What We’re Reading (Week Ending 03 April 2022) - 03 Apr 2022
Reading helps us learn about the world and it is a really important aspect of investing. The legendary Charlie Munger even goes so far as to say that “I don’t think you can get to be a really good investor over a broad range without doing a massive amount of reading.” We (the co-founders of Compounder Fund) read widely across a range of topics, including investing, business, technology, and the world in general. We want to regularly share the best articles we’ve come across recently. Here they are (for the week ending 03 April 2022):
1. Hundreds of AI tools have been built to catch covid. None of them helped – Will Douglas Heaven
“This pandemic was a big test for AI and medicine,” says Driggs, who is himself working on a machine-learning tool to help doctors during the pandemic. “It would have gone a long way to getting the public on our side,” he says. “But I don’t think we passed that test.”
Both teams found that researchers repeated the same basic errors in the way they trained or tested their tools. Incorrect assumptions about the data often meant that the trained models did not work as claimed.
Wynants and Driggs still believe AI has the potential to help. But they are concerned that it could be harmful if built in the wrong way because they could miss diagnoses or underestimate risk for vulnerable patients. “There is a lot of hype about machine-learning models and what they can do today,” says Driggs.
Unrealistic expectations encourage the use of these tools before they are ready. Wynants and Driggs both say that a few of the algorithms they looked at have already been used in hospitals, and some are being marketed by private developers. “I fear that they may have harmed patients,” says Wynants…
…Many of the problems that were uncovered are linked to the poor quality of the data that researchers used to develop their tools. Information about covid patients, including medical scans, was collected and shared in the middle of a global pandemic, often by the doctors struggling to treat those patients. Researchers wanted to help quickly, and these were the only public data sets available. But this meant that many tools were built using mislabeled data or data from unknown sources.
Driggs highlights the problem of what he calls Frankenstein data sets, which are spliced together from multiple sources and can contain duplicates. This means that some tools end up being tested on the same data they were trained on, making them appear more accurate than they are.
It also muddies the origin of certain data sets. This can mean that researchers miss important features that skew the training of their models. Many unwittingly used a data set that contained chest scans of children who did not have covid as their examples of what non-covid cases looked like. But as a result, the AIs learned to identify kids, not covid.
Driggs’s group trained its own model using a data set that contained a mix of scans taken when patients were lying down and standing up. Because patients scanned while lying down were more likely to be seriously ill, the AI learned wrongly to predict serious covid risk from a person’s position.
In yet other cases, some AIs were found to be picking up on the text font that certain hospitals used to label the scans. As a result, fonts from hospitals with more serious caseloads became predictors of covid risk.
Errors like these seem obvious in hindsight. They can also be fixed by adjusting the models, if researchers are aware of them. It is possible to acknowledge the shortcomings and release a less accurate, but less misleading model. But many tools were developed either by AI researchers who lacked the medical expertise to spot flaws in the data or by medical researchers who lacked the mathematical skills to compensate for those flaws.
A more subtle problem Driggs highlights is incorporation bias, or bias introduced at the point a data set is labeled. For example, many medical scans were labeled according to whether the radiologists who created them said they showed covid. But that embeds, or incorporates, any biases of that particular doctor into the ground truth of a data set. It would be much better to label a medical scan with the result of a PCR test rather than one doctor’s opinion, says Driggs. But there isn’t always time for statistical niceties in busy hospitals…
…What’s the fix? Better data would help, but in times of crisis that’s a big ask. It’s more important to make the most of the data sets we have. The simplest move would be for AI teams to collaborate more with clinicians, says Driggs. Researchers also need to share their models and disclose how they were trained so that others can test them and build on them. “Those are two things we could do today,” he says. “And they would solve maybe 50% of the issues that we identified.”
Getting hold of data would also be easier if formats were standardized, says Bilal Mateen, a doctor who leads the clinical technology team at the Wellcome Trust, a global health research charity based in London.
Another problem Wynants, Driggs, and Mateen all identify is that most researchers rushed to develop their own models, rather than working together or improving existing ones. The result was that the collective effort of researchers around the world produced hundreds of mediocre tools, rather than a handful of properly trained and tested ones.
“The models are so similar—they almost all use the same techniques with minor tweaks, the same inputs—and they all make the same mistakes,” says Wynants. “If all these people making new models instead tested models that were already available, maybe we’d have something that could really help in the clinic by now.”
2. The Pendulum in International Affairs – Howard Marks
Because psychology swings so often toward one extreme or the other – and spends relatively little time at the “happy medium” – I believe the pendulum is the best metaphor for understanding trends in anything affected by psychology. . . not just investing…
…The first item on the agenda for Brookfield’s board meeting was, naturally, the tragic situation in Ukraine. We talked about the many facets of the problem, ranging from human to economic to military to geopolitical. In my view, energy is one of the aspects worth pondering. The desire to punish Russia for its unconscionable behavior is complicated enormously by Europe’s heavy dependence on Russia to meet its energy needs; Russia supplies roughly one-third of Europe’s oil, 45% of its imported gas, and nearly half its coal.
