These models are proclaiming near AGI, so they should be smarter than hallucinating an answer.