FACTS Grounding: A new benchmark for evaluating the factuality of large language models