302: Constellation's 714 Acquisitions Since 2005, Ringelmann effect, Microsoft Azure Troubles, Netflix Austerity, Predicting AI Progress, and Gaming $$$

"Our mind loops back and tries to model itself"

There is a tendency to think we did everything right when we succeed, when often we only made the next-to-last mistake.

—Garry Kasparov ♜

🔧👨🏻‍🔧🧰👷‍♂️🪚 As someone who has made a living largely as a symbol manipulator (as friend-of-the-show and supporter (💚 🥃) Jim O’Shaughnessy puts it so well), I sometimes find myself daydreaming about how satisfying it must be to earn a living doing something a bit more concrete.

I don’t think I *really* want to do these things all day. Like anything, it’s trade-offs.

But if you’re a carpenter or cobbler or car mechanic, there must be some satisfaction is going from nothing or a broken thing all the way to something useful.

It feels like honest work.

A few weeks ago, I repaired my clothes dryer by replacing the broken heating element, and it felt very different from writing or investing. Maybe the fact that I do that kind of stuff relatively rarely has a novelty factor that would wear off if it was a daily occurrence, but I certainly encourage you to give that kind of construction/repair work a try if you don’t do much of it.

📱🔐 Here’s a good-to-know tip if you have an iPhone.

Since our mobile devices have become largely extensions of our brains and contain a lot of info about us, it’s important to keep that info secure.

You may think that biometric authentication is pretty bulletproof, but in at least the US, and probably many other jurisdictions, it’s a lot easier for authorities to get you to put your thumb on the fingerprint scanner or your face in front of the phone than to compel you to give information that is inside your brain (passwords and PINs), thanks to protections against self-incrimination.

How can you quickly make your phone go from biometrics-unlock to password/PIN-unlock?

If you squeeze the power button on one side and one of the volume buttons on the other and press for 2 seconds, the phone will go in ‘hard-lock’ and biometrics will be disabled until you’ve entered your passcode again.

Probably a good idea to do whenever people you don’t trust may get their hands on your phone (airport security, getting arrested by police, etc). John Gruber (🍎) has more details on this.

🔁 🧠🫀🫁 Brian Norgard tweets:

We live inside feedback loops.

That’s correct, but also *we are feedback loops*.

The more you learn about biology, the more you realize that life happens within certain ranges, and homeostasis is what keeps us within those ranges. ie. Too much electrolytes and some processes are triggered to lower them, not enough and the reverse takes place, your body’s temperature gets too high or too low, you get dehydrated or drink too much water, your blood glucose gets too high or too low, etc.

Not only that, but when it comes to us as sentient beings, it seems like we pretty literally *are feedback loops*, in a Douglas Hofstadter I Am a Strange Loop kind of way.

Our mind loops back and tries to model itself, which is pretty special.

It appears that only very social animals developed the capacity to model the minds of other individuals in their groups, and that over time this capability was turned inward, on our own minds, further increasing our mental capabilities.

Not something we can observe in real-time, but that’s a theory that has explanatory power to me... ¯\_(ツ)_/¯

💚 🥃 For the price of one alcoholic drink, you get 12 emails per month (plus 𝕤𝕡𝕖𝕔𝕚𝕒𝕝 𝕖𝕕𝕚𝕥𝕚𝕠𝕟𝕤 and podcasts!) full of eclectic ideas and investing/tech analysis.

That’s pennies per edition for the Serendipity Engine!

If you make just one good decision per year because of something you learn here (or avoid one bad decision — don’t forget preventing negatives!), it'll pay for multiple years of subscriptions (or multiple lifetimes).

