MIT Teaches AI Models to Say "I Don't Know

Researchers at MIT have solved a dangerous problem in AI: models that sound confident even when they're guessing.

Their breakthrough helps AI admit uncertainty, making it safer for use in medicine, finance, and law.