136: Nvidia Q1 Highlights, 33 Bits, Druck & Soros' Adventures, Thoughts on Amazon Unbound, Richard Feynman, Bitcoin Electricity Theft, 1997 Wifi Security Hole, and Mega Drive

"knowing your hometown gives me 16 bits of entropy about you"

If you're having a good time, you already won. 

—Jack Butcher

It doesn't matter
What you create
If you have no fun

—Michelle Gurevich, Party Girl

💸 One of the good — but sometime exhausting — things about investing is that it always keeps you on your toes. Or rather, it does if you have any hope of being any good at it over the long-term

You never feel like you’ve arrived, you never feel like “ok, now I’ve got it, cracked it, it’s going to be easy going forward”... Sometimes it’s like, am I even an investor? should I just index?

But then, I mostly invest because I enjoy the process.

I’m going to do the seemingly egotistical thing and quote myself1 from edition #83:

[Investing is] an endless playground for someone who’s curious and likes to learn about how things work. That you can make money doing it, that makes it practical. But the money is not the reason why it’s interesting.

I can’t spend my days thinking about money and how it’ll be when I sell that stock for a higher price than you bought it at some point in the future. That’s boring. Thats’ just the mechanics of paying the bills, and keeping playing the infinite game.

What’s interesting is the people and technologies and systems and patterns and relationships and puzzles and constant change and trends and counter-trends and dynasties and disruptions and breakthroughs and fallen angels and risen phoenixes…

🛀 It’s often the vivid details that hit us as humans.

One person’s tragic story, with details about their hobbies, where they live, what their kids look like, a closeup photo where you see that familiar look in their eyes, can make the front-page and obsess most of us for weeks, while thousands of anonymous deaths to some faraway tragedy won’t feel as real, even though each one could be turned into that frontpage human-interest story.

I think it’s important to periodically remind ourselves of what others are living through, not lose touch, and do our best to help our fellow humans. (yea, I’m going soft on you)

🌎 🚮 🔥 I’m pretty sure that someday, I’m going to tell my kids about how when I was growing up, we used to take all the trash, and all the recyclables, and all the compostable stuff, put it all in the same plastic bag, and have a big 100-decibel diesel truck billowing black smoke come pick it up and go bury it somewhere.


Oh yeah, not always. Sometimes we just threw it all in a big pile and burned it too ¯\_(ツ)_/¯

🤔 I posted this photo on twitter. We’re trying to grow an avocado tree to show the kids (it won’t grow outside or bear fruit in Canada, but as a potted plant indoors, it can work — my parents did this when I was a kid and it grew to be enormous and curved around the living room, suspended to the ceiling with wires because it was too skinny to support its own weight..).

Anyway, looking at it, I thought of two things:

  1. My countertop looks a bit like Jupiter’s clouds.

  2. A granite countertop has to be like a fingerprint, right?. I wonder if a crime has ever been solved by some opsec person because part of a countertop was showing on a photo, and that was used to confirm that the photo had to have been taken in a certain house...

It also reminds me of this thing about how it takes only 33 bits of information to identify anyone on Earth (a of 2008 — slightly higher population now, but point remains):

there are only 6.6 billion people in the world, so you only need 33 bits (more precisely, 32.6 bits) of information about a person to determine who they are.

This fact has two related consequences. First, a lot of traditional thinking about anonymous data relied on the fact that you can hide in a crowd that’s too big to search through. That notion completely breaks down given today’s computing power: as long as the bad guy has enough information about his target, he can simply examine every possible entry in the database and select the best match.

The second consequence is that 33 bits is not really a lot. If your hometown has 100,000 people, then knowing your hometown gives me 16 bits of entropy about you, and only 17 bits remain. But the real danger is that information about a person’s behavior, which was traditionally not considered personally identifying, can be used to cause serious privacy breaches in a variety of different contexts. (Source)

h/t to friend-of-the-show and Extra-Deluxe supporter (💚💚💚💚💚 🥃 ) Byrne Hobart for this great 33-bit stuff

🍽 Planning to do a 3-day fast this week with a friend (always more motivation when someone does it with you), starting today (ie. last meal was Sunday night).

