So if you fine-tune a model with your own data... you get answers based on that data. Such a groundbreaking revelation