“The hallucination rate in large language models like ChatGPT has been reduced to almost nothing over the past two years.”