If anyone wants to join us, feel free to share your progress in the comments. Of course, by publicly talking about it, I’m helping myself make sure I go through with it — gotta make those cognitive biases work for you!

Someday, I’ll write more about both my restricted-feeding routine and my longer water-fasts.

In the meantime, for a good primer on a lot of it, this conversation between Peter Attia and Jason Fung is quite good (I’ve listened to it multiple times).

🛀 If you see someone do something really quickly, and it’s really good, they probably have been doing it for years, so how quickly did they really do it?

Do you only count that part that you saw, or the years of practice? It’s like this:

Jacob Riis: “When nothing seems to help, I go and look at a stonecutter hammering away at his rock, perhaps a hundred times without as much as a crack showing in it. Yet at the hundred and first blow it will split in two, and I know it was not that last blow that did it—but all that had gone before.”

💚 🥃 Thank you for your support, it means a lot.

And if you aren’t part of the club yet, it takes 19 seconds (see, I spent all weekend optimizing everything and shaved off a second — the things I do for you! Oh, and if you’re on mobile and use Apple Pay or Google Pay, it takes closer to 2 seconds):

💙 Subscribe now 💙

Investing & Business

‘Cookie Licking’

That was the unofficial birth of the service that would come to be known first as Prime Air, then Amazon Prime Air, and finally, Amazon Air. The confusion was born from Bezos using the former name in his infamous 60 Minutes interview in 2013 to announce that the company was working on aerial drones to ferry individual packages to people’s backyards. Several operations executives told me they were embarrassed by that stunt (which seven years later, had progressed no further than private tests). They used an old internal Microsoft term to describe it: “cookie licking,” or the act of claiming to do something before you actually do it, in order to capture notoriety and prevent others from following.

Source: Brad Stone, Amazon Unbound

Nvidia Q1 Highlights

I won’t do all the numbers this time, you can look them up if you haven’t seen them yet (“Nvidia is a 28-years-old company that grew revenue 84% and GAAP earnings 106%.”).

Some stuff I found interesting from the transcript:

Channel inventories are still lean, and we expect to remain supply-constrained into the second half of the year.

Makes me curious how they’d be doing right now if they somehow had endless supply…

we expect the RTX upgrade cycle to kick into high gear as the vast majority of our GPU installed base needs to upgrade.

That must be the thing with these generations that both have big performance jumps, but also have new technologies like RTX hit maturity; it takes a while for RTX to get escape velocity, because the first implementation isn’t quite fully baked yet, and then few games support it, and game studios need to learn to use it properly too (it’s like adding a new brush to the artist’s palette — there’s a learning curve until you are fully comfortable with it and get the best results), etc.

It seems like the Ampere generation is that sweet spot where the performance is a big enough jump to pull in a lot of upgrades, and RTX is at the point where it’s ready for prime time, enough games support it, and it looks like game-makers are starting to be more comfortable with it.

So that’s a pretty strong gravitational pull for gamers. (“That’s no moon.”)

We also announced that DLSS is now available in Unreal Engine 4 and soon in the Unity game engine, enabling game developers to accelerate frame rates with minimal effort.

And that is also very important for adoption. This is like having built-in support in both Windows and MacOS. These two platforms cover the waterfront…

(DLSS is Nvidia’s tech to upscale images using deep-learning algos to have better performance at higher resolutions, basically “faking” very convincingly that higher rez in a way that is less computationally-intensive for the GPUs than actually running the game at that rez)

We estimate that about 75% of GeForce gamers play eSports games, and 99% of eSports pros compete on GeForce.

99%? Can it really be that high?

Our GeForce NOW cloud gaming platform passed 10 million registered users this quarter. GFN offers nearly 1,000 PC games from over 300 publishers, more than any other cloud gaming service, including 80 of the most popular free-to-play games. GFN expands the reach of GeForce to billions of underpowered Windows PCs, Macs, Chromebooks, Android devices, iPhones and iPads. GFN is offered in over 70 countries with our latest expansions including Australia, Singapore and South America

This is an interesting part of the business that I don’t hear that much about, but I think it’s such a win-win.