Since it can be hard to arrange for alternative sources of energy on short notice, sanctioning Russia by prohibiting energy exports would cause a significant dislocation in Europe’s energy supply. Curtailing this supply would be difficult at any time, but particularly so at this time of year, when people need to heat their homes. That means Russia’s biggest export – and largest source of hard currency ($20 billion a month is the figure I see) – is the hardest one to sanction, as doing so would cause serious hardship for our allies. Thus, the sanctions on Russia include an exception for sales of energy commodities. This greatly complicates the process of bringing economic and social pressure to bear on Vladimir Putin. In effect, we’re determined to influence Russia through sanctions . . . just not the potentially most effective one, because it would require substantial sacrifice in Europe. More on this later.
The other subject I focused on, offshoring, is quite different from Europe’s energy dependence. One of the major trends impacting the U.S. economy over the last year or so – and a factor receiving much of the blame for today’s inflation – relates to our global supply chains, the weaknesses of which have recently been on display. Thus, many companies are seeking to shorten their supply lines and make them more dependable, primarily by bringing production back on shore.
Over recent decades, as we all know, many industries moved a significant percentage of their production offshore – primarily to Asia – bringing down costs by utilizing cheaper labor. This process boosted economic growth in the emerging nations where the work was done, increased savings and competitiveness for manufacturers and importers, and provided low-priced goods to consumers. But the supply-chain disruption that resulted from the Covid-19 pandemic, combined with the shutdown of much of the world’s productive capacity, has shown the downside of that trend, as supply has been unable to keep pace with elevated demand in our highly stimulated economy.
At first glance, these two items – Europe’s energy dependence and supply-chain disruption – may seem to have little in common other than the fact that they both involve international considerations. But I think juxtaposing them is informative . . . and worthy of a memo…
…U.S. companies’ foreign sourcing, in particular with regard to semiconductors, differs from Europe’s energy emergency in many ways. But both are marked by inadequate supply of an essential good demanded by countries or companies that permitted themselves to become reliant on others. And considering how critical electronics are to U.S. national security – what today in terms of surveillance, communications, analysis and transportation isn’t reliant on electronics? – this vulnerability could, at some point, come back to bite the U.S. in the same way that dependence on Russian energy resources has the European Union.
How did the world get into this position? How did Europe become so dependent on Russian exports of energy commodities, and how did such a high percentage of semiconductors and other goods destined for the U.S. come to be manufactured abroad? Just as Europe allowed its energy dependence to increase due to its desire to be more green, U.S. businesses came to rely increasingly on materials, components, and finished goods from abroad to remain price-competitive and deliver greater profits.
Key geopolitical developments in recent decades included (a) the perception that the world was shrinking, due to improvements in transportation and communications, and (b) the relative peace of the world, stemming from:
- the dismantling of the Berlin Wall; the fall of the USSR;
- the low perceived threat from nuclear arms (thanks to the realization that their use would assure mutual destruction);
- the absence of conflicts that could escalate into a multi-national war;
- and the shortness of memory, which permits people to believe benign conditions will remain so.
Together, these developments gave rise to a huge swing of the pendulum toward globalization and thus countries’ interdependence. Companies and countries found that massive benefits could be tapped by looking abroad for solutions, and it was easy to overlook or minimize potential pitfalls.
As a result, in recent decades, countries and companies have been able to opt for what seemed to be the cheapest and easiest solutions, and perhaps the greenest. Thus, the choices made included reliance on distant sources of supply and just-in-time ordering.
3. How a Founder’s Childhood in India Inspired His Fight Against Climate Change – Annie Goldsmith and Shashank Samala
Samala grew up outside Hyderabad in southeast India, where he saw firsthand the effects of wealth inequality and climate change.
The house that I grew up in was around 200 square feet for five people. And I was exposed to droughts and cyclones and so forth. People don’t call it climate change there, which shows the [lack of] education. But the worst impacts of climate are faced by the world’s most vulnerable people, and these are them.
When Samala was nine, his father moved to the U.S. with hopes of bringing his family along when he was financially stable.
My dad was away for many years, working any minimum wage job he could get, trying to figure out how to bring us here. He actually went to pharmacy school—a pharmacist is just not a valuable job in India. In his early 40s, he moved to Reston, Va., where there’s this massive mall, Tysons Corner Center. He got a job serving Dippin’ Dots ice cream and he would send money back for [me, my mom and my two siblings] to live. He eventually realized that if you pass a bunch of exams, you can become a pharmacist here. It took him a few years, and the first job he got was as an intern at a Rite Aid in Old Town, Maine. He didn’t know where Maine was, but at that point he would take anything. So he just went to the bus station. I don’t know how he managed to do it.
He and his mother and siblings moved to Bangor, Maine, when Shashank was 12, relocating near the pharmacy where his father worked.
The high school was 1,300 kids and I was one of few nonwhite folks. It was overwhelming to me—there was a big culture shock. I really wanted to go back to my friends and everyone in India, but I somehow stuck through it. I remember flying into Boston Logan [International] Airport and you couldn’t see outside—it was just white. I asked my dad, “What is that?” It was four feet of snow…
…After nearly seven years at the company, Samala left in 2020 to found his current venture, Heirloom Carbon.
I just realized, at the end of the day, [Tempo was creating] more gadgets in the world. I mean, important gadgets in many cases—like medical devices and rockets and so forth. But I think the inequity issue just kept coming back to me. I thought about where I wanted to spend time. That’s when I started thinking about other things.
Increasing climate crises intensified Samala’s desire to work in environmental technology.
