Observational Scaling Laws and the Predictability of Language Model Performance

Neural Information Processing Systems 

Self-Consistency as language model capabilities continue to improve.