Yes. Without something providing freshly crawled data into a prompt all that ChatGPT can do is predict based on patterns what is statistically the most probable result. But this vide is, I think, meant to be showing AIPRM first taking a URL from the user, then crawling that url, extracting the text from the URL, putting that into a dynamic prompt, and only then passing that prompt, containing the crawled data, over to ChatGPT.
Incidentally, some kinds of URLs are far easier to predict. Anything to do with most standard formats, for example has a standard pattern - press releases, news stories, all then to have very clear, easy to predict patterns.
Take something like a news page about one company acquiring another. If the URL is something that included [Company 1] acquires [Company 2] then we can predict tons of information on that page just from the fact those types of stories have such a clear and standard format.
One company acquires another. The one doing the acquiring is obviously the bigger, because it could afford to buy out company 2.
The page will summarize who company 1 are and what they are famous for, which is likely to be the same stuff they were famous for (or working on) just 2 years ago.
The page will summarize who company 2 are in the same sort of way.
They’ll usually stress the thing that both companies are famous for as a shared interest, since statistically that common interest is why company 1 will have bought our company 2. Especially if it is an interest that is a main focus for company 2, and something that company 1 are less known for but known to be expanding in (2 years ago).
There’s always a statement about how Company 1 expects to expand its capabilities in [insert shared interest that was a primary fame point for Company 2] thanks to the acquisition.
See how that pattern is so predictable?
So to show that this is not just prediction, we need to see details that could not be predicted from data 2 years ago.