So I decided to do a ChatGPT search for your question to Google. Not a bad answer for less than one second.
Firstly, thank you.
I was toying with the idea of opening a thread discussion on ChatGPT but it's not really a Camino topic and I figured that if I opened it then it would get closed and probably deleted rather quickly. It looks like there is some interest in ChatGPT on this forum so lets go.
I see that you have characterised your interaction with ChatGPT as a "search". This is also the characterisation used by the recently departed poster.
Of course, everyone is free to use words and language in any manner they like but usually we want to communicate our ideas when we use language and so it is useful to agree the meaning of words that we use. I can, of course, call a monkey a wrench but that doesn't help the people looking for ways to undo a nut that is screwed on a bolt.
For the next bit I am going to completely anthropomorphise ChatGPT and so apologies in advance for anyone who has an issue with that.
It is correct that ChatGPT is trained on a huge aggregation of data found on the Internet and from other sources but the output that it produces is not a precis of that data. It can best be characterised as directed word selection or in more understandable (anthropomorphised) language, an opinion.
The output produced is completely made up. This can be hard to see and it is why people often, mistakenly, think that it is operating like a search. This is especially apparent when ChatGPT's opinion aligns with our own. The mechanics behind ChatGPT are a bit easier to understand when we look at similar tools that use different media from language. I did a quick search for AI image generators, found one called Craiyon and had it create the image below by typing in a few terms. It is important to understand that this image tool didn't search the Internet to find an already existing image that matched my words but rather it generated a completely new image for me using my words and it's training images (that were undoubtedly taken from the Internet).
Leaving aside the low quality of the image, I didn't want to waste time on it as it is just an illustration. The words that I used were "three hikers on the Camino". I had to adjust my words because I initially typed in "three pilgrims on the Camino" and all its opinions had people dressed in medieval garb.
While it can be harder to spot that ChatGPT is making up stuff when it produces its output opinion it gets a lot easier when you know how to trip it up. When I last used ChatGPT a couple of months ago I noted that it was trained using data that was available up until 2019 (I now see that it is trained up until 2021) and so if you want to see how it really works then ask a question that requires data that is not available within its training set (after 2021).
Now, let's address the value of ChatGPT's opinions. How would you value the opinion of a friend or acquaintance who constantly scanned the Internet looking for ideas and chose to base their opinions on the most frequently occuring posts that they read?
Sometimes your friend's opinions would be useful, especially when the majority of Internet posts on a subject were saying reasonable things but on the other hand when your friend was basing their opinions on the latest nutty conspiracy theory that someone had made up then it might be time to stop listening to your friend's opinions.
ChatGPT has some built in rules to help with this but basically what it bases its opinions on is the most numerous posts it can find on a particular "subject".
Note: Currently I don't think that ChatGPT has semantic knowledge and so it doesn't know "subjects". It is much simpler and relies on statistical word association.
Okay, feel free to use the output of ChatGPT as you wish, many people do and it can be useful when used with thought.
BTW, the scary bit about these tools isn't what they are capable of now but rather the speed with which they are improving. Soon, my criticisms of ChatGPT will be overcome. Smiley face or sad face according to your own thoughts on this matter.
Edited: 7:13pm 21st May 2023 to remove an incorrect reference to another member.