In 2020, in Hyderabad, there was this massive flood, like a once-in-a-century type of flood, and those are just happening even more and more. That’s my hometown, completely flooded. The airwaves [in America] don’t catch that stuff, but that’s very prominent in a lot of these people’s lives. Now, basically Heirloom has a bunch of these carbon removal contactors and these kilns, and our first site is going to be deployed this time next year. It’s going to be in the [San Francisco] Bay Area. Call it a demonstration, I guess, but it’s the first deployment of its kind in North America, second one in the world. It’s going to be capturing a meaningful amount of CO2.
4. How People Think – Morgan Housel
2. What people present to the world is a tiny fraction of what’s going on inside their head.
The Library of Congress holds three million books, or something like a quarter of a trillion words.
All of the information accessible on the internet is estimated at 40 trillion gigabytes, which is roughly enough to hold a high-def video lasting the entire 14 billion years since the big bang.
So much of history has been recorded.
But then you remember, that’s just what’s been publicly shared, recorded, and published. It’s a trivial amount of what’s actually happened, and an infinitesimal amount of what’s gone through people’s heads.
As much as we know about how crazy, weird, talented, and insightful people can be, we are blind to perhaps 99.99999999% of it. The most prolific over-sharers disclose maybe a thousandth of one percent of what they’ve been through and what they’re thinking.
One thing this does is gives a false view of success. Most of what people share is what they want you to see. Skills are advertised, flaws are hidden. Wins are exaggerated, losses are downplayed. Doubt and anxiety are rarely shared on social media. Defeated soldiers and failed CEOs rarely sit for interviews.
Most things are harder than they look and not as fun as they seem because the information we’re exposed to tends to be a highlight reel of what people want you to know about themselves to increase their own chances of success. It’s easiest to convince people that you’re special if they don’t know you well enough to see all the ways you’re not.
When you are keenly aware of your own struggles but blind to others’, it’s easy to assume you’re missing some skill or secret that others have. Sometimes that’s true. More often you’re just blind to how much everyone else is making it up as they go, one challenge at a time…
…4. We are extrapolating machines in a world where nothing too good or too bad lasts indefinitely.
When you’re in the middle of a powerful trend it’s difficult to imagine a force strong enough to turn things the other way.
What we tend to miss is that what turns trends around usually isn’t an outside force. It’s when a subtle side effect of that trend erodes what made it powerful to begin with.
When there are no recessions, people get confident. When they get confident they take risks. When they take risks, you get recessions.
When markets never crash, valuations go up. When valuations go up, markets are prone to crash.
When there’s a crisis, people get motivated. When they get motivated they frantically solve problems. When they solve problems crises tend to end.
Good times plant the seeds of their destruction through complacency and leverage, and bad times plant the seeds of their turnaround through opportunity and panic-driven problem-solving.
We know that in hindsight. It’s almost always true, almost everywhere.
But we tend to only know it in hindsight because we are extrapolating machines, and drawing straight lines when forecasting is easier than imagining how people might adapt and change their behavior.
When alcohol from fermentation reaches a certain point it kills the yeast that made it in the first place. Most powerful trends end the same way. And that kind of force isn’t intuitive, requiring you to consider not just how a trend impacts people, but how that impact will change people’s behavior in a way that could end the trend…
…7. We are pushed toward maximizing efficiency in a way that leaves no room for error, despite room for error being the most important factor of long-term success.
The world is competitive. If you don’t exploit an opportunity your competition will. So opportunity is usually exploited to its fullest extent as soon as possible.
That’s great – it pushes the world forward. But it has a nasty side effect: When all opportunity is exploited there is no room for error, and when there’s no room for error any system exposed to volatility and accident will eventually break.
Describing the supply chain fiasco of the last year, Flexport CEO Ryan Petersen explained:
What caused all the supply chain bottlenecks? Modern finance with its obsession with “Return on Equity.”
To show great ROE almost every CEO stripped their company of all but the bare minimum of assets. Just in time everything. No excess capacity. No strategic reserves. No cash on the balance sheet. Minimal R&D.
We stripped the shock absorbers out of the economy in pursuit of better short term metrics. Now as we’re facing a hundred year storm of demand, our infrastructure simply can’t keep up.
The global logistics companies have no excess capacity, there are no reserves of chassis (trailers for hauling containers), no extra shipping containers, no extra yard space, no extra warehouse capacity. The brands have no extra inventory. Manufacturers have no extra components or raw materials on hand.
He’s right, but part of me can also empathize with the CEOs who maximized efficiency because if they didn’t their competition would have and put them out of business. There’s a weird quirk of human behavior that incentivizes people to maximize potential all the way up to destroying themselves.
So many people strive for efficient lives, where no hour is wasted. But when no hour is wasted you have no time to wander, explore something new, or let your thoughts run free – which can be some of the most productive forms of thought. Psychologist Amos Tversky once said “the secret to doing good research is always to be a little underemployed. You waste years by not being able to waste hours.” A successful person purposely leaving gaps of free time on their schedule can feel inefficient. And it is, so not many people do it.
The paradox that room for error is essential to survival in the long run, but maximizing efficiency in a way that eliminates room for error can be essential to surviving the short run, is a strange one.
