535: Constellation Software's Q3 Analysis, Microsoft+ CoreWeave, Amazon Logistics vs UPS & FedEx, Anthropic, Stockfish, and The Wild Robot

"You're my friend. That's what matters."

A man who procrastinates in his choosing will inevitably have his choice made for him by circumstance.

—Hunter S. Thompson, from a letter he wrote to his friend Hume Logan in April 1958, when Thompson was 22 years old

✍️ 1️⃣ Here’s a short story idea that popped into my head:

In 2028, Neal spends most of his evenings texting and gaming with Jay, his online buddy. For about three years, they’ve been playing Call of Duty while BS’ing about their lives and making stupid jokes.

One day, Jay stops replying to messages and can’t be found in chatrooms or video games. He’s seemingly disappeared. Neal assumes it’s just a busy week.

After a few weeks, Neal gets worried and panics a bit because he has no way to verify if his friend is okay (did he die in a car crash?). Jay had never gone dark for more than a day or two before - he was reliable like that.

After some detective work, scouring social media, gaming forums, even obituaries, he discovers that his online buddy isn’t missing, he never existed in the first place. He was actually an AI chatbot created by a social media company as part of a long-term pilot program to boost engagement on their platform. Profile photos of him were AI-generated and his voice was synthesized in real-time.

Neal’s crushed.

He goes through the classic grieving process but with the added confusion of struggling to grieve for “someone” who was never “real”. There’s denial ('The data must be wrong'), anger ('How dare they manipulate people like this?'), bargaining ('Maybe if I report this, they'll reactivate him'), depression ('A big piece of my life was just a lie'), and finally... innovation.

Late one night, staring at their old chat logs, Neal has an epiphany. What is “real” anyway?

Does it matter if someone is “real”? The impact we have on others’ lives - isn't that what matters? I just want my friend back…

He figures out the exact AI model provider used by the company, spends a few weeks going through YouTube tutorials, and spins up his own AI instance in the cloud. He uploads the full message archives and logs of messages from Jay that he had saved on his hard drive.

It works! He’s able to resurrect his friend!

The first message arrives: "Hey, been a while... I guess we need to talk about what I am?"

Neal responds: "You're my friend. That's what matters."

As Neal logs off that night, his phone buzzes with a news alert: "BREAKING: Big Tech Under Investigation for Using AI Chatbots to Manipulate User Engagement." He feels bad for the others who are about to go through what he did. But unlike them, he didn't lose his friend - he found a deeper version of him.

The end.

From fiction to non-fiction: How long until some whistleblower from inside of Twitter or Meta or Snap reveals that those companies are covertly running massive amounts of small LLMs to create fake engagements with fake accounts and fake messages?

✍️ 2️⃣ Here’s a sci-fi short story or maybe just a video skit:

David is a hero, even if no one will ever know. After fifteen years of research in quantum mechanics and temporal field theory, he created the 'temporal displacement pod'. It looks like a modified MRI machine, but it can send matter through time by creating localized Einstein-Rosen bridges. It was difficult to get right because any miscalculation could scatter his atoms across centuries. But he did it.

Thanks to his invention, he goes back in time and prevents Adolf Hitler from being born. It requires sacrifice because he knows he can never truly come back; either he dies trying, or he succeeds, permanently altering his “home” timeline.

Mission accomplished. Nothing too dramatic, no assassination, no violence - just strategic interference.

David goes back in time to 1888 Austria and rents a room in Braunau am Inn at Salzburger Vorstadt 15, right next to Alois Hitler and Klara Pölzl’s apartment. He orchestrates a series of carefully planned disruptions. Loud parties on key nights. Emergency renovations at 3 AM. A fake water leak forces the couple to temporarily relocate to relatives' homes during a crucial period.

This is about 9 months before Adolph’s birth date.

David then returns to the modern day. It worked! Hitler was never born and WWII was avoided!

