Обоснованное поколение

AbstractDespite recent success, large neural models often generate factually incorrect text. Compounding this is the lack of a standard automatic evaluation for factuality–it cannot be meaningfully improved if it cannot …...