In recent years, a lot of progress has been made in the development of machine learning potentials (MLPs) turning them into a standard tool for atomistic simulations in chemistry and materials science. 

While the first generation of MLPs has been restricted to small molecules with only a few degrees of freedom, the second generation extended the applicability of MLPs to high-dimensional systems containing thousands of atoms by constructing the total energy as a sum of environment-dependent atomic energies. Long-range electrostatic interactions can be included in third-generation MLPs employing environment-dependent charges. Only recently limitations related to the underlying locality approximation could be overcome by increasingly popular fourth-generation potentials, which are able to describe non-local charge transfer and multiple charge states. 

In this talk, the evolution of MLPs will be illustrated using high-dimensional neural networks, an important and frequently used type of MLPs. Some examples concerning their applicability will be presented.