The Five Best Things: Sept 5, 2020
Exponentials are everywhere, more on AI at scale, and Japan makes moves in the vacuum
Thanks to everyone who read the inaugural issue of the newsletter and provided feedback!
A brief note on why I wanted to do this: Many of you know that I found little of tangible value in business school. However, the experience allowed me to be in the presence of some of best thinkers on a variety of topics. And while I chose to remain within the technology industry after school, I am now able to view it from many different angles. Some of the large themes I’ve been mulling over in the last few years include - how traditional gatekeepers are being dismantled in the internet economy, and the new gatekeepers forming as the benefits of scale accrue in other areas such as AI and Machine Learning. Overlaid on top of all this is the role of macro-economics and geopolitics, in particular of South Asia. I’ve always been fascinated with the time scale impact of seemingly innocuous decisions and policies. This newsletter is an attempt to explore these themes a bit deeper, filtered through the many sources of media I consume. I would love for this to be a dialog and for you all to keep me honest!
Without further ado, here are the top 5 things I read this week -
This week’s Import AI newsletter highlighted a couple of different research paths (Berkeley, Google, and CMU & Facebook AI Research, and DeepMind) which show how a single good enough algorithm or policy, allows a robot to learn on multiple tasks in parallel much faster, compared to attempting to perfect a task.
Continuing on the theme of scale in Natural Language Processing (NLP) last week, there are now signs of how we might achieve scale in robotics with a good enough base algorithm, on top of which domain experts can apply their special sauce.
Back in February (seems so long ago, doesn’t it!) Andreessen Horowitz published an article about how AI companies have much lower margins that software businesses, due to the need for large cloud budgets, several rounds of experimentation, and domain specific customization. Recently they published an update with practical tips for companies to deal with this problem. Solutions include determining when to cut bait on long tail exploration, defining the problem being solved and constraints at hand before choosing the ML technique for it, and being zealously focused on operations.
This aligns very closely with my experience working in ML the last 3 years. Peter Wang, CEO of Austin-based Anaconda got on a podcast with Martin Casado of a16z to explore this further. It was a great, nerdy listen!
The Santa Fe Institute has been the leading research center in the science of complexity the last 40 years. This article by researchers at the institute argues that in the aftermath of the Covid-19 crisis, teaching complexity theory must become mainstream. Linear and uniform distributions can no longer predict extreme events, nor model the outcomes of systems with many large interconnected parts (such as healthcare, supply chains, politics, global travel etc.). We should take inspiration from biological systems which cope, adapt and thrive under such conditions.
This is a topic I got deeply interested in the past couple of years after I read this book about the Sante Fe institute - it’s like a heist movie with scientists - how could I not love it! One of the reasons why Deep Learning works so well is that it captures non-linearities in relationships between entities in complex systems. While the theories have existed for decades, it’s only in the present moment that compute, storage and networking has advanced enough and allows us to model these phenomena cost-effectively. We do our kids a huge disservice by continuing to teach them averages and normal distributions in a world that has always been about exponentials - compound interest, population growth, virality on the internet and in real-life, winner-take-all, and how the downward spiral comes at companies fast and hard if they misstep (more on this below). Exponentials are everywhere.
If someone has tips on how I can explain exponential phenomena to my 4 year old in simple terms, please pass it along! The best idea I have right now is to gift her boy and girl bunnies……
This twitter thread quotes a departing analyst at Sanford Bernstein on how virtuous and vicious cycles develop in the tech industry, and it’s very hard to get off that path. Thus, contrary to the old adage, timing does matter when investing in a tech company.
Staying on the exponential train of thought - I thought this was a very concise articulation of how very few can predict why an idea that’s failed in the past suddenly becomes mainstream. All of a sudden, there is a spark in the primordial ooze and giant corporations appear almost out of nowhere. I’m thinking of this in the context of Zoom this week and its meteoric rise. Video conferencing seemed to be a solved problem until Zoom Apple-fied it (“it just works!”). When you think about all those flights not going anywhere, and the $90B we’ve saved now that we’re not commuting, it sort of makes sense.
This article talks about how Japan is seeking to expand its influence in SE Asia, beginning with Cambodia and Vietnam, to counter China. A range of policies, including diplomatic, economic and defense, are being deployed.
Japan has been in the news a bunch this week, with the 75th anniversary of V-J day, Watten Buffett’s investment in 5 Japanese companies, and Shinzo Abe’s abrupt retirement. A pair up between the high-growth emerging economies in SE Asia, having been dealt a temporary blow due to Covid-19, and the legendary Japanese saver coming out of 30 years of stagflation could yield some very interesting results in the medium- to long-term. A strong US ally to counter China’s outreach in this region.
Honorable mentions -
How the US de-industrialized over decades and lost its manufacturing edge, leaving it unable to deal with the present moment.
How cap-and-trade for carbon emissions might finally take off, with a market starting to form and big-name players participating.
Kalyan jewelers, an Indian jewelry retailer is preparing to IPO. I thought this was notable as this store has become the only place my mom will buy jewelry from anymore, due to the professional experience they provide. Think of it as the Nordstrom’s of jewelry for India’s rising middle- and upper-class, who think of jewelry as an invest-able asset.
