Language models hallucinate because their training and evaluation reward guessing over admitting uncertainty. Models are unable to say “I don’t Know” because they focus on accuracy. Guessing can impro