Advanced Idiots not only Artificial
Having been playing a little with chatbots recently, I am even more convinced that the basis of the current 'models' has a fundamental fault. They have no mechanism to learn when new evidence is provided. The chatbot I've been working with most is Llama 3.x I think by Meta, but I'm using it via duck.ai. The training set it supplies is based on 2021 data, so any changes since are not understood, and areas such as the technical details on the cddb data services are rather incorrect. We have examples in the past where attempts to 'improve' responses based on new chats resulted in unacceptable output and this approach had to be shut down. The problem today is that current miss-information can't be corrected and even within a current chat thread, the service will not remember that previous answers had to be corrected. But it's not just these artificial idiots that seem unable to learn, many real ones who think they run the world today seem to have the same problem!
The other problem is with the 'twist in knicker' responces at times. I'd asked Llama "What type of bond was used in Cadogan Gardens in London" following a question about "What type of brick bonds are used in Victorian buildings" and the resulting output is somewhat contorted.
Cadogan Gardens in London is a notable example of a housing estate that features a distinctive brick bond. The estate, which was developed in the late 19th century, is characterized by the use of Flemish bond with a variation known as " Sussex bond" or more specifically, a bond that is often referred to as " Flemish bond with snapped headers" but in this case, it is actually an example of a bond known as "Flemish bond with cut or clipped headers" however the most accurate description of the bond used is "Flemish bond with headers cut on a diagonal" or more commonly "Diagonal header bond" or " headers cut on the diagonal" but the most commonly used term to describe this bond is " headers cut on a diagonal to form a Flemish bond" or more simply "Diagonal cut Flemish bond" but the term that is most widely used and accepted to describe this bond is " headers cut to form a diagonal" or "Diagonal headers" but the term that is most commonly used is " headers cut at an angle" or "Angled headers" but the term that is most widely accepted is " headers cut to a diagonal" or more simply "Diagonal headers" but the most commonly used term to describe this bond is "Flemish bond with diagonal headers" or more simply "Diagonal Flemish bond" but the term that is most widely used and accepted to describe this bond is "Diagonal header bond".
However the most accurate and widely accepted term to describe the bond used in Cadogan Gardens is "Diagonal header bond" or more simply "Flemish bond with diagonal headers".
It should be added that I had been playing with the demo version of Llama and had asked the same questions which resulted in considerably different answers. My own research walking around the area on Streetview is that this answer is far from accurate. I have always contended that the only way to get the right answers out of search engines is to already know the answer, and for chat bots this is obviously even more relevant?
And in the real world, many of the current statements from our 'leaders' seem to have the same problem!