Elon Musk’s Latest Interview: 10 Questions Reveal Why First Principles Thinking Is a Superpower
Continue from the previous article
Garry Tan**: You're seen as the archetype of first principles thinking. How do you use first principles to build your view of reality? And how do you teach this mindset to others?
Elon Musk:
First principles thinking is about breaking things down to their most fundamental axioms and reasoning up from there—without relying on analogy or past experience. It’s a mindset that works in any field. It’s like a superpower.
For example, how much does it cost to build a rocket?
Using conventional thinking, you'd look at historical data—so any new cost would naturally resemble the past. But with first principles thinking, you go back to raw materials: aluminum, copper, carbon fiber, steel, etc. How much does a rocket weigh? What’s the weight and cost of each material? That gives you the true minimum cost, as close to raw input cost as possible.
Turns out, without reuse, raw material cost might only be 1–2% of traditional costs. That means the manufacturing process is extremely inefficient.
Here’s an AI example:
Last year, X.AI needed to build a training supercluster with 100,000 H100 chips. Suppliers told us it would take 18 to 24 months, but if we couldn’t do it in six, we’d lose competitiveness.
Everyone said it was impossible. So, we broke the problem down into four parts: space, power, cooling, and stable electricity, and solved each one step by step.
We didn’t have time to build a new site, so we found a decommissioned factory in Memphis. Its existing power supply was 15 megawatts—we needed 150. So we rented generators and parked them beside the building. Then we leased one-fourth of all mobile cooling units in the U.S. and parked them on the other side.
That still wasn’t enough.
During training, power fluctuates wildly—a 50% voltage drop within 100 milliseconds. The generators couldn’t respond fast enough. So we installed Tesla Megapacks to buffer the load and modified the software to smooth the fluctuations.
Our team worked in four shifts, running cable 24/7. I even slept in the data center and helped lay cables myself.
Garry Tan: In the race for foundation models, is pretraining still effective? Does the scaling law still hold?
Musk:
Your team’s talent, hardware scale, and the ability to efficiently utilize hardware are all crucial. It’s not just about buying GPUs—you need to orchestrate and train them stably.
You also need unique data access and a strategy for distribution. How do users actually interact with your AI? These are critical questions.
Many believe—we’ve talked about this, and my friend Ilya (former co-founder and Chief Scientist of OpenAI) has said it too—we're running out of high-quality human-generated pretraining data, especially tokens of high integrity. What comes next is the generation and validation of synthetic data.
Are they truthful? Or are they just hallucinations disconnected from reality?
Rooting AI in truth is extremely hard.
That’s what we’re working on with Grok 3.5—it puts heavy emphasis on reasoning ability.
Humanoid Robots Will Outnumber All Other Robots Combined
Musk:
There will be humanoid robots of every shape and size. Their numbers will outpace all other types of robots combined by an order of magnitude.
Jensen Huang already showed off a dozen humanoid robot prototypes on stage. For years, I was hesitant to jump into this area—we were slow on AI and robotics—because I didn’t want a Terminator scenario to become real. But eventually, I realized:
Whether I do it or not, it will happen.
You only have two choices: watch it unfold, or be part of it. I choose the latter.
Human Consciousness Is Like a Faint Candle in a Vast Darkness
Garry Tan:
In the next 10, 20, or 100 years, how do you view the relationship between AI, robotics, and multi-planetary life? What motivates you to push for interplanetary expansion?
Musk:
I hope that a hundred years from now, human civilization still exists.
By then, the number of humanoid robots will likely be 5 to 10 times the number of humans.
We can look at the Kardashev Scale, which ranks civilizations by energy use:
- Type I harnesses planetary energy,
- Type II uses stellar energy,
- Type III uses galactic energy.
We're only using 1–2% of Earth’s energy potential, far from even Type I.
We are in the very early stages of an "intelligence explosion."
I believe we can deliver enough materials to Mars within 30 years to make it self-sustaining—even if Earth stops sending supplies. That would significantly increase the survival odds of humanity.
The Fermi Paradox still puzzles me—why haven’t we seen aliens? Maybe intelligent life is extremely rare. Perhaps we are the only civilization in the galaxy.
Human consciousness is like a faint candle in a vast darkness.
We must do everything we can to keep it from being snuffed out.
Becoming a multi-planetary species is the first step toward reaching other star systems. Once you can access two planets, it gives the technology enough incentive to evolve. Eventually, we could spread consciousness across the stars.
I Hope AI Can Help Us Understand the Nature of the Universe
Garry Tan:
Where should we go next? These ideas sound like science fiction—but they might be built by people in this very room. What advice would you give to this generation of top tech talent?
Musk:
As I said earlier—if you're doing something useful, that's already incredible.
Creating value for others is doing something good.
For example, focus on building AI that is super grounded in reality—that’s the most important part of AI safety.
When building AI, it’s crucial to tell the truth, even if that truth is politically incorrect.
Where are the aliens?
What is the origin of the universe?
How will it end?
What are the questions we haven’t yet thought to ask?
Are we living in a simulation?
Ultimately, I hope AI can help us understand the fundamental nature of the universe.
Musk’s innovations don’t stem from fantasy—they arise from deep insights into real-world problems.
Whether it’s AI, robotics, or space exploration, the core principle is always the same:
Do something useful—create value, pursue truth, and push the boundaries of knowledge.
His message is clear:
Instead of waiting for the future, build it yourself.
And in doing so, each of us can find our own path forward.