Those who fight it – the rare company or employee or economy willing to sacrifice short-term gain for long-term survival – are the oddballs, rarely understood, easily belittled, who underperform most of the time but survive long enough to get the last laugh, and the highest returns…
…16. We are blind to how fragile the world is due to a poor understanding of rare events.
John Littlewood was a mathematician who sought to debunk the idea of miracles being anything more than simple statistics.
Physicist Freeman Dyson explains:
Littlewood’s law of miracles states that in the course of any normal person’s life, miracles happen at the rate of roughly one per month.
The proof of the law is simple. During the time that we are awake and actively engaged in living our lives, roughly for eight hours each day, we see and hear things happening at a rate of one per second. So the total number of events that happen to us is about 30,000 per day, or about a million per month.
With few exceptions, these events are not miracles because they are insignificant. The chance of a miracle is about one per million events. Therefore we should expect about one miracle to happen, on average, every month.
The idea that incredible things happen because of boring statistics is important, because it’s true for terrible things too.
Think about 100-year events. One-hundred-year floods, hurricanes, earthquakes, financial crises, frauds, pandemics, political meltdowns, economic recessions, and so on endlessly. Lots of terrible things can be called “100-year events”.
A 100-year event doesn’t mean it happens every 100 years. It means there’s about a 1% chance of it occurring in any given year. That seems low. But when there are hundreds of different independent 100-year events, what are the odds that any one of them will occur in a given year?
Pretty good, in fact.
If next year there’s a 1% chance of a new disastrous pandemic, a 1% chance of a crippling depression, a 1% chance of a catastrophic flood, a 1% chance of political collapse, and on and on, then the odds that something bad will happen next year – or any year – are … uncomfortably high.
Littlewood’s Law tells us to expect a miracle every month. The flip side is to expect a disaster roughly as often.
Which is what history tells us, isn’t it?
History is “just one damn thing after another,” said Arnold Toynbee. Dan Carlin’s book The End is Always Near highlights periods – from pandemics to nuclear war – where it felt like the world was coming to an end. They exist in every era, every continent, every culture. Bad news is the norm.
Even during what we remember as prosperous periods, like the 1950s and 1990s, there was a continuous chain of grief. Adjusted for population growth, more Americans lost their jobs during the 1958 recession than did in any single month during the Great Recession of 2008. The global financial system nearly fell apart in 1998, during the greatest prosperity boom we’ve ever seen.
The world breaks about once every ten years, on average. For your country, state, town, or business, once every one to three years is probably more common.
Sometimes it feels like terrible luck, or that bad news has new momentum. More often it’s just Littlewood’s Law at work. A zillion different things can go wrong, so at least one of them is likely to be causing havoc in any given moment.
5. An Interview with Nvidia CEO Jensen Huang about Manufacturing Intelligence – Ben Thompson and Jensen Huang
Well, it’s interesting because I mean, not to hop ahead, but I was going to ask you about the shift to memory bandwidth and super wide just being more and more important. One of the things that was really striking in your keynote this time was every time whether you talked about chips, or you talked about your new CPU, or you talked about your systems, you basically just spent the whole time talking about memory, and how much stuff can be moved around. It’s interesting to hear you say that that was actually a key consideration really from the beginning. Everyone thinks about the graphics part of it, but you have to keep those things fed, and that’s actually been important all along as well.
JH: Yeah, that’s exactly right. It turns out that in computer graphics, we chew through more memory bandwidth than just about anything because we have to render to pixel, and because it’s a painter’s algorithm, you paint over the pixels over and over and over again, and each time, you have to figure out which one’s in front of which, and so there’s a read-modify-write, and the read-modify-write chews up more memory bandwidth, and if it’s a blend, that chews up more memory bandwidth. So, all of those layers and layers and layers of composition just chews up a ton of bandwidth, and as we moved into the world of machine learning and this new era of computing where the software is not written just by a human, the architecture’s created by the human, but the architecture’s tuned by the machine studying the data, and so we pump in tons and tons of data so that the machine learning algorithm could figure out what the patterns are, what the predictive features are, and what the relationships are. All of that is just memory bandwidth and so we’re really comfortable with this area of computation, so it goes all the way back to the very beginning as served as well.
I have always said the most misnamed product in tech is the personal computer, because obviously the personal computer is your phone and not the PC you leave on your desk, but we’ve wasted such a great name. I feel like GPUs is like the opposite direction. We call this a graphics processing unit, but to your point, the idea of keeping it fully fed, doing relatively simple operations and massively parallel all at the same time, that’s a specific style of computing that happened to start with graphics, but we’re stuck calling it GPU forever instead of, I don’t know, advanced processing unit or whatever it should be. I mean, what should the name be?
JH: Once GPU took off and we started adding more and more capabilities to it, it was just senseless to rename it. There were lots of ideas. Do we call it GPGPU? Do we call it an XPU? Do we call it a VPU? I just decided that it wasn’t worth playing that game, and what we ought to do is assume that people who buy these things are smart enough to figure out what they do, and we’ll be clever enough to help people understand what the benefits are, and we’ll get through all the naming part.
The thing that is really remarkable, if you look at TNT, it was a fixed function pipeline, meaning every single stage of the pipeline, it did what it did, and it moved the data forward, and if it ever needed to read the data from the frame buffer, the memory, if it ever needed to read the memory data back to do processing, it would read the data, pull it back into the chip, and do the processing on it, and then render it back into the frame buffer, doing what is called multipass. Well, that multipass, a simple fixed function pipeline approach, was really limiting, which led to the invention several years later of the programmable shader which —
This is great. You are literally walking down my question tree on your own, so this is perfect. Please continue.
