Researchers have developed a benchmark called the LAMBADA dataset to rigorously test how well AI models can leverage broader discourse context when predicting an upcoming word.