180: Berkshire Incentives, Yandex & Google, Twilio, Drug R&D Costs, DeepMind AlphaFold2, Integrated Circuit Design File Format, Early Apple Stories, and Perverse Academia
"that speech about the terminator that Kyle Reese gives to Sarah Connor"
We know the past but cannot control it. We control the future but cannot know it.
🛀 The way to look for what is ‘Lindy’ is not to look at specific products/services, but to look at the human traits and desires that make those desirable.
Jeff Bezos had it right when he talked about how people are always going to want better prices, faster deliveries, and better inventory choices…
✏️ Maybe it’s a cliché because of ‘I, Pencil’, but I was just thinking that I don’t actually know how pencils are made. I should look it up on Youtube… How do they get the round graphite core and the wood to be so tightly assembled..? I can think of some ways to do it, but they doesn’t feel simple enough, there must be a better way…
Update: I looked it up, here it is:
🙍♂️🙍♀️🙍♂️🙍♀️ Friend-of-the-show and supporter (💎🐕) Rishi Gosalia recently mentioned something about a potential investment:
But in the end the bet is on people imo
This made me think about how often that’s what investing is, a lot more often than many people would want to admit... And it’s why developing good taste in people is such a valuable skill to have, IMO.
(Though even if it wasn’t useful for investing, it would still be among the very best life skills to have for obvious reasons. Who you surround yourself with today shapes your future will be to a very large degree, both positively and negatively).
📱If the original iPhone somehow had had a camera system as good as the current iPhone, they may have called it the iCamera.
So much of the engineering efforts, both hardware and software, and so much of the value of the device to users, is camera-centric nowadays that it’s certainly a lot more important than the “phone” part (which at the time it came out still meant something else — today the word ‘phone’ has morphed into something else…).
h/t John Gruber for mentioning something about this on his most recent podcast.
💚 🥃 Thanks to the 5,298 of you who subscribe to this paddle steamboat, and welcome to the show to the 205 new subscribers who got on board in the past 30 days. 👋
If you like what you see, if you feel like you’re learning something and are having a good time, please consider becoming a supporter. It’s quick and painless and costs less than 77-cents per edition:
A Word from our Sponsor: In Practise
Do you want to invest in high-quality businesses? In Practise is a research service for fundamental investors to learn about high-quality companies.
In Practise conducts hundreds of high-quality interviews every year with CEO’s from the best companies globally. All interviews are available on mobile and desktop.
All for the price of only $40 per month.
✨ Sign up here for paid and free plans. ✨
Two great recent interviews that you’ll love:
Investing & Business
How to design management incentives, Berkshire Subsidiaries Edition
Supporter (💚🥃) Adam Mead has a good post where he explains how he thinks Berkshire managers are incentivized to create value while taking into account how much capital is employed (too many companies aren’t so good at that last part):
Let's say a BRK subsidiary uses $100m of capital. Buffett perhaps sets a base salary for the person and then incentivizes them by giving them 10% of any profits over a 15% return on capital. The business earns 20% or $20m and the manager gets a bonus of $500k ($20m minus $15m bogey = $5m x 10% = $500k).
Use of Additional Capital:
If that manager wants an additional $100m, say, they're going to want to be very sure they can earn at least 15% on that capital. If they can't then their bonus could be in jeopardy. If profits went from $20m to $30m but they employed $200m capital, then earnings would just meet the threshold and the manager wouldn't get a bonus. Still with me?
Now let's say that manager is working under the original scenario with $100m in capital but finds a way to do business with $90m. So they've sent $10m back to Omaha but still earn $20m. Well, that's just another way of saying return on capital increased from 20% to 22.2% ($20m / $90m). What does the manager earn for a bonus?
A 15% return would be $13.5m. So $20m - $13.5m = $6.5m. The bonus, at 10%, is $650k. So that manager has earned an additional $150k by figuring out how to earn the same amount of money with less capital. The manager gets $150k more and Berkshire gets $10m to allocate elsewhere.2
Simple, rational, and tied to the business.
This sounds similar to how Constellation Software business units are managed (bonuses based on ROIC+growth above some hurdle, I think), except that in recent times, the decentralization of M&A means that they can’t just send back all of their capital back to the group level or HQ so that their ROICs stay super high, they have to deploy some or all of the capital they generate at the business unit level.
Yandex & Google, Why Search is a Great Business Edition
Friend-of-the-show and Extra deluxe (💚💚💚💚💚 🥃 ) supporter Byrne Hobart has a good free post about Russia’s Yandex and search in general:
I didn’t know the origin of Yandex:
Yandex is actually older than its other search engine competitors; the company dates back to a 1990 project to systematically catalog and search through Russian patent data.
Also a good reminder that it wasn’t always obvious that search would be such a great business model:
Search was not an obviously good business when this was happening, in the late 90s. When online ads were mostly sold by the pageview, the only thing worse for business than a search engine was a good search engine that would promptly redirect people to another site, which would then capture all of those lucrative pageviews.
