We demonstrate that the proposed GPT2-based generative adversarial network framework can be used for simulating human-like brain signals and build the EEG classification model that can be transferred between different tasks and different subjects