By expanding who can play games (to people with non-gaming hardware), they are doing a solid to game makers, and they benefit both from the subscription revenue and by having this other huge sink for their hardware to be deployed into, increasing their economies of scale and making their platform an even bigger target for game-makers to optimize against (using RTX ray-tracing, DLSS, etc).

I mean, that last one probably won’t make a huge difference since they’re already huge, but at the margin, it can’t hurt, and does improve the competitive positioning from yet one more angle.

And it’s one more thing where they’re competing on both the software/hardware/ecosystem rather than just on the hardware as a more “pure” semi-player would, and gets them more direct relationships with gamers.

Our automotive design win pipeline now exceeds $8 billion through fiscal 2027. Most recently, Volvo Cars announced that it will use NVIDIA DRIVE Orin, building on our next great momentum with some of the largest automakers, including Mercedes Benz, SAIC and Hyundai Motor Group. [...]

In robotaxis, we added GM Cruise to the growing number of companies adopting the NVIDIA DRIVE platform, which includes Amazon Zoox and DiDi. 

We've also had a great traction with new energy vehicle makers. Our latest wins include Faraday Future, R Auto, IM Motors and VinFast, which joined previously announced wins with SAIC, NIO, Xpeng and Li Auto. In trucking, Navistar is partnered with TuSimple in selecting NVIDIA DRIVE for autonomous driving, joining previously announced Volvo Autonomous Solutions and Plus.

A billion here, a billion there, and pretty soon you have a real business.

Their new automotive SoC is pretty impressive too, at least when it comes to raw compute specs:

Atlan, which targets automakers' 2025 models, will follow the NVIDIA DRIVE Orin SoC, which delivers 254 TOPS that has been selected by leading vehicle makers for production time lines starting next year. [...] Atlan will deliver an unrivaled 1,000 trillion operations per second to performance and integrate data center class NVIDIA BlueField networking and security technologies to enhance vehicle performance and safety, making it a true data center on wheels.

A thousand tera-OPS here, a thousand tera-OPS there…

Pretty soon you have a real data-center on wheels.

On our Arm acquisition, we are making steady progress in working with the regulators across key regions. We remain on track to close the transaction within our original time frame of early 2022.

I’m still not sure if this one will be able to get to the finish line, but I hope so, because I feel like ARM technology will become more interesting if owned by Nvidia.

We're democratizing AI. We're bringing it out of cloud. We're bringing it to enterprises, and we're bringing it out to the edge. And the reason for that is because the vast majority of the world's automation that has to be done has data that has data sovereignty issues or data rate issues that can't move to the cloud easily. And so we got -- we have to move the computing to their premise and oftentimes all way out of the edge.

I’ve been hearing more and more about geo-compliance/data-sovereignty/etc lately.

Seems like one of the major trends of the coming decade, one that will reshape large parts of the internet.

This last part does sound a lot like their recent partnership with Cloudflare.

You know that the world's cloud data centers are moving to deep learning because every small percentage that they get out of predictive inference drives billions and billions of dollars of economics for them. And so the movement towards deep learning shifts the data center workload away from CPUs because accelerators are so important. And so hyperscale, we're seeing great traction and great demand.

That’s the thing, right. At the scale that many of the big tech companies operate, small improvements in performance represent gigantic amounts of money, so there’s quite the incentive to make it happen.

Jensen had a good riff about why data-centers are becoming more and more software-defined, and what that means for DPU-type accelerators:

you don't want a whole bunch of bespoke custom gear inside the data center. You want to operate the data center with software. You want to be software-defined.  [...]

depending on how you want to think about it, how much security you want to put on it, if you assume that is a zero-trust data center, probably half of the CPU cores inside that data center is running not applications. And that's kind of strange because you created the data center to run services and applications, which is the only thing that makes money.