JH: (laughing) So, we invented a programmable shader, which put a program onto the GPU, and so now there’s a processor. The challenge of the GPU, which is an incredible breakthrough, during that point when we forked off into a programmable processor, to recognize that the pipeline stages of a CPU was, call it umpteen stages, but the number of pipeline stages in a GPU could be several hundred, and yet, how do you keep all of those pipe stages and all of those processors fed? You have to create what is called a latency tolerant processor, which led to heavily threaded processors. Whereas you could have two threads in a microprocessor going in any CPU core, hyper-threading, in the case of our GPU, at any given point in time, we could have 10,000 threads in flight. So it’s 10,000 programs, umpteen thousand programs, that are flying through this processor at any given point in time, which really reinvented the type of this new style of programming, and our architecture called CUDA made it accessible, and because we dedicated ourselves to keeping every generation of processors CUDA-compatible, we invented a new programming model. That was all started right around that time.
I’m actually curious about this, because what is fascinating about NVIDIA is if you look backwards, it seems like the most amazing, brilliant path that makes total sense, right? You start by tackling the most advanced accelerated computing use case, which is graphics, but they’re finally tuned to OpenGL and DirectX and just doing these specific functions. You’re like, “Well, no, we should make it programmable.” You invent the shader, the GeForce, and then it opens its door to be programmed for applications other than graphics. NVIDIA makes it easier and more approachable with CUDA, you put SDKs on top of CUDA, and now twenty-five years on NVIDIA isn’t just the best in the world at accelerated computing, you have this massive software moat and this amazing business model where you give CUDA away for free and sell the chips that make it work. Was it really that much on purpose? Because it looks like a perfectly straight line. I mean, when you go back to the 90s, how far down this path could you see?
JH: Everything you described was done on purpose. It’s actually blowing my mind that you lived through that, and I can’t tell you how much I appreciate you knowing that. Just knowing that is quite remarkable. Every part of that you described was done on purpose. The parts that you left out, of course, are all the mistakes that we made. Before there was CUDA, there was actually another version called C for Graphics, Cg. So, we did Cg and made all the mistakes associated with it and realized that there needed to be this thing called shared memory, a whole bunch of processors being able to access onboard shared memory. Otherwise, the amount of multipassing-
Yeah, the coherence would fall apart, yeah.
JH: Yeah, just the whole performance gets lost. So, there were all kinds of things that we had to invent along the way. GeForce FX had a fantastic differentiator with 32 bit floating point that was IEEE compatible. We made a great decision to make it IEEE FP32 compatible. However, we made a whole bunch of mistakes with GeForce FX —
Sorry, what does that mean? The IEEE FP32 compatible?
JH: Oh, the IEEE specified a floating point format that if you were to divide by zero, how do you treat it? If it’s not a number, how do you treat it?
Got it. So this made it accessible to scientists and things along those lines?
JH: So that whatever math that you do with that floating point format, the answer is expected.
Right.
JH: So, that made it consistent with the way that microprocessors treated floating point processes. So we could run a floating point program and our answer would be the same as if you ran it on a CPU. That was a brilliant move. At the time, DirectX’s specification of programmable shaders was 24 bit floating point, not 32 bit floating point. So, we made the choice to go to all 32 bits so that whatever numerical computation is done is compatible with processors. That was a genius move, and because we saw the opportunity to use our GPUs for general purpose computing. So that was a good move.
There were a whole bunch of other mistakes that we made along the way that tripped us up along the way as we discovered these good ideas. But each one of these good ideas, when they were finally decided on, were good. For example, recognizing that CUDA was going to be our architecture and that we would, if CUDA is a programmable architecture, we have to stay faithful to it and made sure that every generation was backwards compatible to the previous generation, so that whatever install base of software was developed would benefit from the new processor by running faster. If you want developers, they’re going to want install base, and if you want install base, they have to be compatible with each other. So, that decision forced us to put CUDA into GeForce, forced us put CUDA into Quadro, forced us to put CUDA into data center GPUs, into everything, basically, and today, in every single chip that we make, it’s all CUDA compatible…
…You talked about there being four layers of the stack in your keynote this week. You had hardware, system software, platform, and then application frameworks, and you also have said at other times that you believe these machine learning opportunities require sort of a fully integrated approach. Let’s start with that latter one. Why is that? Why do these opportunities need full integration? Just to step back, the PC era was marked by modularity, you had sort of the chip versus the operating system versus the application, and to the extent there were integrations or money to be made, it was by being that connective tissue, being a platform in the middle and the smartphone era on the other hand was more about integration and doing the different pieces together. It sounds like your argument is that this new era, this machine learning-driven era, this AI era is even more on the integrated side than sort of the way we think about PCs. Why is that? Walk me through that justification.
JH: Simple example. Imagine we created a new application domain, like computer graphics. Let’s pretend for a second it doesn’t run well on a graphics chip and it doesn’t run well on a CPU. Well, if you had to recreate it all again, and it’s a new form of computer science in the sense that this is the way software is developed, and you can develop all kinds of software, it’s not just one type of software, you can develop all kinds of software. So if that’s the case, then you would build a new GPU and a new OpenGL. You would build a new processor called New GPU, you would build a new compiler, you would build a new API, a new version of OpenGL called cuDNN. You would create a new Unreal Engine, in this case, Nvidia AI. You would create a new editor, new application frameworks and so you could imagine that you would build a whole thing all over again.