U.S. Drug R&D Costs, Moore’s Law Edition
It’s not because Mike Pearson liked to talk about something that it’s not something worth paying attention, I suppose. (note the log scale on the Y axis)
My reference to Moore’s Law is about the fact that in the semiconductor world, the price of each node shrink (ie. going from 7nm to 5nm) means exponentially higher costs for R&D, fabs, and equipment required to keep up (which is why basically only TSMC remains at the leading edge).
h/t Alex Stapp
Twilio @ Scuttleblurb
Friend-of-the-show and Extra-Deluxe (💚💚💚💚💚 🥃) supporter David Kim (check out the interview we did) wrote a deep dive on Twilio. I still need to write more about this very interesting company at some point… But this is a great introduction to the company:
Twilio (inexpensive sub $ required)
I recently read Jeff Lawson’s book ‘Ask Your Developer’ and it was pretty good.
As a long-time observer of the space, I didn’t learn a ton from the book. But anyone who’s not a nerd should probably check it out, it’s a good primer on a lot of concepts related to software development and organizational design.
🍌🍌🍌🍌🍌 🦍 No Comment
h/t Modest Proposal (“NFT that pays a dividend to entice the boomers”)
Science & Technology
Integrated Circuit Design File Format Sent to Foundries like TSMC
In the intro of edition #179, I wondered how big the files were that fabless chip designers like Nvidia and Apple send to fabs like TSMC. I still don’t have an exact answer, but reader durga sadan pointed me to the file format (GDSII) and guessed that it’s probably on the order of tens to hundreds of gigabytes.
Initially, GDSII was designed as a stream format used to control integrated circuit photomask plotting. Despite its limited set of features and low data density, it became the industry conventional stream format for transfer of IC layout data between design tools of different vendors, all of which operated with proprietary data formats.
It was originally developed by Calma for its layout design system, "Graphic Design System" ("GDS") and "GDSII".
GDSII files are usually the final output product of the IC design cycle and are handed over to IC foundries for IC fabrication. GDSII files were originally written on magnetic tapes; this step was fittingly called tape out, although that term goes back even farther.
Objects contained in a GDSII file are grouped by assigning numeric attributes to them including a "layer number", "datatype" or "texttype". While these attributes were designed to correspond to the "layers of material" used in manufacturing an integrated circuit, their meaning rapidly became more abstract to reflect the way that the physical layout is designed.
As of October 2004, many EDA software vendors have begun to support a new stream format, OASIS, which may replace GDSII
Looks like the future probably belongs to OASIS, which modernizes GDSII and uses a few tricks to reduce file sizes.
DeepMind paper on AlphaFold2 (Computation Protein 3D Prediction/Design)
It was a *real breakthrough* when they kicked everyone’s butt at CASP14. I can’t wait to see how this tool is further refined, and how it is used around the world to do all kinds of great things for humanity.
we develop the first, to our knowledge, computational approach capable of predicting protein structures to near experimental accuracy in a majority of cases. [...]
In CASP14, AlphaFold structures were vastly more accurate than competing methods. [...]
AlphaFold greatly improves the accuracy of structure prediction by incorporating novel neural network architectures and training procedures based on the evolutionary, physical and geometric constraints of protein structures. [...]
Early Apple Stories & Anecdotes From Those Who Were There
For those of you interested in what it takes to build great products, small company anecdotes, etc, I recommend Folklore.org. It’s a site full of anecdotes about the early days of Apple (many from when they were working on the world-changing original Macintosh).
Here’s for example the origin story of the rectangles-with-rounded corners that have been so prominent in Apple’s history.
Many great anecdotes about a young Steve Jobs — as an aside, there’s a lot more publicly known about him as a 20-something than about his later years, because he became a lot more private over time, so a lot of people assume that he was always the way he was when he was young. He still had very sharp edges in his later years, but he also clearly grew a lot over his life.
Perverse Incentives of the Scientific Publication System
I was listening to an episode of the Cortex podcast and I thought this line by co-host CGP Grey was great, on the topic of the replication crisis:
You have to know that when people’s careers depend on publish papers that other people cite, there is going to be an incentive to make interesting, citable papers, which doesn’t always align with the incentive of accurately describing how the world works.
If you can write a paper that #1 gets a lot of attention, and #2 people in your field will love and want to link to, there’s a big incentive to write that paper, even if it’s not based on anything.
Context for this was recent controversy surrounding Dan Ariely and — ironically — some of his research on honesty…
Speaking of perverse…
The Arts & History
Industrial-Scale, Assembly-Line Oil Paintings
a village in China called Dafen that produces 60 percent of the world’s oil paintings, many of them copies of works by great masters. These are basically fine art factories, with assembly lines cranking out hand-painted copies of paintings by Vincent van Gogh, Leonardo da Vinci, Andy Warhol, and others. The painters work in teams. Each person walks down the aisle of easels and paints a few brushstrokes on each canvas. The next artist adds another element. More than eight thousand artists work in Dafen. They crank out three to five million paintings a year.
Source: Ask Your Developer by Jeff Lawson.
More details on Dafen in this piece.
h/t RabbiJacob for pointing out that there’s even a book about this
30 Year Anniversary?!
I almost can’t believe that 2022 is going to be the 30th anniversary of this album. And if “next year” sounds far away, this is your reminder that 2021 is 75%+ over. I know, I haven’t gotten over 2020 either…
Time is relentless. It’s like that speech about the terminator that Kyle Reese gives to Sarah Connor:
Listen. Understand. That Terminator is out there. It can't be reasoned with, it can't be bargained with...it doesn't feel pity of remorse or fear...and it absolutely will not stop. Ever. Until you are dead.
Back in 2003: "Current generation chips can require up to 40 masks" … "A single mask can require 55 gigabytes of data"