Jim Keller is one of the most influential chip designers on the planet and he worked for Tesla and led the hardware business between 2016 and 2018. Here is his comment about Elon:
"You think you have an understanding about what the first principles of something are, and then you talk to Elon about it, and you didn't scratch the surface. You know he has a deep belief that no matter what you do is a local maximum."
Andrej Karpathy is shiny star in AI and computer vision, and he led the computer vision team of Autopilot in Tesla from 2017. Here is his comment about Elon:
"I‘m still trying to really map out his superpowers, he has incredibly well-developed intuition. I would say many aspects where he makes the right judgment in what I perceived to be a lack of information because he's not fully in detail of all the things but yet his judgment is extremely good.
I still haven't fully sort of understood how that happens. He has a way of taking a very complex system and simplifying it. It's just like the fundamentals and they're really the first principle components of what really matters about the system and then making statements about those things.
It's a very different way of thinking that I find kind of fascinating by default. For example, sometimes I get sort of overwhelmed by the system. I feel like I need to know the system in its full detail to make the correct decisions. But that's not how he operates he somehow as a way to distill the system into a much simpler system in which he operates and so I think I've learned a lot about just how to approach problems."
Elon is one of the most famous people in the world and he has made inconceivable contributions to humanity, but I don't want to discuss more about his achievements. Instead, it's more inspiring and meaningful to learn the real reasoning process, mental models and logic in his brain.
Someone asked Elon what his real angle was and what he really thinks. His response would always be boring: “I think exactly what I said.” Thus, I think the most effective and efficient way to learn about Elon is directly reading his words and getting some raw data.
Let's walk through some fragments or examples from Elon's public speeches, interviews or other fragments. (it's like "raw data" to learn about Elon) to get some intuitive understanding about first principles or how Elon thinks.
Example 1: Twitter acquisition
Recently, lots of people expressed negative comments or confusion because Elon suspended his plan of acquisition of Twitter. But what matters is understanding the reasons behind that. The following content comes from his interview in All-in podcast.
Elon mocked that Twitter claimed that 95% of the daily active users of platform are real unique people and monetizable users are 217 million. He believed that fake accounts or robots should take at least over 20% of daily active users and it's a lower bound.
The reason is, he has a most liked tweet of any living human and that tweet had roughly 4.9 million likes. He thought it's impossible that the most liked tweet only takes about 2.5% of the entire user base and maybe 10% is more reasonable. The real user number is important since Twitter's business is brand advertising as opposed to purchase advertising.
I think Elon's doubts on Twitter make lots of sense, it's all based on quantitative estimation align with common sense. There is not that much conspiracy but just logical reasoning.
Example 2: Dark
Elon: “When I was a little kid, I was really scared of the dark. But then I came to understand, dark just means the absence of photons in the visible wavelength—400 to 700 nanometers. Then I thought, well it’s really silly to be afraid of a lack of photons. Then I wasn’t afraid of the dark anymore after that.”
I think Elon really, really reached the essence of things. Dark is "absence of photons in the visible wavelength—400 to 700 nanometers", it's really about how universe works and more convincing and informative than almost any other explanation in this case.
Example 3: Electric vehicles
In Elon's initial master plan, he explained details why electric vehicles are far superior than fuel cars. The content is as following:
"The H-System Combined Cycle Generator from General Electric is 60% efficient in turning natural gas into electricity. "Combined Cycle" is where the natural gas is burned to generate electricity and then the waste heat is used to create steam that powers a second generator. Natural gas recovery is 97.5% efficient, processing is also 97.5% efficient and then transmission efficiency over the electric grid is 92% on average. This gives us a well-to-electric-outlet efficiency of 97.5% x 97.5% x 60% x 92% = 52.5%.
Despite a body shape, tires and gearing aimed at high performance rather than peak efficiency, the Tesla Roadster requires 0.4 MJ per kilometer or, stated another way, will travel 2.53 km per mega-joule of electricity. The full cycle charge and discharge efficiency of the Tesla Roadster is 86%, which means that for every 100 MJ of electricity used to charge the battery, about 86 MJ reaches the motor.
Bringing the math together, we get the final figure of merit of 2.53 km/MJ x 86% x 52.5% = 1.14 km/MJ. Let’s compare that to the Prius and a few other options normally considered energy efficient.
The fully considered well-to-wheel efficiency of a gasoline powered car is equal to the energy content of gasoline (34.3 MJ/liter) minus the refinement & transportation losses (18.3%), multiplied by the miles per gallon or km per liter. The Prius at an EPA rated 55 mpg therefore has an energy efficiency of 0.56 km/MJ. This is actually an excellent number compared with a “normal” car like the Toyota Camry at 0.28 km/MJ."