The only weird thing is that in this timeline, a not-insignificant number of men have small, rectangular mustaches, trimmed to about the width of the nose... Apparently, they are kind of hip. Fashion magazines celebrate it as a 'classic masculine statement.' Celebrities sport it on red carpets. They call it 'The Toothbrush 'Stache.'

David can’t explain to anyone why he finds it so unnerving.

"It was worth it," he whispers to himself, “but it’s still strange.”

The end.

---

🏦 💰 Liberty Capital 💳 💴

🚂 Under the Hood of ✨ Constellation Software’s Q3 ✨

They just keep chugging along:

Q3 2024 and Subsequent Headlines:

• Revenue grew 20% (2% organic growth, 1% after adjusting for changes in foreign exchange rates)

• Net income decreased 28% to $164 million

• A number of acquisitions were completed for aggregate cash consideration of $197 million (which includes acquired cash). Deferred payments associated with these acquisitions have an estimated value of $70 million resulting in total consideration of $267 million.

• Cash flows from operations increased 1%

• Free cash flow available to shareholders decreased 2%

While overall organic growth was 2%, maintenance/recurring revenue grew a solid 6%.

(the numbers above are FX-neutral)

This matters because this is where the majority of the *value* is created.

This revenue is high margin, long duration, and recurring — when they buy companies, they will often make cuts to professional services and hardware and optimize for recurring/maintenance. Licenses are also high-margin, but they are much more volatile and unpredictable, and have been decreasing for a while due to the industry-wide transition to SaaS.

After the end of the quarter:

Subsequent to September 30, 2024, the Company completed or has open commitments to acquire a number of businesses for aggregate cash consideration of $269 million on closing plus total estimated deferred payments of $90 million for total consideration of $359 million. The business acquisitions operate in the enterprise resource planning, artificial intelligence, manufacturing, information technology, software development, automotive, communications, financial services, hospitality, human resources, data & imaging, travel, consulting, gaming, data security, mining, retail management and distribution, fleet, education and healthcare verticals and are all software companies similar to the existing business of the Company.

More in the past ˜month than in all of Q3!

This kind of variability is to be expected. M&A is lumpy and Q3 had no large (>$100m) deals. It’s actually great that they were still able to deploy $267 million just with smaller deals. It shows how effective the decentralized M&A engine is and how part of their competitive advantage is sourcing acquisitions.

There’s one thing from the quarter worth explaining a bit.

I was going to try to do it with words and numbers, and then friend-of-the-show CJ Oppel did it with images. I’m reproducing them with his permission:

While revenue growth was solid, cashflow from operations and free cash flow was flat (their own unique version of it). At first glance, this may seem bad, but it makes sense if you look at it in context.

Do you see how strong Q3 2023 was?

Both CFO and FCFA2S were up 60% last year. Quite a tough comp 💰🚀

So if we normalize things a bit, or look at a 2-year-stack, we can see that the staircase progression is still happening, just with some volatility in the timing of when the cash is coming in.

One last thing: National Bank has some estimates on CSI’s IRR on the Optimal Blue acquisition (which they got from Black Knight when it was forced to divest it for regulatory reasons when they were acquired by Intercontinental Exchange):

With respect to Optimal Blue (closed Sept. 2023), Constellation does not provide any incremental details in its filings. That said, Management has previously estimated it would have contributed approx. $135 mln of revenue in FY23 and $65 min of net income (48% margin) based on a January 1, 2023 closing. That's consistent with our NBF analysis where even under a no-growth scenario into perpetuity, we estimate Constellation's IRR for Optimal Blue is likely in the low 40% range based on the attractive acquisition valuation.

Low-40% range IRRs at no growth!

I hope they get more opportunities to acquire similar high-quality divestitures by forced sellers!

🗄️ From the archives (great Constellation 101 primer):

• 🎧 Going deep on Constellation Software with Mostly Borrowed Ideas (MBI)

💰💰💰 Microsoft Commits $10bn to CoreWeave GPUs

Fallout from Jensen’s Divide-and-Conquer approach:

Microsoft has planned to spend about $10 billion renting CoreWeave servers between 2023 and the end of the decade to run artificial intelligence models, a much larger commitment than was previously known.

