Correcting misinformation on social media with a large language model