniques that integrate both covalent and noncovalent forces more effectively.