Large language models have become the public face of artificial intelligence, but a growing group of researchers and practitioners argue that these systems are nowhere near genuine understanding.