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

UC Berkeley researchers say large language models have gained "metalinguistic ability," a hallmark of human language and ...