Every research project has a barometer on which they judge their research on. Mine is performance, and more specifically, loss as it's the best metric for understanding how my language model is learning the underlying dataset.