“…From this perspective, the actual developments of first-principles machine-learned interatomic potentials are probably most exciting. 81,82 Explicit AIMD data 43,44,56,58,66,68,69,335,368,369,[459][460][461][462][463][464] has long been used to validate and improve implicit solvation methodology. If machine-learned interatomic potentials allow to generate comparably accurate, but orders of magnitude longer trajectories and in larger simulation cells, then this will be an invaluable asset that might even ultimately enable to validate and refine implicit solvation schemes for application outside the domain of ab initio thermodynamics, notably the modeling of kinetic reaction barriers.…”