It's easiest way to see what these "AI" models are doing by looking at the "art" models. The systems are very good at finding the source material to steal (and there are lawsuits filed), but the rest is just a merge/morph operation with no real understanding of the material. Sticking with the art model, if you tell it to "draw" a woman with red hair wearing a black dress, you'll get several reasonable representations. However, if you tell it to draw a Christmas parade, you'll get something that looks ok from a distance, but the people all have warped faces, or too many arms or some such issue. This is because the system doesn't actually understand the material. It's the same thing with the text models. You tell it you want a paper on the theology of bed bugs, and it'll dutifully go out and find a bunch of source material on theology and bed bugs and attempt to merge these concepts into something that "sounds right". It will result in a final product that is as non-sensical as the original input. GIGO. Now, if you take this initial technology and use that to train the next model on the concepts of "person", "dog", "car", "love", etc. you might get another step closer. However, there still isn't a reasoning engine in the mix. Until then, these toys won't be able to pass the Turing test. For all the fluff and thunder in the news, there are just as many stories of how easily these simple models can be tripped up, fooled and twisted. The true danger of "AI" at this point in time is how much people believe that it exists.

You can get a good visual reference to what these "AI" systems are doing by playing with the various AI "art" programs. It's much more clear there. The fancy part is in how it selects the source material to steal (and yes there are lawsuits filed). After that it's just merging those bits together. In images, it's just an extension of the old programs where they morph one image into another one. The chatbots are doing the same thing. They don't have anything resembling a thread of consciousness or even an understanding of the material. It's all just "find and merge".

I think that's covered in the Sprites section.