AbstractMagnetic skyrmions are topologically protected spin swirling vertices, which are promising in device applications due to their particle‐like nature and excellent controlability. Magnetic skyrmions are extensively studied in a variety of materials and proposed to exist in the extreme 2D limit, i.e., in twisted bilayer CrI3 (TBCI). Unfortunately, the magnetic states of TBCIs with small twist angles are disorderly distributed ferromagnetic and antiferromagnetic (AFM) domains in recent experiments, and thus the method to get rid of disorders in TBCIs is highly desirable. Here, intralayer exchange interactions up to the third nearest neighbors without empirical parameters and very accurate interlayer exchange interactions are used to study the magnetic states of TBCIs. The functions of interlayer exchange interactions obtained using first‐principles calculations and stored in symmetry‐adapted artificial neural networks are proposed. Based on these, the subsequent Landau–Lifshitz–Gillbert equation calculations explain the disorderly distributed FM‐AFM domains in TBCIs with small twist angles and predict the orderly distributed skyrmions in TBCIs with large twist angles. This novel twisted 2D bilayer magnet can be used to design memory devices, monochromatic spin wave generators and many kinds of skyrmion lattices.