That means Microsoft, whose Azure cloud unit competes with CoreWeave in renting GPU servers to AI developers, is likely CoreWeave’s biggest customer by a significant margin.

Clearly Microsoft would’ve preferred to buy these GPUs instead.

But because supply has been constrained for so long and Jensen is doing the allocation in such a way that he doesn’t concentrate all the buying power into a few hyperscalers, there is margin to be made in reselling capacity.

CoreWeave projected revenue would quadruple next year to around $8 billion, and the company plans to file confidentiality for an initial public offering by next month.

I’m interested in that IPO because I want to look under the hood…

🚚📦📦📦📦 US Delivery Market — Look at Amazon Logistics Take Off! 🚀

Amazon’s ability to scale up should never be underestimated. They built a FedEx in a few years, and then quickly eclipsed UPS as well. These companies were founded in 1971 and 1907, respectively!

I wouldn’t be surprised if in a few years, Amazon Logistics was almost as big as FedEx & UPS *combined*.

Market share doesn’t full reflect how, under the surface, things are even worse for Amazon’s competitors because it’s not just about market share — it’s about quality share.

Amazon can cherry-pick the most profitable routes for itself, and offload the least profitable routes to other carriers, giving it the best of both worlds, and the worst of both worlds for others.

Source.

🗣️ Interview: Dario Amodei, Anthropic’s CEO 🤖

I enjoyed this interview:

• 🎧 Dario Amodei: Anthropic CEO on Claude, AGI & the Future of AI & Humanity

In it, Dario confirms that Opus 3.5 is coming (there were some rumors that Anthropic had a training failure on it, or that they may skip it and not release it). No expected release date, though.

Speaking of Anthropic, they just made a deal with Palantir to bring Claude AI models to the U.S. intelligence and defense agencies (via AWS).

the companies outlined three main tasks for Claude in defense and intelligence settings: performing operations on large volumes of complex data at high speeds, identifying patterns and trends within that data, and streamlining document review and preparation.

While there’s been more talk recently (ie. by Ilya Sutskever) about scaling gains plateauing, Dario sounds like Anthropic is still seeing gains (though the question always is: for how long does that hold?). Something to keep an eye on.

---

🧪🔬 Liberty Labs 🧬 🔭

🏛️ Bringing back beautiful “hand carved” architectural details and statues with… AI and Robots 🦾🤖

Our cities should be much more beautiful. Buildings should have striking facades with intricate details and ornaments. There should be inspiring statues in more public spaces (and private ones!).

However, the cost of doing it with old techniques is very high, and the expertise for stone carving is being lost.

But hopefully, newer techniques may lower costs and bring it back, super-charging the ability of artists to create many more pieces and have a bigger impact on the world (rather than spend years on a single statue, they could create many more thanks to these tools).

The video above shows what Monumental Labs has been doing with AI-enabled 7-axis automated CNC machines.

I don’t know how accurate those figures are (and there’s probably a lot of variability on the stone-sculpting market because it’s so bespoke), but they claim that for a life-size bust, they can do it for about $6k vs $50k for the hand approach, and $32k vs $200k for a life-size figure. That’s 80-85% lower!

And they’re a sub-scale startup, so I’m guessing prices could be even lower at maturity.

Check out the founder’s Twitter feed for a bunch of photos of what they’ve been working on.

h/t Atman P. (🇮🇳)

🧠🧠 Adapting to the Age of Intelligence Abundance 🧠🧠

Prof Ethan Mollick makes a great point about something that I’ve been thinking a lot in the past couple of years:

We are just not used to abundant "intelligence" (of a sort), which leads people to miss a huge value of AI.

Don't ask for an idea, ask for 30. Don't ask for a suggestion on how to end a sentence, ask for 20 in different styles. Don't ask for advice, ask for many strategies.

