LEAPFROGGING AIs

Musk's Grok3 Clears OpenAI and DeepSeek

Last night, on Monday, February 17th, Musk and his xAi team unveiled Grok3.

The new model has leaped over its competitors in math, science, and coding benchmarks.

In the chart you can see Grok3, in blue, compared to the offerings from OpenAI, DeepSeek, and Google Gemini.

Given that xAi started from scratch a year ago, the achievement is extraordinary and unprecedented.

XAI MOVES FASTER THAN ITS COMPETITION, A KEY COMPETITIVE ADVANTAGE

"All you need to know to understand which company will win a technology competition is look at the first and second derivatives of the rate of innovation." - Elon Musk, February 18, 2025

For those of you whose calculus is rusty, let me break down Musk's comment for you.

He talks about a "rate of innovation". Here, he is thinking of some measurable progress, milestones, per unit of time. I want to make this super simple, so let's talk about something we know. When we drive our car, we travel a distance, in miles, over a measure of time - - think about traveling "35 miles per hour". This is a rate of travel we call "speed": How we change our position "r" over time. In physics, we say, how does our position ("r") differ ("d") as time ("t") differs ("d")? So, speed is "dr/dt", or "difference in position over difference in time".

When Elon mentions the "first derivative" of the rate of innovation, he wants us to look at how that speed changes OVER time. So first we are going to get back in our car and stomp on the gas pedal, hard. We call this "acceleration". That is, "dr/dt", or speed, increases as time clicks by: So, "dr/dt" divided by time is a positive value, greater than zero. If we were to stomp on our brakes, we would "decelerate" and "dr/dt" divided by time goes negative. Got it?

So, first, Elon is saying that the winner in this AI race - - and any tech race - - is the one who NOT ONLY innovates faster, BUT ALSO increases their speed of innovation over time, that is, they ACCELERATE.

ELON MOVES FASTER AND MOVES "BETTER"

But what else is he saying … what is the "second derivative"? [Hint: Remember that Elon owns a rocket company.]

The second derivative of change in position is "jerk".

Back to our car. You actually know what "jerk" is. An inexperienced driver at the wheel of a powerful car might be unable to smooth the transitions, gear-to-gear, giving you a "jerk" as each change is made. So you are accelerating, AND the rate of the acceleration is not smooth, it is "jerky". Got it? If you are in an accident and experience "whiplash", that is the same phenomenon, in a negative way.

My father was an amateur stock car driver, and I inherited his "need for speed" along with, I hope, some of his skills. Have you noticed how good racers move away from a tight turn, toward the outside edge, and then "dive" into the apex, on the inside edge, and then return outside? This "racing line", or optimal path, is using the full width of the lane to avoid "jerk" - - We are driving in a way to make the radius of the turn constant. Applied physics.

Jerk, via whiplash, destroys the human body. Jerk, in racing and in rocketry, destroys pilots and vehicles. Jerk, in innovation, destroys people and organizations! I hear you, Elon.

Elon is saying that the winner in technology not only has acceleration but also manages the process to avoid "jerk".

MOST IMPRESSIVE, AND LEAST NOTICED, IN THE GROK ANNOUNCEMENT IS THE DATA CENTER ACHIEVEMENT

Grok3 was able to leapfrog its competition by training Grok3 in a new, purpose-built data center, nicknamed “Colossus”, the most powerful computer ever built.

In April 2024, Musk and his team decided to build the center. They consulted outside data center providers and were told that it would take 18 to 24 months. For xAi this meant they would lose the Ai race.

They decided to do it themselves. They had no time to build a site, so they looked at good, abandoned factories. An Electrolux plant in Memphis, Tennessee fit the bill. They used used generators while being connected to the electrical grid.

They connected 100,000 NVIDIA H100 GPUs in 122 days, a Herculean feat. They added another 100,000 GPUs in 92 days. The chips are liquid cooled, presenting never-before-encountered plumbing challenges. The mass of GPUs created huge swings in power demand. Musk likened this to an orchestra, the "whole orchestra will go quiet and then loud". Musk and Tesla engineers reconfigured Tesla Mega Pack batteries to buffer the loads. There were enormous networking and debugging challenges.

In the end, xAi created a data center in a fraction of the time anyone else could do it. They plan a new cluster five times the size for their next Ai model.

In my view, this competence, that displayed by the creation of the data center, will be the competence most difficult for competitors to replicate. If this is indeed true, xAi will dominate the Ai space. If there is no “breakaway” with this competence, leapfrogging will be the norm, move-by-move.

EXPECT MORE LEAPFROGGING, BOTH BY TECH FIRMS AND NATION STATES, AS THE AI RACE CONTINUES TO HEAT UP

"The United States of America is the leader in AI, and our administration plans to keep it that way. The US possesses all components across the full AI stack, including advanced semiconductor design, frontier algorithms, and, of course, transformational applications. Now the computing power this stack requires is integral to advancing AI technology. And to safeguard America’s advantage, the Trump administration will ensure that the most powerful AI systems are built in the US with American designed and manufactured chips." - US Vice President JD Vance, keynote speech Paris AI Summit, Feb 11, 2025