I think it's rare to see such a detailed reasoning process in most cases, even in some professional materials. Reading such analysis could directly increase your comprehension of things. Normal analysis of EVs just generally argues with vaguely defined concepts and rough reasoning process.
Example 4: Engineering methodology
The following content comes from the field interview about starship and starbase. Elon summarized his methodology about engineering.
Elon: What I am trying to implement rigorously is sort of a 5-step process.
“Step one: Make the requirements less dumb. The requirements are definitely dumb; it does not matter who gave them to you. It’s particularly dangerous when they come from an intelligent person, as you may not question them enough. Everyone’s wrong. No matter who you are, everyone is wrong some of the time. All designs are wrong, it’s just a matter of how wrong
Step two: try very hard to delete the part or process. If parts are not being added back into the design at least 10% of the time, [it means that] not enough parts are being deleted. The bias tends to be very strongly toward 'let’s add this part or process step in case we need it'. Additionally, each required part and process must come from a name, not a department, as a department cannot be asked why a requirement exists, but a person can
Step three: simplify and optimize the design. This is the most common error of a smart engineer — to optimize something that should simply not exist, in school people are trained to answer the questions and converge the logic even the questions don't make any sense
Step four: accelerate cycle time. You’re moving too slowly, go faster! But don’t go faster until you’ve worked on the other three things first, if you’re digging your grave, don’t dig it faster.”
The final step is automation. An important part of this is to remove in-process testing after the problems have been diagnosed; if a product is reaching the end of a production line with a high acceptance rate, there is no need for in-process testing. I have personally made the mistake of going backwards on all five steps multiple times. In making Tesla’s Model 3, I literally automated, accelerated, simplified and then deleted"
This process of manufacturing or engineering is actually applicable to lots of fields. It's distilled, summarized and abstracted from large-scale industrial production and contains insights. After learning about his thoughts, it helps people to effectively reflect on their work and life and generate improvements.
Example 5: Boring Company
In a TED talk in 2017, Elon explained some details and vision of Boring Company.
Chris: But people, seen traditionally it’s incredibly expensive to dig, and that would block this idea.
Elon: Yeah. Well, they’re right. To give an example, the LA subway extension, which is, I think it’s a two-and-a-half mile extension that was just completed for two billion dollars. So roughly a billion dollars a mile to do the subway extension in LA and this is not the highest utility subway in the world. So, yeah, it’s quite difficult to dig tunnels normally. I think we need to have at least a tenfold improvement in the cost per mile of tunneling.
Chris: And how could you achieve that?
Elon: I guess actually if you just do two things you can get to approximately an order of magnitude improvement. And I think you can go beyond that. So the first thing to do is to cut the total tunnel diameter by a factor of two or more. So it’s a single road lane tunnel according to regulations has to be 26 feet maybe 28 feet in diameter to allow for crashes and emergency vehicles and sufficient ventilation for a combustion engine cars. But if you if you shrink that diameter to what we were attempting which is 12 feet which is plenty to get an electric skate through, you drop the diameter by a factor of two and the cross-sectional area by a factor of four and the tunneling costs scale with the cross-sectional area. So that’s roughly a half order of magnitude improvement right there. Then tunneling machines currently tunnel for half the time then they stop, and then the rest of the time is putting in reinforcements for the tunnel wall. So if you design the machine instead to do continuous tunneling and reinforcing, that will give you a factor of two improvement. Combine that and it’s a factor of eight. Also, these machines are far from being at their power or thermal limits, so you can jack up the power to the machine substantially. I think you can get at least a factor of two, maybe a factor of four or five improvement on the on top of that. So I think the there’s a fairly straightforward series of steps to get somewhere in excess of an order of magnitude improvement in the cost per mile.
We see the similar reasoning process again. Elon just breakdown the problems into sub-problems and tackle them. He just maps out the cost structure, tries to figure out the solution to reduce cost (time and space). It 's all about simple but reliable math, facts and logic.
Example 6: Neuralink
The following content comes from a podcast hosted by Lex Fridmen with Elon in 2019. I clip some discussion about Neuralink to show what it really is in Elon's mind.
Elon: "What are the problems that we face? Material science, electrical engineering, software, mechanical engineering, microfabrication. It's a bunch of engineering disciplines.
