The degree to which a language model generates factually accurate information rather than false or hallucinated content.