Only because the Google paper crossed their desk did they even think to try it on matrix multiplications.
And then more or less by luck, it happened to turn out to be slightly better. If it had produced exactly the machine learning algorithm, it would have been less interesting.
The speed of the response makes me feel that the authors are extremely good at their domain, and that finding this algorithm was a reasonably trivial problem for them. The slowness of the follow up paper, I believe is that the authors understand that their mathematical domain is wildly far outside of what most people can easily understand, so they're going to take their time to write it up well, instead of drop a whole bunch of maths definitions that nobody else in the world understands.