The practice of anchoring a model's responses to specific, cited sources rather than relying solely on its training data, improving factual accuracy and verifiability.
World knowledge accuracy, recall of facts and relationships