Essentially, that's what it comes down to, is that you have to have a tiny electrode, so small it doesn't hurt neurons, but it's got to last for as long as a person could just last for decades. And then you've got to take that signal. You've got to process that single signal locally at low power. So we need a lot of chip design engineers because we got to signal processing and do so in a very efficient way so that we don't hit your brain up because it's very heat sensitive.
And then and then we've got to take those signals. We've going to do something with them, and then we've got to stimulate and stimulate back to, you know, so you could bidirectional communication. So it's very good at material science, software, mechanical engineering, electrical engineering, chip design, fabrication. That's what those are the things we need to work on. We need to be good at material science so that we can have tiny electrodes that last a long time.
And that's the tough thing with the science behind a tough one, because you're trying to read and stimulate electrically in an electrically active area. Your brain is very electrically active and electric, chemically active. So how do you have a coating on the electrode that doesn't dissolve over time and is safe in the brain?
This is a very hard problem. And then how do you collect those signals? In a way that is the most efficient, because you really just have very tiny amounts of power to process those signals, you know, then we need to automate the whole thing. So it's not if this is done by neurosurgeons, there's no way it can scale to a large number of people. And it needs to scale large numbers of people because I think ultimately we want the future to be determined by a large number of humans."
What did you feel after reading it? Elon is thinking about the problem very seriously and realistically. He elaborated on the existing science and engineering barriers to solve the problem, and there are already some general frameworks in his mind. For instance, he described the " tiny electrode", "bidirectional communication", "dissolve over time" and so on. This technology is definitely challenging, but I think Elon seems to already have some high-level understanding of it and you can feel the vague logic behind it.
Example 7: Science vs. Engineering
Tim Urban is one of my favorite writers and he has an article elaborating how Elon thinks. In the article, he asked whether Elon considered going into scientific discovery instead of engineering. The following is a quote from Tim Urban.
Elon's response:
“I certainly admire the discoveries of the great scientists. They’re discovering what already exists—it’s a deeper understanding of how the universe already works. That’s cool—but the universe already sort of knows that. What matters is knowledge in a human context. What I’m trying to ensure is that knowledge in a human context is still possible in the future. So it’s sort of like—I’m more like the gardener, and then there are the flowers. If there’s no garden, there’s no flowers. I could try to be a flower in the garden, or I could try to make sure there is a garden. So I’m trying to make sure there is a garden, such that in the future, many Feynmans may bloom.”
In other words, both A and B are good, but without A there is no B. So I choose A.
He went on:
“I was at one point thinking about doing physics as a career—I did undergrad in physics—but in order to really advance physics these days, you need the data. Physics is fundamentally governed by the progress of engineering. This debate—“Which is better, engineers or scientists? Aren’t scientists better? Wasn’t Einstein the smartest person?”—personally, I think that engineering is better because in the absence of the engineering, you do not have the data. You just hit a limit. And yeah, you can be real smart within the context of the limit of the data you have, but unless you have a way to get more data, you can’t make progress. Like look at Galileo. He engineered the telescope—that’s what allowed him to see that Jupiter had moons. The limiting factor, if you will, is the engineering. And if you want to advance civilization, you must address the limiting factor. Therefore, you must address the engineering.”
A and B are both good, but B can only advance if A advances. So I choose A.
I think the interesting and awesome thing about Elon is, almost all the arguments he proposed are backed by logic and facts which are so solid and even indisputable. There is whole reasoning and fact architecture or stack in his mind and his decison-making porcess build ontop of it. All the ground-breaking achievements or first principles thinking grow and prosper on this foundation.
Example 8: SpaceX
The following content comes from an interview with Elon in 2019. Elon explained some details about SpaceX and engineering problems.
Elon: "And we've just recently been successful in catching the nose cone of the rocket. We that is a crazy exercise with boat that's basically a giant catcher's mitt. The actual complexity of recovering the fairing [the nose cone apparatus which holds rocket payloads being jettisoned into orbit] is so nuts. Like, I'm not sure we should've done it. We have done it. But each fairing half is like a tiny spacecraft with little thrusters on it. So when it's coming in from space it's in vacuum. And little thrusters controlling the fairing because it's got to come down round side-down because that's where it's got a heat shield. It's got a thermal protection or heat shield on the outer surface but not on the inner surface. So it's gotta come down with the rounded surface coming down. And it's gotta maintain its attitude as it comes in through space. And it comes in hot. If you look at the fairing entry video you see super-heated plasma and sparks and stuff flying off of it. It's coming in at basically five times faster than a bullet from an assault rifle. It's insane."
I think he is describing what really happened in the process. Even as a layman of rocket science, you can feel he takes you straight to the front lines of reality without any nonsense.