Just to jump in though, because there was another part in the keynote where I think you were talking about Nvidia DRIVE and then you jumped to Clara, something along those lines, but what struck me as I was watching it was you were like, “Actually all the pieces we need here, we also need there”, and it felt like a real manifestation of this. Nvidia has now built up this entire stack, they almost have all these Lego bricks that they can reconfigured for all these different use cases. And if I’m Nvidia, I’m like, “Of course these must be fully integrated because we already have all the integrated pieces so we’re going to put it all together with you”. But is that a function of, “That’s because Nvidia is well placed to be integrated” or is that “No, this is actually really the only way to do it” and if other folks try to have a more modular approach, they’re just not going to get this stuff working together in a sufficient way?
JH: Well, deep learning, first of all, needed a brand new stack.
Just like graphics once did. Yeah.
JH: Yeah, just like graphics did. So deep learning needed a brand new stack, it just so happened that the best processor for deep learning at the time, ten years ago, was one of Nvidia’s GPUs. Well, over the years, in the last ten years, we have reinvented the GPU with this thing called Tensor Core, where the Tensor Core GPU is a thousand times better at doing deep learning than our original GPU, and so it grew out of that. But in the process, we essentially built up the entire stack of computer science, the computing, again, new processor, new compiler, new engine and new framework — and the framework for AI, of course, PyTorch and TensorFlow.
Now, during that time, we realized that while we’re working on AI — this is about seven years ago — the next stage of AI is going to be robotics. You’re going to sense, you’re going to perceive, but you’re also going to reason and you’re going to plan. That classical robotics problem could be applied to, number one, autonomous driving, and then many other applications after that. If you think through autonomous driving, you need real-time sensors of multiple modalities, the sensors coming in in real-time. You have to process all of the sensors in real-time and it has to be isochronous, you have to do it consistently in real-time and you’re processing radar information, camera information, Lidar information, ultrasonics information, it’s all happening in real-time and you have to do so using all kinds of different algorithms for diversity and redundancy reasons. And then what comes out of it is perception, localization, a world map and then from that world map, you reason about what is your drive plan. And so that application space was a derivative, if you will, of our deep learning work, and it takes us into the robotic space. Once we’re in the robotic space and we created a brand new stack, we realized that the application of this stack, the robotic stack, could be used for this and it could be used for medical imaging systems, which is kind of multi-sensor, real-time sensor processing, used to be traditional numerics.
Right. Well, it’s like you started out with like your GPU like, “Oh, it could be used for this and this and this”. And now you built a stack on top of the GPU and it’s like, it just expands. “It could be used for this and this and this.”
JH: That’s exactly right, Ben! That’s exactly right. You build one thing and you generalize it and you realize it could be used for other things, and then you build that thing derived from the first thing and then you generalize it and when you generalize it, you realize, “Hold on a second, I can use it for this and as well”. That’s how we built the company.
6. Rule #1: Do No Harm – Permanent Equity
At Permanent Equity, “Do No Harm” is a primary cornerstone of our approach, defined within the context of our values and priorities. “Do No Harm” sounds simple. But what counts as harm?
The ethics of defining harm have always been challenging. Even Hippocrates, commonly credited with penning primum, non nocere (first, do no harm) didn’t actually include it in the Hippocratic Oath. If a doctor were actually going to do no harm, active surgical intervention would never happen, and the ongoing harm of a tumor would be allowed to fester and spread.
In any type of decision-making, from healthcare to investing, the possibility for harm is everywhere, and Do No Harm is not a neutral statement.
By approaching our partnerships with humility and curiosity, by prioritizing outcomes in which everyone wins, by building trust in our relationships with operations teams, and by understanding the value we can provide, we’ve figured out what harm means to us. And we work like hell not to do it…
…In G.K. Chesterton’s 1929 book The Thing (bear with us), two reformers come across a fence in the middle of a field. The first leaps into action, proclaiming “I don’t see the use of this; let us clear it away.” The second quickly replies “If you don’t see the use of it, I certainly won’t let you clear it away. Go away and think. Then, when you can come back and tell me that you do see the use of it, I may allow you to destroy it.”
We like to think we’re the second kind of reformer, even if we tend to think in terms of strategic changes to small business operations instead of fences and fields (although if you’ve got a fencing company you’re interested in selling, by all means get in touch).
Every process, procedure, and position a small business has was initially created (and probably with some pain and suffering); it had a purpose, even if that purpose isn’t clear to someone new coming in the door. Just as Chesterson’s fence wasn’t beamed down by aliens, a small business’s invoicing system wasn’t cooked up in a fever dream.
One of the basic tenets of Do No Harm is that everything in a small business was put in place because somebody thought it would be good for something. If we can’t figure out what that something is, we’ve likely missed some other thing and misplaced our humility. And if we start making changes from a place where we’ve decided things are meaningless and mysterious, chances are that harm will follow–whether that be alienating existing leadership or employees, gumming up processes we thought could run smoother, shifting customer profiles to a market that doesn’t exist, deleting some operationally critical Excel sheet that looked like it was outdated lunch orders, or some other harm we can’t even think of because potential harms are innumerable–and specific to company and context.
