Computing neural nets isnt something a CPU is optimized for (matrix calculations), This tech is based on gpu optmized LSTM's, acka Transformers, they can run on cuda, reducing time. GPU's perform those calculations fast for huge matrixes.
Doing it on a CPU (they're no good in fat matrix caclulations) it takes forever.
And yes.. essentially those neural nets are extreme huge matrixes with some predictive math in between.