The ‘Thinking’ AI Breakthrough

During Intel’s long stretch of breakthroughs in the 1990s and early 2000s, it released new, faster chips every 1 to 2 years.

It was an incredible feat, and rapidly swept us into the internet age.

OpenAI, creator of ChatGPT, was moving at a similar speed early on. A new, wildly improved model every 1-2 years. But recently the company has been releasing a new model every 3 to 4 months

image 1

Additionally, its latest models, o1 and o3, are new animals. They don’t instantly output an answer based on training data. These ponder, consider, review, calculate, and then give an answer.

You could say this type of AI is the first truly “thinking” one.

Thinking AIs are new and rough around the edges. But they’re already making incredible progress.

OpenAI’s o1 recently scored 133 on Mensa Norway’s IQ test.

image 2

Source: TrackingAI

Now, it’s important to realize that this IQ test is probably in o1’s training data, meaning it should technically know the answers. But you can see just how large of an improvement the new o1 models (in yellow) are over previous ones.

In an offline IQ test, created specifically to test AI, OpenAI’s o1 Pro scored a 110. And those answers were not in the training data (you can see all the IQ results here).

Recently, the brand new o3 model scored 2,727 on Codeforces, a competitive coding test. That places it as the 175th-best competitive coder on the planet. Incredible.

These breakthroughs are a big deal. We are about to enter a stage of major disruption. Soon these models will be smart enough to handle complex tasks, and they will begin to compete for coding and some other white-collar jobs. It’s a disturbing idea, but seems inevitable based on current trends.

For now, these thinking models are expensive, slow, and limited in availability. But as they come down in price, speed up, and the market begins to understand how to utilize them, things are going to get weird for a while.

The Snowball

Each new model helps train and build the next model. As the models become smarter, this creates a compounding effect.

And of course, new breakthrough techniques are being discovered by hundreds of worldwide teams researching AI. Some are kept secret, but many are being shared in research papers and through open-source code.

The NVIDIA GPUs used to build these models are also rapidly advancing. As the latest hardware is deployed in massive data centers, the pace of acceleration will only increase.

In other words, we’re about to enter a very disruptive period.

Investment and Economic Implications

Companies that wisely employ these new AI technologies will see a productivity boost unlike anything we’ve ever seen in centuries.

Likewise, individuals who learn to harness these tools will become super-productive.

Companies and people who refuse to adopt this new tech, however, may find themselves falling behind.

Ultimately AI will lead to far higher productivity, and eventually, growth. But the adoption period will be challenging. Some white-collar teams will require less employees, and the disruption this causes will be no small matter.

In many ways, it reminds me of the Industrial Revolution, when new manufacturing tech changed the entire world’s economy. The old way of crafting items by hand with skill and artistry became a rare curiosity.

As investors, we must be keenly aware of what’s coming. Picking stocks which make excellent use of AI will be crucial.

In 2025, we’ll be devoting plenty of attention to AI, so stay tuned. And if you haven’t yet, do check out Jim Rickards’ new book, MoneyGPT: AI and the Threat to the Global Economy.

I hope you all are having an excellent Christmas and Holiday season.

The Daily Reckoning