Some time ago I formulated a paradigm that I called jolting technologies. Jolting technologies are characterized by an increasing rate of acceleration. And as one of the premier examples of jolting technologies, I brought artificial intelligence. At the time, there were not many data points available. I kind of had to leap to the conclusion.
16030335560000000
I started from a critique of the report published at the time, almost two years ago by Stanford University, where they talk about the two eras of AI. And I pointed out that rather than simplistically interpolating the data with two lines, broken at the knee and talk about these two errors. They should have instead embraced this interpretation of increasing rate of acceleration, ie of jogging artificial intelligence. And today we have another data point at the. Developers Conference recently held by Nvidia. They announced that the rate of doubling is now every two months in the power of AI applications, as opposed to the rate of doubling that. Two years ago, was four months in the report published by Stanford University.
16030335560000000
So let's go back and recap a little bit. In order to better illustrate what this means, and why we are talking about jolting technologies in general and jolting AI, in particular, and what the implications are for the future.
16030335560000000
We are accustomed to be seeing Moore's law being implemented around us in the various gadgets, whether our computers, mobile phones, smart speakers or, everything that the world is now containing, and that which surrounds us these gadgets are based on Moore's law, which has been formulated over 50 years ago, and it talks about our ability to design, electronic components that include an ever increasing number of transistors in a square centimeter or a square inch. And our ability is increasing. At the same cost, or even at decreasing cost. It is increasing, not based on the same technologies, but on a series of generations, so that our chips, have gone through different ways of implementing these transistors. Originally when Gordon Moore formulated. What we today call more slop. He only had a handful of data points. Actually, three data points, and he was extremely courageous. In, claiming that what he saw wouldn't be possible. Engineers, the world over. Have exerted every possible
16030335560000000
trick and smarts, and passion they had in order to prove him right now. Moore's law is not a natural law. It is a self fulfilling prophecy. And it doesn't follow a precise curve. But it is a statistical plot of this increase in our ability to create ever denser computational platforms, which in turn are correlated to better performance higher speeds, all kinds of programs and and products and applications that can be as a consequence, implemented. Now, a lot of people have been talking about the death of Moore's law. Oh yeah. Moore's law has died several times over. We could have felt of the equivalent of Moore's law for computers based on mechanical relays, or vacuum tubes. And these evidently, hit the wall of their maximum usefulness. We couldn't have computers based on those technologies with billions of transistors. And the same, the various ways with which we are designing and etching, and producing at massive scale, our chips, have gone through, generation after generation of
16030335560000000
improvements. And you can fully expect Moore's law to die. Again, several times in the future. But the engineers that are working on the problems that they are seeing every day. And they want to overcome in order to improve the performance of these computing platforms have many other tricks up their sleeve. And they are deploying these tricks. They may be wrong about the timeline precisely coming next year maybe it will come. Two years later, but statistically speaking, the power of our computers is going to keep increasing. Just yesterday, Apple Computer and nounce, sorry, it is not that old computer anymore, just Apple. Apple announced the new iPhone 12. They are designing the chip. That is then fabricated by either Samsung, or tsmc, and it is based on the newest five nanometer technology. Yes. This is the latest and Moore's law is still active. We are still reducing the size of our transistors we are still improving the way we are designing our chips. Apple's latest chip actually
16030335560000000
includes several subunits that improve the graphical capabilities that are specialized in AI and machine learning tasks. Now, if we are talking about Moore's law, and its doubling of the transistor count every two years. Then what did Stanford University say they looked at how specifically AI applications have been improving. And they said, starting at approximately, 2012. If AI applications, followed Moore's Law, we would have had. By this time in eight years, an increase of little more than 30 times in their power. Instead, what we observed is an improvement of 300,000 times. So they interpolated, the two. And they said, okay, the power of AI is increasing every four months now. That is fine, and you can definitely do that. But what I am saying is that instead of talking about two constant rates of acceleration. One every two years. The second every four months, what we should be talking about is a statistical curve, expressing the increasing grade of acceleration. So if you have a
16030335560000000
logarithmic chart, which is what typically we use, where on the vertical axis, you have orders of magnitude. 10 101,000 10,000 and so on. At equal distances, and an exponential curve will be a straight line on that kind of chart. When we are looking at an increasing rate of acceleration or jolting quantity. When we are looking at how we chart, the data points in order to then find the kind of function. the doubling great of that particular technology. We have an exponential curve. So here we are.
16030335560000000
Two years after Stanford University published their report, talking about a doubling of every four months. And we have Nvidia in their worldwide developer conference announcing artificial intelligence in dozens of different platforms and environments and application programmer interfaces and eco systems from embedded chips to data centers, and what they are saying is that they are observing the doubling of performance of AI systems to be every two months.
16030335560000000
So, as it looks like based on just a few data points. The law of jolting artificial intelligence is that of every two years, doubling the rate of doubling. So, we can expect. And this is the forecast that we can come back in 2022 and see if I was right. The power of AI applications to double every month by 2022. Let's put it in our calendars. And let's see if we can find data points to confirm the power of AI applications, doubling every month by 2022. And that is jolting AI, one of many jolting technologies, whose power is increasing, rather than just exponentially. In adulting great. You can go to jolting.co and sign up to be informed. When the jolting technologies seminar series is going to be available. It is already available on an enterprise level. And there are companies, both in Europe and in the Americas and elsewhere, who are training their mindset in order to recognize jolting trends, not only in AI but elsewhere too. And what my team is now busy with is to configure the
16030335560000000
seminar series in order to be able and make it available to anyone, including you, so that you can also train your mindset reconfigure it to understand and recognize the jolting trends, around you and take advantage of that. Thank you.
16030335560000000
| Word Count | Unique Lemmas | Characters | Lemmas Density |
0 |
0 |
0 |
0 |
| Modularity | 0 |
Influence Distribution | 0 % |
|---|
| Topics | Nodes in Top Topic | Components | Nodes in Top Comp |
0 |
0 % |
0 |
0 % |
| Nodes | Av Degree | Density | Weighed Betweenness |
0 |
0 |
0 |
0 |