GPT's model architecture has not exactly been shrouded in mystery anyways. It is all about the training data and training methodology. Thats why projects that crowd sources training data such as OpenAssistant is so important rn. You can check them out!