Imagesource: http://visual6502.org/

The topic that Swen Kalski raises here, could easily trigger a controversy on HN that I would definitely like to follow.

As a proof of concept, Swen has implemented an ML approach in Assembly on a C64, and even if it is hard to believe at a clock speed of 1MHz, the runtime behavior of his regression implementation is absolutely acceptable. 

The point Swen now makes in his article, however, will divide minds. Swen's argument - in summary - is: With the many abstraction layers in modern operating systems, libraries and runtimes we slow down modern CPU's so dramatically, that we could have gigantic speed advantages with pure assembly implementations instead.

Swen digresses here and there into quite controversial theories, but I don't want to give away too much. His article hits a good point, and I agree with it at least partially. On the other hand, in my opinion, it is unfortunately absolutely illusory to develop modern software with all the capabilities, UX and UI of today, still completely in Assembly.

But how and if this dog bites its tail, you best find out for yourself. 🐕