Evaluating AI language models just got more effective and efficient

Assessing the progress of new AI language models can be as challenging as training them. Stanford researchers offer a new approach.