Whatever the “fence,” we prioritize studying it and talking to the people who built it. Only once we know fully why it’s there, what it’s supposed to do, and whether it is or isn’t doing that, then we collectively can decide whether to keep it, replace it, repair it, double down on it, or practice our hurdle sprints over it…
…Let’s also be clear that harm to a company can start well before investment boots hit the ground and reverberate after exit. In many cases, harm–or at least the potential for harm–is baked into the system from the get-go.
Do No Harm, scaffolded by our commitment to humility, extends to the price we pay, our aversion to transactional debt, our default to asking questions, and our unwillingness to let non-business-related events like a change in ownership drive changes in the business.
A business operates with certain economic pressures (e.g., payroll) consistently over time. When an equity transaction happens, depending on the way that transaction is structured, the magnitude and source of those pressures can shift enormously, very quickly. For example, traditional private equity’s model frequently hinges on debt and complicated fee structures. In this case, it’s possible for a business to go from having total autonomy over its cash flows to having 60% or more of cash flows earmarked for debt payments.
Whether such a structure inflicts direct financial harm is specific to that business, but it changes the very rules of the game. That business must now shift their decision-making to prioritizing repaying debts first. Debt is not a productive operating priority. This change to the operating core, prompted by an investment system based in debt, has the potential to do significant harm.
In practicing the principle of let great be great, we go back to Rule #1 and typically employ no debt. We have no transaction fees or other “gotchas.” In principle, we aim to design deals that do not put financial and operating priorities at odds–and therefore enable operators to focus on operating to build durable value.
The leveraged buyout incentive system also ramps up the potential for harm post-close. When the primary objective is to provide investors a return in three to five years, decision-making is made in that context. Dramatic increases in EBITDA are inevitably the aim because that’s the most straightforward path to a higher valuation exit. But if you’re moving fast enough, you can keep adding to a leaky bucket and have it feel full. But someone at some point will be left with a mess.
It’s hard to square these foundational choices with a principle of Do No Harm; when we say “Do No Harm,” we’re not just talking about not making major changes before we’ve forged relationships and gotten a genuine read on the landscape–it means, from the very beginning, we aim to make rational investments that acknowledge embedded risks in reality–without introducing new forms of harm.
7. Elon Musk discusses the war in Ukraine and the importance of nuclear power — and why Benjamin Franklin would be ‘the most fun at dinner’ – Mathias Döpfner and Elon Musk
Mathias Döpfner: Before we talk about the future, let’s look at the present. There is war in Europe. If you see the horrible images of Putin’s troops invading Ukraine, killing people. What are your thoughts?
Elon Musk: It is surprising to see that in this day and age. I thought we had sort of moved beyond such things for the most part. It is concerning. If you can get away with it, then this will be a message to other countries that perhaps they could get away with it too…
…Döpfner: You did something very concrete, 48 hours, upon the request of the digital minister of Ukraine. And that was delivering Starlink material in order to grant internet access. What was the motivation, and how is it developing?
Musk: We did think that Starlink might be needed, and we took some preemptive actions to ensure that it could be provided quickly. When the request came, we acted very rapidly. It is worth noting that the satellite internet connectivity of Ukraine was taken offline by a cyberattack on the day of the invasion permanently. The cell towers are either being blown up or they are being jammed. There is a major fiber backbone which the Russians are aware of. It was quite likely that they will sever that fiber link. This would leave Ukraine with very few connections open. So Starlink might be, certainly in some parts of Ukraine, the only connection.
Döpfner: What happens if the Russians and Chinese are targeting satellites? Is that also a threat for Starlink?
Musk: It was interesting to view the Russian anti-satellite demonstration a few months ago in the context of this conflict. Because that caused a lot of strife for satellite operators. It even had some danger for the space station, where there are Russian cosmonauts. So why did they do that? It was a message in advance of the Ukraine invasion. If you attempt to take out Starlink, this is not easy because there are 2000 satellites. That means a lot of anti-satellite missiles. I hope we do not have to put this to a test, but I think we can launch satellites faster than they can launch anti-satellites missiles.
Döpfner: Russia said that they are going to stop the delivery of rocket engines. Is that a threat or an opportunity for SpaceX?
Musk: At SpaceX, we design and manufacture our own rocket engines. So we did not really own any Russian components at all…
…Döpfner: With knowledge, products and services, Elon Musk is almost a strategic weapon in modern warfare. How do you see your role in that context?
Musk: I think I can be helpful in conflicts. I try to take a set of actions that are most likely to improve the probability that the future will be good. And obviously sometimes I make mistakes in this regard. I do whatever I think is most likely to ensure that the future is good for humanity. Those are the actions that I will take…
…Döpfner: History doesn’t repeat itself, but it rhymes. And we see a rhyme these days. Back to the big strategic picture. The terrible actions of Putin are, to a certain degree, also a result of strategic mistakes that Europe, particularly Germany, has made, the dropout of nuclear energy in 2011.
Musk: It is very important that Germany will not shut down its nuclear power stations. I think this is extremely crazy.
