Artificial Intelligence
Teaching AI to admit uncertainty
Published June 26, 2025Johns Hopkins researchers show how different “odds” can teach AI models to admit when they’re not confident enough in an answer, rather than risk a wrong answer