The issue is that gpt is trined on previously collected data and is not kept up to date. It might be able to tell you if an article from 2020 is fake news because it might know what actually happened that day from news articles from that time. But gpt has no idea of what happened today so it won't be able to tell what is real and what is fake. You'd need to use some sort of continuous online learning to do this properly. Obviously, it might be able to detect the real crazy stuff but it might even produce false negatives or real news if they are unexpected. For example, gpt probably has no idea that there currently is a war going on in ukraine, so how should it know whether or not an article about this topic is fake?