You can even sign up in single digits seconds on your phone with Apple/Google Pay by clicking the button below (if you don’t see paid options, it’s because you’re not logged into your Substack account):

---

🏦 💰 Liberty Capital 💳 💴

✨ Constellation Software’s ✨ 714 Acquisitions since 2005

We’re only in early July and they’ve already deployed almost as much capital as in all of 2021, which itself was about double the biggest year before that. 🚀

The decentralization and scaling of capital deployment appears to be humming along! Topicus is doing well too.

h/t LC Investing

The Ringelmann effect 🗄🗄🗄🗄📉

This is one that I think most of us understand intuitively, but it’s nice to have a name to put on it (the more modern version when it comes to software is Fred Brooks’ Mythical Man-Month and Brooks’ law):

The Ringelmann effect is the tendency for individual members of a group to become increasingly less productive as the size of their group increases.

This effect, discovered by French agricultural engineer Maximilien Ringelmann (1861–1931), illustrates the inverse relationship that exists between the size of a group and the magnitude of group members’ individual contribution to the completion of a task.

The reasons for the effect seem to come from both loss of motivation, and loss of coordination as the group becomes bigger.

When your own particular contribution becomes less identifiable, chances are you won’t work as hard. And it also makes it easier for freeloaders to blend in…

While studying the relationship between process loss (i.e., reductions in performance effectiveness or efficiency) and group productivity, Ringelmann (1913) found that having group members work together on a task (e.g., pulling a rope) actually results in significantly less effort than when individual members are acting alone.

Ringelmann discovered that as more and more people are added to a group, the group often becomes increasingly inefficient, ultimately violating the notion that group effort and team participation reliably leads to increased effort on behalf of the members

The coordination costs are also very real. Regardless of the ‘pool of skills’ that are available, if they can’t be effectively coordinated, you won’t get a productive use out of them.

It’s a bit like if you have a car with a very powerful engine, but there are problems with the transmission and the tires. You may not get much forward movement… 🚙

Microsoft Azure facing hardware supply issues, capacity running in the “yellow” 🚦☁️

Interesting piece at The Information (Thanks N.S.!) about how tight things are getting at some Azure datacenters (and possibly also at AWS and Google, though this article is based on conversations with MS employees):

as of early June Microsoft’s internal rating for Azure capacity globally was yellow, a step down from the green rating it designates for data centers that are operating with normal levels of reserve server capacity [...]

more than two dozen Azure data centers in countries around the world are operating with limited server capacity available to customers, according to two current Microsoft managers contending with the issue and an engineer who works for a major customer. And in more than half a dozen Azure data centers—including a key one in central Washington state and others in Europe and Asia—server capacity is expected to remain limited until early next year, said one of the Microsoft managers.

The piece says that MS has even been shifting some of its own internal cloud use around to help spread demand around and free up capacity at the most bottlenecked data-centers.

The capacity issues appear to stem from a number of factors. The main one is a global shortage of chips and other components that has made it harder to get new hardware, the current managers say. Microsoft isn’t alone in this regard, as Amazon Web Services and Google Cloud have also been dealing with server capacity shortages in some of their data centers, said a person who works for a customer of both cloud providers.

But the chip shortage seems to be hitting Microsoft harder because it was struggling to provide enough capacity to Azure customers long before the pandemic, and because it has a large number of aging servers, storage and networking equipment that are due for upgrades, said one of the Microsoft managers.

Victim of its own success because it has been growing so fast? Bad planning? Bad luck because of the pandemic? Probably a mix, but I don’t know the ratio…

🇩🇪 Germany facing a huge gas shock to its economy

Well, you paint yourself in a corner, things may get messy…

It took years for Putin shills like Gerhard Schröder (and many others) to make Germany as dependent on Russia as possible, and now comes the time to pay for those mistakes:

“Because of the gas bottlenecks, entire industries are in danger of permanently collapsing: aluminum, glass, the chemical industry,” said Yasmin Fahimi, the head of the German Federation of Trade Unions (DGB), in an interview with the newspaper Bild am Sonntag. “Such a collapse would have massive consequences for the entire economy and jobs in Germany.”

Check this out:

Ouch!

