Every day when I chat with one AI or the other. You have to pay attention because assumptions or mistakes happen all the time. The prompt(s) you give the AI probably have something to do with it, but I usually try not to write a novel as a prompt. It’s hard to get all the pertinent details in there and have it all make sense.
Maybe, but the transforner model may not capture all the intended meaning in a text.
I have looked, but I have been unable to find information on how good a statistical fit the model is.
They have a variety of models, trained for different purposes … ie they fit different sets of weights. That suggests that a universal fit is not achievable.
Noone can tell me if the fitted weights explain say 90% of the variance in any model fit.
What are they hiding?
I live in the Netherlands. The weather people actually think that predicting the weather a few days in advance is possible, even with the help of computers. Ha!They’re almost never right! Sometimes even when doing a prediction on the same day.
Too deep for me so I asked ChatGPT about your question. There was a 2 or 3 page answer that I asked for a brief summary. Here is the summary I got from AI.
Well, not surprised, you got nothing from AI… it waffled and admitted defeat.
The first thing that would be asked in academic circles is “how well does your model fit the data”
Why has the question not been asked?
It is true that it is strange model and it may be difficult to measure the fit, but there is no excuse for hiding it or not trying.
Yes it is difficult.
We are still getting excuses rather than answers.
The whole world is going to use this AI tool, and they cant quantify how well it works.
I found another one caused by the moon. A bit old, but interesting.
“On October 5, 1960, the U.S. nuclear command center NORAD received signals from its early warning radar in Thule, Greenland, indicating that a massive Soviet nuclear attack on the U.S. was underway—with a certainty of 99.9 percent.”