Linking model outputs to verifiable data or tools to reduce hallucinations.