That’s why I won’t be surprised if a new LNG export terminal is soon announced on Canada’s East coast, and hopefully Germany also corrects its major blunder of getting rid of its 12 nuclear power plants by keeping the last 3 operational and reviving as many of the others as possible (yes, it’d be very hard, but there are no easy choices now).

🚚 📦📦📦 📫 FedEx going all-in on the cloud ➡ ☁️

FedEx’s CIO:

“We’ve been working across this decade to streamline and simplify our technology and systems,” he said. “We’ve shifted to cloud...we’ve been eliminating monolithic applications one after the other after the other...we’re moving to a zero data center, zero mainframe environment that’s more flexible, secure, and cost-effective.”

“Within the next two years we’ll close the last few remaining data centers that we have, we’ll eliminate the final 20 percent of the mainframe footprint, and we’ll move the remaining applications to cloud-native structures that allow them to be flexibly deployed and used in the marketplace and business. While we’re doing this, we’ll achieve $400 million of annual savings.”

The company is already a customer of Microsoft Azure and Oracle Cloud, so it’s likely that they’ll move their stuff there.

How big is the gaming industry? 🎮 👾 💰💰💰

Let’s compare it to cinema: In 2019, the global box office was $42.2 billion.

Tencent and Sony, when taken together, are significantly bigger than that!

h/t Ho Nam

📺 From TV Content Tsunami to Austerity 🎥 🎬

After years of massive spending and growth in content production, the TV world appears to be shifting gears:

The sudden decline in Netflix’s share price and the growing fear of a recession have forced Hollywood into a new period of fiscal austerity. This manifests in ways big and small. A TV director who made $4 million a year is now getting $750,000. Mid-budget movies are being shelved. Broadcast TV budgets have dropped more than 30%.

In the latest new trend, networks are canceling shows that they already agreed to make. Peacock shelved plans for a “Field of Dreams” TV series from the co-creator of “Parks and Recreation,” while HBO Max canceled “Demimonde” from “Lost” co-creator J.J. Abrams.

Too bad, I kinda enjoyed the tsunami of content.

I’m not watching more than before, but if I have time to watch 3-4 hours a week, I’d rather pick the best 3-4 hours that I can from a larger absolute number of shows.

More shots on goal should in theory result in a better quality of shows that make the cut for me (though it’s always hard to know if it works like that in practice, and what the dilution effect if the number of shows grows faster than the talent pool of actors, editors, directors, artists, craftspeople, etc).

This is a significant departure from the last few years when media companies tripped over themselves to produce any halfway decent idea. The industry made 559 scripted shows last year, up more than 200 since 2013, the year “House of Cards” debuted. That doesn’t include all of the unscripted programs, including competing docuseries on the same subject.

Who’s the last player standing?

Apple is just about the only place that can still spend like money grows on trees.

Source.

---

🧪🔬 Liberty Labs 🧬 🔭

🌏 Global Population Density in 3D 👨‍👩‍👧‍👦👨‍👩‍👧‍👦👨‍👩‍👧‍👦👨‍👩‍👧‍👦👨‍👩‍👧‍👦👨‍👩‍👧‍👦

This viz was created by Alasdair Rae. You can click the image above for larger versions. He explains how he did it here.

Predicting AI Progress

You may think AI progress is fairly easy to predict.

We can extrapolate compute with Moore’s Law or by looking at GPU roadmaps, and we can look at the growth in the size of existing models (number of parameters, dataset size) and have a good idea of where we’ll be in X years.

But it turns out (IT TURNS OUT! what a lot of weight these few words have been carrying in recent years… They must be tired and looking forward to retirement):

People seem to be continually surprised, over and over again, by the new capabilities of big machine learning models, such as PaLM, DALL-E, Chinchilla, SayCan, Socratic Models, Flamingo, and Gato (all in the last two months!).

