A universal potential for all-purpose atomic simulations has been pursued for decades, but remains challenging due to limitations in model expressiveness and dataset construction. Now, writing in the journal Science China Chemistry, the research team led by Professor Zhi-Pan Liu at Fudan University has developed a generalized global neural network potential (GG-NN) based on a newly proposed low-cost, high-accuracy High-order Pair-reduced Neural Network (HPNN) model and a comprehensive global potential energy surface dataset of 5.84 million configurations covering 83 elements. GG-NN achieves high-precision predictions for both energies and forces, offering the potential to significantly accelerate the prediction and design of molecular and materials systems.