Compared to humans, artificial intelligence (AI) language models like OpenAI's GPT-3 can produce accurate tweets that are easier to understand but also fake ones that are harder to detect, according to a recent study.