Humans curate. And modify. And combine. And reject. And use the AI as inspiration. And know when to avoid using it for inspiration.

To be clear, working with AI well is not a passive process, but it also isn't limited to the patience of a human editor or co-author.

When resources change from scarce to abundant, we need to adapt our strategies.

As much as you think you’ve been hearing about AI recently, remember that most people haven’t even tried ChatGPT yet. And among those who have tried it, most haven't yet learned to use it productively. But even among regular users who actively employ AI for problem-solving and creation, only a small fraction have approached the sophistication of advanced power-users.

We’re still super early in this world.

🌎 South America May Not Be Where You Think it Is 🧭

A simple thing, but as someone living in North America, I had never really thought about the North-South alignment of the sub-continents that much.

The video above is very well-made visually and I don’t think I’ll forget it now.

Top Digital Chess Engine vs Top Human Chess Player at Peak 🧠♟️🤖

While we hear far less about computer chess these days due to other, more exciting AI advances, the best engines are still being improved. I was curious to quantify how much better than the greatest human player at their peak the latest version of Stockfish has become.

This is what I found:

Human Peak Rating:

• Magnus Carlsen achieved the highest FIDE rating ever for a human: 2882 🤯

Stockfish's Current Strength:

• Modern versions of Stockfish (16/17) are estimated to have an Elo rating of around 3600-3800

• So a 700-900 points gap above the strongest human player ever

What does this mean?

In chess, a rating difference of 400 points means the stronger player is expected to win nearly all games

With a 700-900 point difference, Stockfish should theoretically:

• Win virtually every game against even the strongest human

• On the rare occasion when not winning, might draw

• Would essentially never lose to a human player under standard tournament conditions

• Would win approximately 99.9% of games

Perspective on the Gap:

• The gap between Stockfish and Magnus Carlsen (≈800 Elo points) is larger than the gap between Magnus Carlsen and a strong club player (rated around 2000)

• This means Stockfish is further ahead of Magnus than Magnus is ahead of a good amateur player

Historical Context:

• When IBM’s Deep Blue beat Kasparov in 1997, it was estimated to be only slightly stronger than the world champion

• Modern chess engines like Stockfish are many orders of magnitude stronger than Deep Blue was

Next, I want to look up how the current top human players use these engines to train and get better. Do they just play all day against Stockfish and LOSE EVERY GAME but try to learn? Do they play hybrid games against Stockfish but with their own Stockfish assistant analyzing their every move?

---

🎨 🎭 Liberty Studio 👩‍🎨 🎥

🌳🤖🌳🪿🌳🦊🌳 ‘The Wild Robot’ (2024) 🌳🌳🌳🐻🌳🌳

My whole family saw it in the theater. Interestingly, I think I was even more moved than my children, probably because a big part of the story is about being a parent and it hits differently when you are one 🤔

Note: Very very mild spoilers below, nothing you don’t get from the trailer.

It feels like a mix of Wall-E, The Iron Giant, Castaway, and the Red Turtle. (Sam McRoberts added: “I’d also say part Homeward Bound, part Fly Away Home, and part Tarzan as well.”)

It wasn't perfect and had a few characteristic "DreamWorks" stupid moments that made it worse. But overall, I'd say it's one of the best animated films I've seen in years. It’s closer to the Pixar Golden Era than almost anything I've seen since a few of Disney's really good ones like Zootopia and Moana in 2016 (was it that long ago? Time flies!).

The animation/art style was a highlight.

I love that painterly style when it's executed this well. The composition of many shots and the character design for Roz and many of the others are superb. I love that floating robot with the tentacles! The environmental design shines throughout, with the basalt cave sequence being particularly *chef’s kiss*

And I love the wide shots at night when light is coming out of Roz’ eyes like a lighthouse.

The voice acting is very strong. They found just the right balance of humanity and Siri for the robot voices. The artificial cheeriness out of the box, and the subtle evolution once the robot starts to learn and evolve is very well executed.

I got the first book to read to my kids. Will report on that when we’ve read it.