It’s fact.
Please see https://medium.com/data-science-at-microsoft/how-large-language-models-work-91c362f5b78f and specifically this image:
Again, this is on a VERY basic level. Most commercial LLMs have billions of parameters build on a corpus of the internet or subsamples.
In the end, it’s a lot of matrix algebra, weights, and regression functions. It’s not magic, but the hipsters, VC reps, and tech bros would like us to think otherwise.
Do you have any experience using AI platforms or are you just a critic?
That said, it's more than a jumble of wires and nodes and training and inference and preprogramming the same way an automobile is more than just a jumble of metal, rubber, and gasoline.