The other half of the computing is completely soaked up running the software-defined data center just to provide for those applications [...]

so you fundamentally want to change the architecture as a result of that to offload the software-defined virtualization and the infrastructure operating system, if you will, and the security services to accelerate it [...]

Separating it doesn't make it more effective. And so you want to offload that and take that application and software and accelerate it using accelerators, a form of accelerated computing.

And so that's -- these things are fundamentally what BlueField is all about.

BlueField-2 replaces approximately 30 CPU cores. BlueField-3 replaces approximately 300 CPU cores just to put -- give you a sense of it.

If you think about how many datacenters there are in the world, both core cloud and on-prem, and how they’re all moving to that more software-define/zero trust paradigm, that’s going to be a big market for these DPU-type accelerators.

Druck & Soros’ Adventures

These two would make good characters in a buddy-cop movie. Druckenmiller on concentration in good ideas:

[In 1992] when I went in to tell Soros that I was going to short a 100% of the fund in the British pound against the Deutschmark, he looked at me with great disdain. 

He thought the story was good enough that I should be doing 200%, because it was sort of a once-in-a-generation opportunity.  (Source)

Soros is clearly bad cop in the good-cop/bad-cop routine..

h/t Mark

How to value ‘confidence’ and ‘years of cumulative knowledge’?

People underestimate the value of accumulated knowledge and confidence in a company.

Replacing an investment that you’ve followed and learned about for years and have very well-calibrated confidence in for another investment that is — for the purpose of this thought-exercise — equal in business quality and stock valuation still isn’t a fair trade.

You can’t buy having been around for years, seen how management reacts in various situations, how they deliver (or not) on what they promise, how resilient the business is to shocks, etc. You can try to backfill that info with research, but it’s never the same, in my experience.

My Thoughts on ‘Amazon Unbound’ by Brad Stone

You know how sometimes you read a book, and it’s bad, and that’s clear.

But other times, you read a book, and it’s good, but it’s still not the book you wanted it to be, it’s not the book you wanted to read, so it doesn’t quite satisfy fully, even if it feels a bit unfair to blame the book for that.

Well, that’s been my experience with Amazon Unbound.

For my own taste, there’s not enough about AWS despite it being incredibly important to the company and the world.

There’s also not enough on Andy Jassy, both because of his importance to AWS (and thus the world), but also because he’s now the admiral of the whole fleet (and not just captain of the USS AWS aircraft carrier group).

Also for my own taste, there’s too much on whatever made headlines in the press, PR gaffes, politics, and who was offended by what, and not enough about the on-the-ground stuff that Amazon actually does that is impressive, the large-scale systems that have solved novel e-commerce logistics problems that previous era large companies like Walmart didn’t have, etc.

There’s also almost nothing on their software, and the thousands of engineers who create it; it’s an important part of the business that is crucial to understand to get what made Amazon so successful.

We may see the atoms, but it’s the bits that organize the complex dance that make it all happen.

Large sections of the book are also based on anecdotes or what someone claimed, or a reputation for X, with little additional facts or context brought to the table (ie. When you have a million employees, what’s the base rate on having X or Y happen?).

Maybe the author couldn't get much info on certain parts of the business and that’s why they’re missing, I’m certainly not assigning blame. Just saying what I wish I had more of.

What if AWS had come first?

Sometimes I wonder what would’ve happened if somehow AWS had been first…

Brands matter, and with Amazon, everybody is thinking about retail first.

What if AWS had been around for years and well-known as a cloud-giant, and then grown a huge retail arm? It’s an interesting thought-experiment, IMO.

💙 Subscribe now 💙

Science & Technology

Atoms… a less cartoony rendering

h/t Friend-of-the-show and supporter (💚🥃) Brad Slingerlend

1997 Wifi Vulnerability Only Discovered Recently

Or at least, the discovery has only been made public recently… Who knows who else knew about it and kept it private, possibly for decades.

One more for the “security is hard” file:

A Belgian security researcher has discovered a series of vulnerabilities that impact the WiFi standard, with some bugs dating back as far back as 1997 and affecting devices sold for the past 24 years. [...]