Döpfner: If we really want to reduce Putin’s power as well as Europe’s and Germany’s dependence on Russian energy, we have to decarbonize. It’s the only way. Is more nuclear energy the key to free ourselves from dictators and autocrats like Putin.
Musk: I want to be super clear. You should not only not shut down the nuclear power plants, but you should also reopen the ones that have already shut down. Those are the fastest to produce energy. It is crazy to shut down nuclear power plants now, especially if you are in a place where there are no natural disasters. If you are somewhere where severe earthquakes or tsunamis occur, it is more of a question mark. If there is no massive natural disaster risk-which Germany does not have-then there is really no danger with the nuclear power plants.
Döpfner: Aren’t there any safer alternatives that could have a similar effect? Solar and wind won’t do it. Do you have any other ideas in mind about future energy policy?
Musk: I think long term, most of civilization’s energy is going to come from solar, and then you need to store it with battery because obviously the sun only shines during the day, and sometimes it is very cloudy. So you need solar batteries. That will be the main long-term way that civilization is powered. But between now and then, we need to maintain nuclear. I can’t emphasize that enough. This is total madness to shut them down. I want to be clear, total madness…
…Döpfner: How is the climate issue going to look like in 15 years? Better than today?
Musk: From a sustainable energy standpoint, much better.
Döpfner: So we are going to solve this problem?
Musk: Yes, absolutely. We will solve the climate issue. It is just a question of when. And that is like the fundamental goal of Tesla.
Döpfner: You once said that the decrease of birth rate is one of the most underestimated problems of all the times. Why?
Musk: Most people in the world are operating under the false impression that we’ve got too many people. This is not true. The birth rate has been dropping like crazy. Unfortunately, we have these ridiculous population estimates from the UN that need to be updated because they just don’t make any sense. Just look at the growth rate last year. See how many kids were born and multiply that by the life expectancy. I would say that is how many people will be alive in the future. And then say, is the trend for birth rate positive or negative? It is negative. That is the best case, unless something changes for the birth rate.
Döpfner: That is also why we need alternatives. You have recently presented Optimus, a human robot, and shared great expectations, what that could do for the world. I assume it is not only about the first visit to Mars that could be done by Optimus, but it might also be a game changer in AI. Could you share this vision?
Musk: With respect to AI and robotics, of course, I see things with some trepidation. Because I certainly don’t want to have anything that could potentially be harmful to humanity. But humanoid robots are happening. Look at Boston Dynamics. They do better demonstrations every year. The rate of advancement of AI is very rapid.
Döpfner: Concretely, Optimus is going to be used in Tesla factories. That is one of the use cases, but what is the broader use case beyond Tesla?
Musk: Optimus is a general purpose, sort of worker-droid. The initial role must be in work that is repetitive, boring, or dangerous. Basically, work that people don’t want to do.
Döpfner: Why has Optimus two legs? Just because it looks like a human being, or is it more practical? I thought four legs were better.
Musk: Haha, four legs good, two legs bad. Kind of reminds me of Orwell. Humanity has designed the world to interact with a bipedal humanoid with two arms and ten fingers. So if you want to have a robot fit in and be able to do things that humans can do, it must be approximately the same size and shape and capability…
…Döpfner: Could you imagine that one day we would be able to download our human brain capacity into an Optimus?
Musk: I think it is possible.
Döpfner: Which would be a different way of eternal life, because we would download our personalities into a bot.
Musk: Yes, we could download the things that we believe make ourselves so unique. Now, of course, if you’re not in that body anymore, that is definitely going to be a difference, but as far as preserving our memories, our personality, I think we could do that.
Döpfner: The Singularity moment that the inventor and futurist Ray Kurzweil has, I think, predicted for 2025 is approaching fast. Is this timeline still realistic?
Musk: I’m not sure if there is a very sharp boundary. I think it is much smoother. There is already so much compute that we outsource. Our memories are stored in our phones and computers with pictures and video. Computers and phones amplify our ability to communicate, enabling us to do things that would have been considered magical. Now you can have two people have a video call basically for free on opposite sides of the world. It’s amazing. We’ve already amplified our human brains massively with computers. It could be an interesting ratio to roughly calculate the amount of compute that is digital, divided by the amount of compute that is biological. And how does that ratio change over time. With so much digital compute happening so fast, that ratio should be increasing rapidly…
…Döpfner: You have solved so many problems of mankind and presented so many solutions. I’m surprised that one topic does not seem to fascinate you as much: Longevity. A significantly increased life span. Why aren’t you passionate about that? Aren’t you personally interested in living longer?
Musk: I don’t think we should try to have people live for a really long time. That it would cause asphyxiation of society because the truth is, most people don’t change their mind. They just die. So if they don’t die, we will be stuck with old ideas and society wouldn’t advance. I think we already have quite a serious issue with gerontocracy, where the leaders of so many countries are extremely old. In the US, it’s a very, very ancient leadership. And it is just impossible to stay in touch with the people if you are many generations older than them. The founders of the USA put minimum ages for a local office. But they did not put maximum ages because they did not expect that people will be living so long. They should have. Because for a democracy to function, the leaders must be reasonably in touch with the bulk of the population. And if you’re too young or too old, you can’t say that you will be attached.
Disclaimer: None of the information or analysis presented is intended to form the basis for any offer or recommendation. Of all the companies mentioned, we currently have a vested interest in Tesla. Holdings are subject to change at any time.
