By using hypernetworks, researchers can now preemptively fine-tune artificial neural networks, saving some of the time and expense of training.