Before an AI acts on two sensible answers, can it spot that they disagree? A Seattle research community points readers toward a small, revealing test.
