Trust, but verify (with better data): overcoming AI’s hallucination problem | Opinion