The central issue that confronts humanity is the misunderstanding of the exponential function. This is becoming crystal clear in this rapidly advancing digital age.
We have all seen images such as this:
For those who are into investing, this often is used to show how compound returns operate. Over time, which is the x-axis, the total output (in this case value owned) grows at a pace greater as shown by the y-axis.
Naturally, this is not limited to investing. It affects everything we do. When it comes to the digital world, because of the pace, we can see things happening in near-real time.
These days, it is hard to mention the digital world within quickly drifting into AI. This is consuming more of the Internet at, not surprisingly, a growing rate.
In 8 days, we will see the 3 year anniversary of GPT3. Think about how much things have changed in the last 36 months. If we are to extrapolate this out to November 2028 (another 3 years), it will not simply double what we saw.
This is due to the exponential factor, something we will discuss in this article.
To illustrate this point, let us cover something many people have heard about, Moore's Law. This was a theory opined by Gordon Moore in the 1960s, the CEO of Intel at the time. He postulated that the amount of transistors on a wavers were doubling every 24 months.
While that has fluctuated, dropping for a period to 18 months, it held fairly steady for 6 decades.
Moore's Law results in a 1000x over 30 steps. If we contrast that with AI training compute, we see a 1 billionx increase.
Here is what it looks like on a chart that Grok created.
No matter what numbers you ascribe as the starting point, we can see how things get rather large over the course of the entire process.
The situation gets even more mind blowing.
As we said, Moore's Law doubles every 2 years. The training compute for AI models is doubling every 4-6 months. That is why the numbers get so large.
However, there are other aspects to the AI stack that are following similar curves.
Here is a table that Grok generated.
Notice the cumulative impact listed at the bottom. According to these calculations, we are looking at a 10x every 12-18 months. That means we are operating at a 7-10x of Moore's Law just using these parameters. We are not even using a cost per unit basis which, due to declining compute costs, is making these much cheaper to operate.
Consider how this affects jobs.
There was a time, not long ago, where scaling meant adding more humans. Now, instead of scaling people we simply do that with GPUs. If we follow the curves from the data above, even with a slowdown, there is enormous impact.
This is why many believe that we are looking at 3-4 years before knowledge work is crushed. Cognition is moving at a pace not seen to humanity.
My guess is this trend remains at an accelerated rate throughout the next 5 year. As we enter the 2030s, we will have to reassess. It was a short time ago that would could reasonably forecast out 10 years, within a decent range. I am getting to the point where I think this is shortened to 5 years.
The exponential curves simply keep going.