Example 9: Self-driving cars
The following content is from a clubhouse interview. I clip some discussion about self-driving cars.
Elon:"It's a very powerful story because once you have autonomy of self-driving cars, you massively increase the utility of any given car. A typical car is driven about 12 hours a week. Depending on the situation where you live, it's like maybe an hour and a half a day or something like that or in LA it might be two hours a day. So roughly 12 hours a week. And there's 168 hours in a week in a seven-day week, so most likely, cars that are autonomous could maybe do a third of the hours in the week or something like that. So, maybe they do, I don't know, 60 hours a week of usage instead of 12.
So you got basically a 5x increase in asset utilization there and far less need for parking lots, parking garages and that kind of thing. This is in itself is good for the environment, because you need fewer cars to get the same thing done. We would need fewer parking garages and places just to keep cars when they're not in use, because they're just being in use a lot more. The net of having a lot of cars times automation or time self-driving, I think is at the heart of why. A lot of investors think Tesla is worth what it is. They're giving us a lot of credit for future execution, but I think the trend is quite positive."
You can get the same "ElonSense" here.
Example 10: LIRAD and autopolit
The following content is from a clubhouse interview. I clip some discussion about technical development path for autopilot such as LIDAR.
Elon:"First of all, I'm not fundamentally against LIDAR on all things because for the Space X Dragon that talks with space station we actually developed and built our own LIDAR for docking with the space station, so obviously if I hated LIDAR, we would not have done that and this is well. Before this was like 10 years ago. We started doing that. I don't have some sort of weird like, antagonism into LIDAR. However for driving on real-world roads you have to solve vision. For basically understanding objects with passive optical, photons then making sense of those objects, so we need like vision perception what this objects mean, what they are going to do, what is the likely path of travel and then control.
The way we're doing it is by running a bunch of neural nets in our head. So we've got to run a bunch of neural nets to do the same thing in the car and at the point at which you've solved passive optical, and this is better passive optical than a human has because you've got eight cameras, three of which point directly forward two are diagonally forward and two that are diagonally rear and one rear.
So the way we do it right now is we have all eight cameras synchronized you've got eight frames collected simultaneously while moving towards. And have in fact mostly moved towards video training. Having eight surround cameras it's kind of a surreal thing to see, because people really people have two eyes, but really more like one eye because two eyes kind of combine. Anyway, the neural net needs to move to full video training inference, surround video training, surround video inference, and then it will be superhuman, no question about it.
Because people don't have eyes on the back of their head, a human for all intents purposes has like one camera on a slow gimbal and that is often distracted and maybe sort of drunk or you know busy changing the radio or they fall asleep or, there's all sorts of things that go wrong there's no question that you can get be super human with just cameras.
I think if one is going to go with sort of active photon generation, I would recommend something in the occlusion penetrating wavelengths like radar, roughly four millimeter radar or something like that would be better if you're going to really delve into the arena of actual photon generation."
You can get the same "ElonSense" here. It's all about facts, logic and framework.
In the end
After learning about some first-hand information and raw data, I just realized maybe Elon is still the most underrated man in the world. All the fragments we know are just tiny parts of all Elon's public information. All the public information available is just tiny parts of all the information he expresses privately and publicly. And all the things he expressed, are just a tiny part of knowledge in his mind.
I always remember Elon said: "I don't want to blow your mind, but I'm always right."
It's like he stands in the future and we stand in today or even yesterday. He thinks on the ground and we think in the air. His world consists of facts, numbers, probability and logic while most poeple’s world consist of other people’s second-hand experience.
It’s fundamentally different.
It's not blind worship of him I didn't see anyone like him that organically combines so much knowledge from so much disciplines together. There is always some enlightenment in his words such as new facts, new logic, new angles which are unconventional but make sense.
I drew a picture to express what the real situation in my mind about Elon and the rest.
Elon is quite unreserved and straightforward about what he is doing and thinking(it’s true confidence). It’s simply because no one can surpass him or his companies by copying of stealing his thoughts or strategies. The competitive edge of Tesla, SpaceX and Elon’s other companies is innovation quality and iteration efficiency, which crush its competitors by order of magnitutude. Linear thinkers never foresee how Elon think and act, always draw wrong conclusions and interpretations about him.
Last, I take this quote for Elon.
Talent hits a target no one else can hit; Genius hits a target no one else can see.
—Arthur Schopenhauer
基于最基本的原理、事实、数据来探究,而非主观观念,观念总是带有立场和私货。
大牛逼