Mostly it was trained in dataset CNNArticles and ChatGPTWritedNewsArticle See here, so it may has a better behavior on classification about news articles,but now poor in science paper.GPT-3.5 and GPT4.0 are supported.
It only takes a few data,and needs more dataset to behave better in more text fields.You can take some articles from CNN official website or other news website and generate some news article by ChatGPT to test it.
