Large Language Models Actually Haven’t Improved That Much

It’s all in the measurement, and the tricks therein. – Wired