How chtagpt has been trained over these billion parameters in the last 7 years? What was the model and process? Can we implement a similar train up model in our education curriculums