The pure ChatGPT output would probably be garbage. The dataset will be full of all manner of sources (together with their inherent biases) together with spin, untruths and outright parody and it’s not apparent that there is any kind of curation or quality assurance on the dataset (please correct me if I’m wrong).

I don’t think it’s a good tool for extracting factual information from. It does seem to be good at synthesising prose and helping with writing ideas.

I am quite interested in things like this where the output from a “knowledge engine” is paired with something like ChatGPT - but it would be for eg writing a science paper rather than news.