Luckily, there is a famous paper on how AI progress is governed by scaling laws, where models predictably get better as they get larger. Could we forecast AI progress ahead of time by seeing how each task gets better with model size, draw out the curve, and calculate which size model is needed to reach human performance?

I tried this, and apparently the answer is no. In fact, whether AI has improved on a task recently gives us exactly zero predictive power for how much the next model will improve on the same task. The sheer consistency of this unpredictability is remarkable, almost like a law of statistical thermodynamics. No matter what I plug in, the correlation is always zero!

More here.

🏴‍☠️🇨🇳 Police database hack: Info of “as many as a billion Chinese residents” possibly leaked 😬

One of the downsides of centralization is that central authorities then become an attack vector for everyone and everything, either from that authority itself, or from others who go through it:

Unknown hackers claimed to have stolen data on as many as a billion Chinese residents after breaching a Shanghai police database, in what industry experts are calling the largest cybersecurity breach in the country’s history.

The person or group claiming the attack has offered to sell more than 23 terabytes of stolen data from the database, including names, addresses, birthplaces, national IDs, phone numbers and criminal case information

1 billion people!

It may be the biggest disclosed hack so far (though let’s remember that what we hear about and what takes place are two very different things, and there’s more transparency in the West than in places like China and Russia, where a lot of cyber-attacks originate from — but they’re also likely big targets).

Domestic breaches are however rarely disclosed because of a lack of transparent reporting mechanisms. In 2016, personal information on dozens of Communist Party officials and industry figures from Jack Ma to Wang Jianlin was said to have been exposed on Twitter, in one of the country’s biggest online leaks of sensitive information at the time. In 2020, the Twitter-like service Weibo Corp. said hackers claimed to have stolen account information for more than 538 million of its users, though sensitive data such as passwords was not leaked. And this year, tens of thousands of seemingly hacked files from China’s remote Xinjiang region provided fresh evidence of the abuse of mostly Muslim ethnic Uyghurs, according to a rights group.

Source.

---

🎨 🎭 Liberty Studio 👩‍🎨 🎥

Find your learning patterns, don’t just blindly follow what school gave you 👩‍🏫🎓📝🕵️‍♂️

I’m going to put this here, because the most interesting aspect is the meta-point about learning and developing your craft, not specifically about mathematics.

First, the story of June Huh, who won the Fields Medal (aka the Nobel Prize in math).

His path to getting there includes dropping out of high school to become a poet, and it was only a chance encounter in his sixth year of university that brought him back to math.

But:

That poetic detour has since proved crucial to his mathematical breakthroughs. His artistry, according to his colleagues, is evident in the way he uncovers those just-right objects at the center of his work, and in the way he seeks a deeper significance in everything he does. “Mathematicians are a lot like artists in that really we’re looking for beauty,” said Federico Ardila-Mantilla, a mathematician at San Francisco State University and one of Huh’s collaborators. “But I think in his case, it’s really pronounced. And I just really like his taste. He makes beautiful things.”

“When I found out that he came to mathematics after poetry, I’m like, OK, this makes sense to me,” Ardila added.

Here’s someone else’s journey, getting discouraged by the rigid educational system only to have to find his own way:

• Shaun Maguire talks about his school experience and how he really learned.

I really like the point he makes at the end of the video: To really learn something, the best way probably isn’t to do constant context switching — 1 hour of English, 1 hour of Math, 1 hour of history — but rather by intense focus, like spend four days straight obsessing about something, trying to make it work.

Noah Smith 🐇🇺🇦 @Noahpinion

There's so much value in letting people explore what they like, find themselves, and develop as human beings before they hurl themselves into a career. I am glad that America's education system allows for that kind of exploration.

You’ll probably learn more by intensely concentrating on one thing that really interests you than by constantly nibbling on various things that someone else is telling you to work on — intrinsic motivation is hugely important to learn anything!

That’s why teaching people how to learn and follow their curiosity is probably the most important meta-skill that a school could teach you (right after reading/writing and arithmetics, probably).