This dub over of Nicola Sturgeon, the Scottish premier, admonishing Glasgow partiers is hilarious! I long to return to Scotland when this whole nonsense is over.
If you have some free time this long weekend, do read Ray Dalio’s series of essays on the changing world order. I’ve loved reading his takes over the summer.
Finally, I’ve pasted the free excerpt of an interview with Patrick Collison, the CEO of Stripe below (the article is paywalled).
Stripe Is Not TikTok: An Interview With Patrick Collison
It’s a quirk of Silicon Valley that one of the most seemingly uninteresting businesses is run by one of the most interesting people.
Ten years into running Stripe, its co-founder and CEO, Patrick Collison, still has a voracious appetite for just about everything related to the future of technology—including his payment company.
And while reliably moving around money isn’t as flashy as flying cars or rocket ships, Stripe, which has skyrocketed to a private company valuation of more than $35 billion, is definitely important. If there’s anything this pandemic has shown, it’s the power of digital commerce and the demand for the tools that enable it.
It’s one of the reasons why, while Uber and Airbnb were the poster children for the last wave of the private tech boom, I believe Stripe and other digital infrastructure companies like it will emerge as the next major hits.
But this week, when I sat across the Zoom from 31-year-old Collison, I found a lot more to discuss than payments.
We talked about how the pandemic has affected Stripe, his apparent lack of IPO FOMO, taxes, the Rockefellers, his reading list, Amazon, Twitter and much more. As usual with Collison, he was at times more history professor than CEO, almost allowing me to forget how well he and his team have continued to run Stripe, now almost 3,000 employees strong.
Edited excerpts below.
How has Covid-19 affected Stripe?
Collison: We’ve seen a huge surge in businesses, maybe an existing business that had no online component that’s now trying to figure out what to do online, or maybe that's just completely new online businesses.
E-commerce as a fraction of total consumer spending was actually not that high pre-Covid. The U.S. actually lagged some other countries like the U.K. or China.
Back in January of this year, [e-commerce] was running around 15% of consumer spending. And obviously that was growing with each year. Covid has been a kind of multiyear accelerant.
And so, primarily it’s been keeping up with this sort of huge surge in demand. And then obviously [Covid-19] has had an impact on us as an organization as we figure out how we adapt to be completely remote, which has its own nuances and challenges.
Last year, you announced a large new Stripe headquarters in South San Francisco. What is its future and do you have a longer-term policy around remote work that has come out of this?
Collison: With Covid, so many people in the beginning thought it couldn't possibly last all that long. It was hard to believe the magnitude of the disruption.
Now there’s kind of a subconscious sense that, well, maybe it’ll never go back to normal. The former was clearly wrong. I think the latter was mistaken, too. There will be persistent change in behavior where some people are going to realize they enjoy working remotely more than they expected to or it’s more viable or they’re making some change in their lives. But I do think that’d be a minority. If you were to visit Stripe in three years, if you spent a day at our office, I don’t think you would necessarily know by physical observation that Covid had happened.
Is San Francisco as a tech capital over?
Collison: Pre-Covid real estate costs in California, and San Francisco, were very clearly acting as an increasingly dissuasive force for either tech companies expanding in the region or new companies located there. All the major Bay Area tech companies had either relatively shifted hiring outside the Bay Area or in some cases ceased net hiring in the Bay Area, and most of the expansion was happening elsewhere.
And then you look at [Y Combinator] classes as a longitudinal sample. An increasing practice is those companies being headquartered outside of the Bay Area.
My expectation is it will return to the trend where the Bay Area is a major locus.
Seattle is burgeoning and the locus of increasing tech employment. New York is on the verge of actually growing really strongly. Now versus 10 years ago is incomparable.
Arguably the first VC firm was Venrock, and Venrock was Rockefeller money and Rockefeller was from Cleveland. In [the] 1900s Cleveland was the seventh biggest city in the U.S. and the center of so much manufacturing activity, the nascent automotive industry, chemicals and all the rest.
Then the Depression hit and international competition started to heat up and so on.
San Francisco has somewhat analogous long-term headwinds [with] its version of the local challenges from property prices, and then somewhat similar dynamics and international competition as other people just get good at building technology companies. But I think that’s going to be a multiyear, multidecade story. My expectation is that in 2022 to 2023, its relative position is not really that different. You don’t get many retweets for that, I know.
The other factor people bring up increasingly is the tax situation. Do you see that playing into this?
Collison: The literature, and specifically the analysis of California, says migration does not seem to be that sensitive to tax rates. On the one hand, there are gonna be some proposals introduced for changing tax rates at the state level, but there’s also questions as to sort of what happens to [state and local taxes] and how that plays with with federal level design. How it’s going to evolve is not totally clear.
If you’re talking about people who’ve already made it and who aren’t necessarily tied to an office—successful investors and so on—I think it probably will sway their decisions. But for the people who are actually doing the inventing and the day-to-day work. I think it’ll remain a kind of secondary or even a tertiary factor.
