539: Jim O'Shaughnessy's New Book, Buffett & Munger AGMs, Jeff Bezos, Perplexity, AI Funding, Modest Proposal, ASML, Sora, Sun vs Compost Heap, and Flight Simulator

"spend time with smart friends from all eras"

Dec 11, 2024

The foundational challenge in design is that it's easy to make things hard and hard to make things simple.

—David Holz

🚨🎁📕📖🔔 I’m SO, SO happy to show you this new book 😃

With ‘Two Thoughts: A Collection of Infinite Wisdom’, Jim O’Shaughnessy and Vatsal Kaushik curated a remarkable ✨ constellation ✨ of eclectic thinkers, wordsmiths, maestros, philosophers, innovators, athletes, artists, entrepreneurs, founders, etc.

What makes this book special? It's designed for serendipity — open it to any page for an instant dose of inspiration guaranteed to make you think and smile. It begins with a great introductory essay by Jim about the Tinker & the Prover that resides in all our minds.

They deliberately avoided the dry, serious tone of dusty old texts. Instead, it’s alive and vibrant, a way to spend time with smart friends from all eras and have an active conversation mind-to-mind through the magic of the written word 🧠💬🧠

And the physical book itself? It’s a real BEAUTY! 🤩

The video gives you some idea, but it’s better in person. It’s a very tactile artifact — the quality of the paper, the UV spot highlights on the cover, the black cloth quarter-spine, the orange ribbon, and countless meticulously crafted details that were sweated over…

I own a lot of books, but when I look across my bookshelf, this is in the top 3 nicest books I own. Perfect for book lovers, I suspect it’ll rank up there for you too!

Keep an eye out for playful surprises—including an unexpected appearance on the copyright page! (you can see it in the video) 👀

It’s coming out very soon and you ⭐️ can pre-order it today ⭐️ (it’s a very giftable book)

Special thanks to my friend S. for capturing this video of the first hardcover copies from the Italian printer. It was a real joy for me to work on this book with the rest of my Infinite Books/OSV teammates.

🦋🕸️⛓️ If you haven’t joined me in the bunker yet, come say hello 👋

Liberty.bsky.social

There are more familiar faces arriving each day. Here’s a Fintwit starter pack curated by Terminal Value.

I still think the odds favor Twitter/X, but as I said, you build the bunker and stock it with canned food before you need it.

It’s not a bad idea to replicate at least part of your social/interest graph over there, especially if you prefer the old Twitter (Discovery/Discussion/Curation) to the new one (Memes/Politics/Engagement Bait from people you don’t follow…).

Think of it as your social media insurance policy—worth setting up while things are calm.

⚾️📘🔎🐐🐐 Congrats to friend-of-the-show and long-time supporter Alex Morris (💚 🥃 🎩) on his new book, three years in the making:

I instantly pre-ordered it!

🏦 💰 Liberty Capital 💳 💴

📊 The Great Inflation/Deflation Divide: 2024 Update 📈📉

I still think this is one of the most important charts out there, and I wish entrepreneurs and founders used it as a guide for what to work on (but I understand the barriers to that…).

The crucial observation: sectors that achieve high productivity and price deflation (think electronics, or even agriculture over the long-term) tend to become a smaller part of the overall economy, while less productive sectors with rising costs (like healthcare and education) grow to occupy larger shares.

This means that to move the needle, we have to attack the worst of the worst, the biggest slices of the pie. We can’t squeeze much juice out of making televisions even cheaper than they already are… What % of the average household income is spent on internet access or software? It’s tiny compared to healthcare and housing.

However, there’s something important to keep in mind while looking at such charts, “hedonic adjustments”:

For goods like televisions where the features and sometimes fundamental qualities of the product are changing all the time, the BLS makes “hedonic quality adjustments.” Put simply, when a product that the CPI trackers are observing is either updated or replaced entirely by a clearly superior version, the BLS adjusts its price point based on the new model and its enhanced features, taking the improvements into consideration as part of their assessment of whether the price is rising or falling. In the case of TVs, that means that things like larger displays, higher definition, and smart compatibility all likely weighed on the Bureau’s measurement through the years.
“Obviously televisions don’t literally cost 2% of what they used to,” quality adjustments are major reasons that goods like televisions and computers have trended in the opposite direction to many of the items in the CPI basket. Indeed, while you could quite easily go to any retailer on the market in 2024 and pick up a TV for as much as you’d have paid in 2000, the BLS’s measure accounts for the technology’s advance over the last couple decades, meaning that the standard of the set you could get for the price point has improved massively.

Source.

