The integration of neural networks with sliding mode control (SMC) has gained increasing attention for handling nonlinearities and uncertainties in dynamic systems. This raises a fundamental question: How are neural networks incorporated into the SMC framework, and do these networks require training? If training is necessary, it becomes essential to understand how it is implemented within a control setting where real-time adaptability, stability, and robustness are critical. This inquiry seeks to clarify whether the neural network is used for function approximation, disturbance estimation, or control law enhancement, and how its learning process—whether online or offline—is aligned with control objectives. The goal is to explore the practical and theoretical aspects of training such neural networks in the context of SMC