Researchers from MIT and Harvard have developed a new way to evaluate whether advanced artificial intelligence (AI) systems, particularly large language models, actually understand the world or simply ...