Making AI More Truthful: Doubt, Provenance, and Correction

New research explores adding 'expressed doubt,' 'provenance' (source tracking), and 'belief revision' to AI models to make their responses more believable and accurate, moving beyond uniform confidence.