It is a direct decendent of the same techniques used to help you type by predicting the next word on smartphones and chat programs.

People found they could improve those models using trained artificial neural networks to learn prediction patterns. These evolved into very large models that have to train for very long times on humongous amounts of data on a very large number of computers.

Papers on these have been criticized for showing little architectural innovation or cognitive insight., just scaling things up massively.

But the results certainly produce many hitherto obtainable aspects of human intelligent discourse, even it is impossible yet to tell if any cognitive model building is going on in that artificial neural network.