“Vin Singh states that large language models have a hallucination rate of about 10% in document review tasks.”