“Three of the discovered vulnerabilities are design flaws in the WiFi standard and therefore affect most devices,” said Mathy Vanhoef, the Belgian academic and security researcher who found the Frag Attacks.

“Experiments indicate that every WiFi product is affected by at least one vulnerability and that most products are affected by several vulnerabilities”

More details here, including a video of Vanhoef’s presentation at the USENIX security conference.

Richard Feynman’s Lectures on Physics at Caltech

For those needed a dose of RPF:

These are the tape recordings of Richard Feynman's 1961-64 Caltech Introductory Physics lectures, which form the basis of the book The Feynman Lectures on Physics. [...]

We present entire lecture tapes without any editing or enhancement, including the tape leader. Parts of some lectures edited out of the commercial versions of these recordings are preserved here intact. Recorded material outside the lectures, including discussions between Feynman and his students and/or colleagues, never previously published, can be found in this publication. Three entire lecture recordings never heard before outside Caltech, including two lectures on Quantum Mechanics Feynman gave in 1964, are also included in this publication. [...]

Lecture #30 ‘Interference’, given on February 20, 1962, is of some historical interest: Feynman delays starting by 6 minutes because earlier that day John Glenn became the first American (and second man, after Yuri Gagarin) to orbit Earth, and he was due to splash down! So, there is a prolonged "before-lecture" discussion in the recording. In the background you can hear the excited hubbub of the students, concerned for Glenn's safe return. Feynman begins his lecture, "Mr. Glenn is in orbit and he'll probably come down during this lecture. We'll see if we get any news from it."

Here they are. h/t Rishi Gosalia (✨💚🥃✨)

Bitcoin Miner Discovered Stealing Electricity, Busted by Police on Cannabis-Farm Hunt

Officers had been tipped off about the site on the Great Bridge Industrial Estate, Sandwell, and raided it on 18 May, West Midlands Police said.

Instead of cannabis plants they found a bank of about 100 computer units [...]

"It had all the hallmarks of a cannabis cultivation set-up and I believe it is only the second such crypto mine we have encountered in the West Midlands," she said. (Source)

Electricity theft has always been a thing, either by plugging where you’re not permitted to or modifying the meter. But historically, it was harder to turn stolen electricity directly into cash.

You could offset a bill up to a point, or use it to run commercial equipment that provided a revenue, but crypto-mining is probably the fewest-steps-removed way of turning electricity into money.

Well, as long as there’s proof-of-work cryptocoins out there… Doesn’t matter nearly as much with the less energy-intensive proof-of-stake stuff (I wrote about Ethereum’s move from 🔥 to 🥶 in edition #131).

The Arts & History

Neuroframe by Mega Drive (2021)

I realized I haven’t really featured much — if any — electronic music so far.

Well, why not? I’ve long ago decided that the only two genres I made a distinction between were “good music” and “bad music”.


I’ve been really enjoying the new Mega Drive album recently, so it’s my recommendation to you today.

If you are the kind of person who turns-on-your-heels and slams-the-door after the words “electronic music”, here’s my pitch to you:

This isn’t electronic dance music.

This is more like a retro-futuristic soundtrack to an imaginary video game or 1980s movie, but with 2021 production. And it isn’t just about rhythm, there’s plenty of melody.

You may still not like it, but I encourage you to give it a (virtual) spin:

‘Why I’m Quitting Tobacco’

Recently watched the Mad Men episode where Don writes the “Why I’m Quitting Tobacco” letter (S4E12)….

When you don’t like what they’re saying about you, change the conversation; when you’re losing, try to increase volatility in case it moves your way.

Classic episode.

Sad how Midge turned out, and how Don’s reflex is basically the same that he had with his brother.

Poor Sally. Betty’s repeating how her mother treated her.

One of the tragedies of humanity is how someone’s bad decisions can echo for generations and generations (though the converse is that the good decisions you make now may echo for generations of your descendants, so choose wisely — be a good ancestor!).


But hey, what is writing if not quoting yourself, the voice that is in your head, eh? Bet you have never thought of it like that!