So, don’t you want in on this IPO bonanza? What are you waiting for?
Collison: We’re an infrastructure company, right?
Yes. Those are very hot right now.
Collison: What I tell folks when they’re considering whether or not to join Stripe is the good news and the bad news depends on your disposition. Wherever it is we go or we get to, it’s going to evolve slowly, just by virtue of the nature of the sector.
What’s been so distinct about Amazon is the durability of their compounding. They’re 20%, 30% a year. Now they’re in the third decade. They are a logistics and an infrastructure company. That is the kind of pattern of other companies that operate at lower levels of the stack, as opposed to, for example, something that is entertainment. TikTok is obviously at the center of the news today and has had staggering growth rates.
That is the kind of thing that’s possible when you’re building that kind of product and you do as well as they did.
Just look at the internet itself. The internet actually didn’t grow that quickly in any given year. It has sustained this 30%, 40% compound annual growth rate for many decades at this point.
I don’t assume for a second [we’re] comparing Stripe to either Amazon or the internet in terms of how far we’re actually gonna go. We’re a minnow compared to either of them at this point today. For a platform, a lower-level foundation, these things just tend to evolve more slowly.
As for any sort of IPO, there are core pillars of the product and the functionality we want to build for customers that we just haven’t finished. At some point, it’s likely we’ll either seek to [go public] or have to, but it’s just not a focus right now.
How do you not get bored over time? It’s been ten years. You are at this for the long haul. Do you have a Blue Origin side project or something?
Collison: Not in the slightest. I do tend to think about what I would do if I wasn’t working at Stripe. The things I’m really interested in are the dynamics of entrepreneurship and growth and productivity and innovation. The deployment of technology and how that interacts with globalization and global prosperity is always what I read about in my spare time, and strategy as the center of all these dynamics. If I wasn’t involved at Stripe, I would be trying to find some other way to get back into the same maelstrom.
On the side, I went to college, thinking I’d become a scientist. I never really seriously tried it. It remains something I’m interested in, but I am never now going to be a scientist, and I’m completely fine with that. But it makes for a great sort of adjacent pursuit.
Stripe powers subscription payments for a number of publishers (including The Information). I believe you even started a magazine at one point. What would you like to see from the news business? Oh, and what happened to that magazine?
Stripe has a little publishing arm and magazine. Programming is obviously a pretty big phenomenon in the world today. There were thirteen publications back in the day that really focused on the craft and practice of programming. But for whatever reason there didn’t seem to be as many today.
I love reading domain experts writing about their fields. The field can almost be arbitrary and arcane. But when you get the right experts writing about it, it becomes interesting. There are writers I [have] followed for years now, because you start to vicariously, through them, really come to enjoy whatever it is they’re focused on.
Daniel Swain is the author of a blog called Weather West, a blog about California climate, and he himself is a climatologist. His depth of knowledge and ability to communicate really makes it very compelling. Or Derek Lowe, who writes about drug development and pharmacology and so on.
What I’m describing is generally the writing of current events to at least some degree, but it’s not reporting. And the world also needs to support that. I’m not offering that as a complete solution but rather just a particular dimension I personally really enjoy.
Why do you think the animosity between tech and tech reporters appears to be growing?
Collison: It’s really important that the tech industry does a good job of helping the world understand what it’s doing. And I think we owe that to people around the world more broadly.
In certain ways the relationship between tech and the press appears to be quite adversarial. But in terms of the broader dynamics, I don’t know. It’s easy to complain about this or that, but what was it like in the ’90s? There are all these stories about Larry Ellison or Steve Jobs or Bill Gates calling up every journalist and yelling at them. Are there instances of adversarial behavior today? Absolutely, but obviously the question is how has it changed?
There’s been a lot of introspection about startup culture lately. What do you worry about with Stripe’s culture?
Collison: People at Stripe have always been hungry and efficient, and able to do a lot with a little, and we’ve really avoided a mindset of entitlement and any sense that we deserved any success or usage by any customer. We just have a real determination to build something great. It’s very easy for larger companies to lose that as they grow.
It is possible there is this inevitable arc that dissipates over time. Then maybe you shouldn’t worry about it all that much. You just accept it.
But there are these tantalizing existence proofs that some human agency is possible. Almost everyone in Microsoft seems to agree Microsoft is a more dynamic and effective culture today than it was 10 years ago. And so it’s not just this sort of monotonic dissolution.
And we have tried to, as we grow, remain a warm and caring and supportive place. These are qualities that, just for sociological reasons, can get harder to maintain at more scale. And so, yeah, every week, I’m thinking about them.
What would you do with Twitter if you owned it?
Collison: I’m the last person in the world you should ask that question. We’re a financial infrastructure company. Our specialty is moving money reliably.
Twitter is great. I’ve made so many friendships through Twitter. I’ve learned so much through Twitter. I think Covid in many ways showed Twitter’s tremendous value, as very essential organizations were doing at best a pretty uneven job of keeping us accurately informed. Twitter really enabled us to keep abreast of what was actually happening and what actually mattered. So first, I would try to not mess it up. And then secondly, I would build an edit button to tweets.