🛒📦🚚 A Rare Jeff Bezos Interview and It’s Good! 🤖🚀

Some highlights:

Q: Throughout your whole career at Amazon, it appears to me like you paid yourself about $80,000 a year in total cash comp. And never took additional equity in the company the entire time.
Jeff Bezos: No, I never did. And I asked the comp committee of the board not to give me any comp. My view was I was a founder, I already owned a significant amount of the company, and I just didn't feel good about taking more. I felt I had plenty of incentive.
I owned more than 10% of the company, more early on... before it was diluted by various things-more than 20% of the company. I just felt, how could I possibly need more incentive.
Most founders own big chunks of the company. They're more like owner operators. The way they increase their wealth is not by getting more equity. They just want to make the equity they have more valuable.
I just would have felt icky about it. I'm actually very proud of that decision. I sometimes wish that there were — maybe you can do this, Andrew, or maybe the Washington Post will do this. But somebody needs to make a list where they rank people by how much wealth they've created for other people. And so instead of the Forbes list- it ranks you by your own wealth...
So you know, Amazon's market cap is $2.3 trillion. Today I own about $200 billion-ish of it. So if you take $2.3 billion and subtract out the piece I kept for myself, then I've created something like $2.1 trillion of wealth for other people. That should put me pretty high on some kind of list. And that's a better list.
How much wealth have you created for other people? Like Jensen at Nvidia. He's going to be very high on that list. And that would be a pretty cool list. Somebody should do that list.

Imagine how much richer Bezos would be today if for the past 30 years he had gotten a more “normal” CEO comp (boatloads of options and RSUs). Such grants from the first 10 years of Amazon could be worth tens of billions now.

On what he’s doing these days:

Q: now what is it that you're doing at Amazon?
Jeff Bezos: AI. There are a few things but small, but it's 95% AI.
You have to remember, modern AI is a horizontal enabling layer. It can be used to improve everything and will be in everything. This is most like electricity.
I went to a brewery in Luxembourg many years ago—a trip that was actually one of the little catalysts for founding AWS. The brewery was 300 years old. Interestingly, many of the world's oldest companies are breweries, though I'm not sure why.
They were very proud of their history and had a museum that contained a 100-year-old electric power generator. When they wanted to improve their brewery's efficiency with electricity, there was no power grid, so they had to build their own power station. At that time, that's what everybody did—if a hotel wanted electricity, they had their own electric generator.
I looked at this and thought, "This is what computation is like today. Everyone has their own data center, and that's not going to last. It makes no sense." You're going to buy compute off the grid—that's AWS. We were doing it internally at Amazon for ourselves, and the APIs were created, which is a very interesting story in its own right.
These horizontal layers like electricity, compute, and now artificial intelligence—they go everywhere. I guarantee you there is not a single application that you can think of that is not going to be made better by AI.

There’s a lot more on Blue Origin and space exploration. I recommend that you check out the whole interview.

🔍🤖 Perplexity is Monetizing Through Shopping 🛒 💳

Perplexity is one of my favorite AI-enabled products.

They’ve been shipping like there’s no tomorrow, improving the product, and adding data modules that even Google doesn’t have despite having tens of thousands of people working on the product (scroll down this page to see the financial data that they’ve added when you search for a company ticker — they even have transcripts!).

When it comes to monetization, they’re still iterating and trying things. Subscriptions were the first step, but while that works fine, it probably doesn’t scale — all the biggest services like Google and Meta are free-with-ads.

So it wasn’t surprising that Perplexity has been looking at ways to advertise without ruining the user experience or undermining trust in the query results (ie. is the highest bidder going to shape what the AI tells you when you ask a question? 🤔)

To sidestep that problem, they’ve created shopping modules that allow them to identify products in queries (and even in photos uploaded to Perplexity), and then offer ways to purchase them.

If you had described it to me on paper, I would’ve thought “meh”, but they’ve executed it well enough that I think it will probably work. I doubt it’s the final word on monetization for them, but it’s an interesting step in the direction of going beyond paid subs.

To bootstrap this two-sided marketplace on the merchant side, they’re launching a merchant program:

If merchants enroll in this program, they’ll have a better chance of being a recommended product because the company will have more complete information in its index. People can also use a one-click checkout to buy items from the company’s merchant partners, the company said.
Merchants will have free API access to power search on their own websites, as well, which could help Perplexity gain market share in this space. The company said it’s not taking an affiliate cut from user purchases at the moment. Plus, it clarified that the shopping feature is separate from its ads product, which was introduced last week.

More details here.

🗣️🗣️🤖🏗️💰💸 Interview: Modest Proposal & Chetan Puttagunta

I enjoyed this interview by Patrick O’Shaughnessy (☘️):

🎧 Capital, Compute & AI Scaling

They discuss the impact of AI on the market and recent reports of a scaling slowdown and what the implications may be (it’s not necessarily bad for a lot of players!).

AI Fundraising: The Race between OpenAI, Anthropic, and xAI 🤖🤖🤖💰💰💰💰💰💰💰💰💰

The most impressive thing about it is not technical, but social — Elon Musk’s fundraising abilities:

In just 16 months since its July 2023 launch, xAI has raised ~$11 billion — a milestone that took OpenAI around eight years and the Amazon-backed Anthropic nearly four years to achieve. Indeed, xAI’s latest funding round catapulted its valuation to $50 billion

It’d be interesting to compare the fundraising graph to the usage graph…

OpenAI is probably way ahead of everyone else, and xAI is probably almost a flat line on the graph. But clearly, Musk bets that if he can leapfrog ahead 🐸, the usage will follow.

