Scaling Laws For Neural Language Models Are Few Shot