Half-formed thought: the bubbles in VR, machine learning and cryptocurrency are partly explained by the decline in returns to Moore's Law, which means that parallelizable problems are cheaper/easier to solve than linear ones.

As returns from Moore's Law have declined, MULTIPLE processors continue to get cheaper (GPUs, etc, because of incremental new efficiencies in production, as well as quantity breaks), while FASTER processors remain relatively expensive

(Faster processors are also attaining gains by means of exotic techniques like branch prediction, making them vulnerable to intractable attacks like SPECTRE and MELTDOWN)

This has created a computational hammer in search of nails: "Come and show me your parallelizable needs, and I will solve them for you!"

Meanwhile, firms and researchers who want to solve linear compute-problems have been (relatively) stymied by slow, balky progress in their computing substrate.

(When a resource is abundant, people tend to think of ways of using it. During post-Soviet shortages, Russians had weird abundances. When all you could buy was forks, everything was made of forks!)

Today's computing applications landscape feels like someone on Sand Hill Road sat down with a whiteboard in 2005 and said, "OK, for the next 15 years, parallel computing will get cheaper and better. What businesses should we fund to take advantage of this?"

This has some explanatory power for the disproportionate focus on ML,VR and cryptocurrency (which also have other factors, eg crypto is key to laundering money in an increasingly oligarchic grifter's paradise).

That is, if you want to know why we're chasing these not-very-exciting goals, it's because we've spent a decade and a half looking for our keys under the computationally tractable light-pole.

Sign in to participate in the conversation
La Quadrature du Net - Mastodon - Media Fédéré

The social network of the future: No ads, no corporate surveillance, ethical design, and decentralization! Own your data with Mastodon!