🧪🔬 Liberty Labs 🧬 🔭

🐜🔍 ASML Showcases its Computational Lithography Tech

Just a beautiful video. I’d rather watch this than most films.

It’s amazing that humans have built this stuff and that it works as well as it does! 🧙‍♂️

☀️🌡️ Power Density: The Core of the Sun vs a Compost Heap 🍂🍠 🌱

Did you know that the CORE OF THE SUN — estimated to be around 15,000,000 degrees Kelvin — has about the same power density as a compost heap?

I don’t remember when I first heard this, but it seemed very counter-intuitive, so I had to check it out.

It turns out to be correct:

Yes, the power output of the solar core is about 276.5 watts per cubic metre. However, if we average that power over the whole volume of the Sun it drops to 0.27 watts per cubic metre. (Thanks, ProfRob).

Energy is measured in joules, power is measured in watts. One watt is one joule per second. So (in general) the power tells you how much energy is produced or consumed per second.

A cubic metre of solar core contains a lot of energy, as indicated by its temperature and density, but the rate that it "generates" new energy is rather small. Of course, energy is conserved, so the Sun isn't actually producing new energy, it's merely converting mass (which is a form of energy) into kinetic energy. Some of that kinetic energy is in the form of photons, and some of it is the kinetic energy of the other fusion reaction products.

The primary fusion reactions operating in the Sun are called the proton-proton chain (or p-p chain). Unlike the processes in a hydrogen bomb (which uses deuterium & tritium, not plain hydrogen), the start of the p-p chain is quite slow. When two protons fuse, the resulting diproton is very unstable, and it usually splits apart again. However, in the brief time before the diproton splits there's a tiny probability, on the order of 10^−26, that one of the protons in the diproton converts to a neutron, creating a deuteron (a deuterium nucleus). The probability is low because the conversion relies on the weak nuclear force, which is much slower than the strong nuclear force involved in binding the nucleons together.

A typical solar core proton has a half-life of around 10 billion years. The Sun will last a long time because its main reaction process is so slow.

So the Sun's enormous total energy output comes from having a massive volume of material generating energy at this rate over billions of years, not from having an extremely high power density at any single point.

To be extra clear: The similar power density between the Sun's core and a compost heap only tells us they generate new energy at a similar rate per volume. It has nothing to do with their actual temperature.

🤖🎥 OpenAI Releases Sora (Generative AI for Video) 🎬

It took a while, but it’s finally out (kind of)!

We developed a new version of Sora—Sora Turbo— that is significantly faster than the model we previewed in February. We’re releasing it today as a standalone product at Sora.com to ChatGPT Plus and Pro users. [...]
Sora is included as part of your Plus account at no additional cost. You can generate up to 50 videos at 480p resolution or fewer videos at 720p each month.

I have a Plus account, but I can’t access the model because they are constantly over-capacity 😕

Users can generate videos up to 1080p resolution, up to 20 sec long, and in widescreen, vertical or square aspect ratios. You can bring your own assets to extend, remix, and blend, or generate entirely new content from text.
We’ve developed new interfaces to make it easier to prompt Sora with text, images and videos. Our storyboard tool lets users precisely specify inputs for each frame.

There’s a demo of the interface on this page.

It seems pretty intuitive and well-designed, but the proof will be in using it.

Right now, unless you have more luck than I do trying it for yourself, the best place to get an idea of what Sora can do is to go to Sora’s homepage and look at the gallery of generated videos.

The physics often still get surreal and continuity/coherence is often broken, but the results look very good and get us one more step in the direction of full photorealism.

Prof Ethan Mollick did a comparison of Sora with Kling (the best Chinese model) and Runway, and I happen to think that Runway is the winner, at least for that specific prompt. If Sora had come out last February it probably would’ve been ahead, but in the meantime the competition got a lot better.

MKBHD had early access and spent a week with it. Here’s his review:

Tencent also has a new video model, but I haven’t had a chance to investigate it yet.

🇺🇸 Adjusted for an Older Population, Americans are 33% Less Likely to Die from Cancer than in 1990 👨🏻‍⚕️🧫🧬🪦

Saloni explains:

Americans are now one-third less likely to die from cancer at the same ages as Americans in 1990

The gray line shows the crude rate, which is the rate of deaths from cancer per 100,000 people. It has risen between 1950 and 1990 and has fallen slightly since then.

However, cancer death rates rise sharply with age, and the age of the US population has increased since 1950, so we would expect cancer death rates to rise for that reason alone.

What if we adjust for the increased age of the US population?

The red line, the age-standardized rate, shows this. It shows the cancer death rate if the age structure of the US population was held constant throughout.

This shows a slight rise until 1990 and then a significant decline; rates have fallen by one-third.

This means Americans are now one-third less likely to die from cancer at the same ages as Americans in 1990.

Good start, but more progress remains to be made!

🎨 🎭 Liberty Studio 👩‍🎨 🎥

🛩️ The Evolution of Flight Simulator and How Microsoft Accidentally Made the Most Realistic Map 🕹️

Very cool video, very well-researched.

Even if you don’t care about Flight Simulator, I think there are a lot of cool lessons about iterative development and problem-solving in there, as well as a neat history of the computing industry.