We wanted to see if a large language model had the same sense of meaning of word combinations, so we built a test that ...