You asked an hallucination generator about information, then asked it to hallucinate about the source of the hallucination, and then got satisfied when it hallucinated a reason as to why the rest were hallucinations?
AI is brilliant enough to make even its critics believe that it’s anything more than a slop machine, that’s already impressive.
I know. While I didn’t explicitly say so, I chose my words very carefully.
The one I tried was DuckDuckGo’s, and it provides links to what are supposed to be sources and it can also search the web for itself.
I checked each of its apparent sources and couldn’t find its claim in them, and when I asked it, it did its own search - something it has successfully* done in the past. It found more potential sources and could not find the claim in those either.
My phrasing was then “I was able to get it to tell me that it made that statistic up”. That doesn’t mean that it did make it up, or that it understood, but it was able to join the dots from the conversation and conclude that it, or a previous instance of it, might have made it up, and so one of its probable responses, especially given my inputs to that point, was an admission, and it was all but forced into giving that one.
* - A while ago, it actually found a source for an answer it gave when I asked it to do so. It had misunderstood (or misread, or some other more suitable word for what they actually do) the source and was wrong, but it at least found a page that was very likely where it got the information from in the first place.
(Ironically, I misread a source and was wrong about something very recently. I would say I know how it feels, but you and I both know that would also be wrong.)
You asked an hallucination generator about information, then asked it to hallucinate about the source of the hallucination, and then got satisfied when it hallucinated a reason as to why the rest were hallucinations?
AI is brilliant enough to make even its critics believe that it’s anything more than a slop machine, that’s already impressive.
I know. While I didn’t explicitly say so, I chose my words very carefully.
The one I tried was DuckDuckGo’s, and it provides links to what are supposed to be sources and it can also search the web for itself.
I checked each of its apparent sources and couldn’t find its claim in them, and when I asked it, it did its own search - something it has successfully* done in the past. It found more potential sources and could not find the claim in those either.
My phrasing was then “I was able to get it to tell me that it made that statistic up”. That doesn’t mean that it did make it up, or that it understood, but it was able to join the dots from the conversation and conclude that it, or a previous instance of it, might have made it up, and so one of its probable responses, especially given my inputs to that point, was an admission, and it was all but forced into giving that one.
* - A while ago, it actually found a source for an answer it gave when I asked it to do so. It had misunderstood (or misread, or some other more suitable word for what they actually do) the source and was wrong, but it at least found a page that was very likely where it got the information from in the first place.
(Ironically, I misread a source and was wrong about something very recently. I would say I know how it feels, but you and I both know that would also be wrong.)