The Princeton team developed a "bullshit index" to measure and compare an AI model's internal confidence in a statement with what it actually tells users. When these two measures diverge significantly ...