461: OpenAI Saga, Alibaba, Moats Are Not Static, Ford EV Plans, Microsoft's GPU Capex, Unreal Engine, RAM, and 'No CGI' Lies

"Most things are either improving or deteriorating"

Nov 22, 2023

The chief trick to making good mistakes is not to hide them — especially not from yourself.

—Daniel C. Dennett

🛀💭✨⏳⏳⏳⏳⏳⏳⏳⏳⏳⏳✨ We often think about the vastness of space — measuring distances in units that our brains cannot comprehend (light years) and marveling at deep field photos taken by space telescopes that show countless galaxies, each with countless stars and planets. ✨🪐🔭

But I find the vastness of *time* equally vertigo-inducing.

The question “Are we alone in the universe?” has the quality of being simple, but it hides the fact that we have to search through both space *and* time.

A more useful question would be something like "Are we the only ones in this accessible region of the universe during our moment in time?"

We have only had the tech to be space-faring (and barely — we haven’t gone very far) for a few decades, which is nothing on a cosmic time scale. The same goes for the powerful telescopes and radio-telescopes that allow us to look beyond our own backyard (aka the Solar System).

If advanced intelligent civilizations are common and don’t all go radio-silent relatively quickly (either because they invent better ways to communicate than blasting radio-waves omnidirectionally, or for more sinister reasons), we might be able to identify the electromagnetic signature of intelligence out there.

But even if we do find something, the vastness of space will probably keep us isolated. Any ship or probe that we send would take so long to travel that by the time it got there, our own civilization may be gone or changed beyond recognition, or theirs may be gone, or both.

In the scenario where intelligent civilizations are rare in the universe, the vastness of time also makes things difficult; in order for civilizations to ever notice each other, they have to line up temporally.

The starry sky we see at night doesn't show the universe as it *currently* exists; instead, the farther we peer into space, the older the photons are by the time they reach our retinas or instruments, presenting us with an increasingly outdated view. We look at many stars that have long gone. It’s possible that we could detect signals of intelligence that belong to long-gone lifeforms.

Time is so vast that it’s easy to miss each other by millions of years. Even thousands of years may be enough for a civilization to rise and fall if things go wrong.

When thinking about our place in the cosmos, we have to consider:

Vastness of Space x Vastness of Time.

I know, it’s kind of a sci-fi shower thought, but it happens…

📖🐴🩸 Now *that* is a great book cover design!

Credit to Folio Society books (h/t Lael O. for pointing me to it!).

It’s a good excuse to remind you to check out the *excellent* mini-series ‘The Offer’ (2022, Paramount+) which is largely about the making of the film ‘The Godfather’.

I wrote about it a few times including in Edition #408.

💌📬 Pssst. You should sign up for OSV’s newsletter. There are all kinds of goodies in there, with a lot more coming!

💚 🥃 🐇 The teachers at my kids’ school are on strike, so we’ve got the kids home for an unknown duration. I’m hoping it’ll be just a few days, but who knows? ¯\_(ツ)_/¯

This is why I haven’t yet sent the email to Extra-Deluxe subscribers about setting up a group Zoom call.

(If you sign up for Extra-Deluxe after the email goes out, don’t worry, I’ll reach out individually and loop you in)

🏦 💰 Liberty Capital 💳 💴

🇨🇳 Alibaba (and the 40 thieves) 💸

Friend-of-the-show Trevor Scott:

It’s Sept 19, 2014.

Meghan Trainor is all about that bass and Taylor Swift is shaking it off.

Alibaba is the fastest growing retailer in the world.

You invest $10,000 at the IPO. 

9 years pass.

Revenues grow from $5.5b to $123.6b.

Your investment is now worth $8,426.

There are various lessons tangled up in there.

Entry points and valuation matters. Alibaba IPO’ed at 28.8x EBITDA and now trades at 6.4x EBITDA.

There’s also something to learn about investing in authoritarian regimes where you don’t really “own” what you think you own, and management doesn’t ultimately have control over the assets (ask Jack Ma).

It’s not because a lot of value is created that shareholders will be the ones to benefit.

That said, this isn’t a forward-looking statement. Alibaba may do worse or better in the future, I have no idea. On one hand, if you just look at the number, valuation is less demanding than it has ever been. But on the other, it’s hard to know how Xi and the CCP will feel about things and what they might decide to do tomorrow or in two years.

🤦‍♀️😬🫣 OpenAI Drama 🤖🌧️

Update: This morning it was announced that Altman is back as OpenAI CEO. He didn’t get a completely new board and has apparently agreed to an investigation into the mysterious claims against him. We still don’t know much about what happened, but this whole saga is full of lessons, including about theoretical power vs effective power (ie. on paper the board has power, in practice, the owner of the GPUs and whoever the crucial employees of the company support has power). I’m leaving what I wrote before this news broke below because I think a lot of it still applies:

I picked quite the weekend to go offline in the woods...

🌲🌲🌲🏡🌲🌲🌲

I didn’t follow the ‘Sam Altman is fired’ news cycle in real-time. I don’t have much to add — I think we’ll have to wait to know more facts.

The meta-lesson is how quickly people can spin up narratives based on very little, mostly in ways that reinforce what they already believed going in. When there’s a vacuum, the online mob is very quick to fill the holes with what is most convenient for them to believe. Every side comes out with a stronger belief in their priors.

Reality is usually complex and nuanced.

The same event can appear totally different to observers who don’t frame it the same way.

If you think OpenAI is a regular company, then events seem a certain way. If you think they are an idiosyncratic research lab with both commercial and non-commercial goals mixed together, with a complex priority stack where profit isn’t at the top, then things can be interpreted differently.

Sometimes both sides are wrong (to varying degrees). Sometimes both sides are right (to varying degrees). Sometimes good decisions lead to bad outcomes, sometimes bad decisions lead to good outcomes (be careful about ‘resulting’). It’s hard to judge without knowing the facts ¯\_(ツ)_/¯

Ethan Mollick has very sane commentary and context.

Ben Thompson too (💚 🥃 🎩)

From a purely business point of view, it appears to be a coup for Satya Nadella and Microsoft. Or at the very least, it seems like he’s made the best lemonade that he could out of the lemons he got 🍋🧃

Matthew Prince has an interesting tweet regarding the less obvious downsides for Microsoft to how things turned out, and how it could’ve been much worse, highlighting their original sin of investing billions in an entity they don’t have direct oversight on.

When it comes to the development of AI more generally, I don’t know what the upshot is. Does it ultimately accelerate or decelerate development? Does it make it safer or riskier? I don’t know ¯\_(ツ)_/¯

Of course, events have been moving so quickly and the equilibrium is so unstable that by the time you read this, the situation may be completely different… [I was right.]

Personally, I’d put a higher probability on OpenAI being reunited after a board change than for Altman and most employees to move over to Microsoft. [Didn’t get the board change, but it was a good guess.]

🏰 Business Moats are NOT Static 🐊

In many areas of life, people too often look at snapshots, at static points in time, when they should really be looking at vectors, direction, speed, acceleration vs deceleration…

Matt Franz wrote something interesting looking at business moats through that lens:

Everyone talks about “moats” but almost no one talks about how moats change. Moats get a little bit wider or narrower every day. A widening moat is even more valuable than a wide moat. A moat’s direction is more important than its width.
Why? A widening moat implies longevity. It’s one thing to have high returns on equity. Plenty of businesses do. Long-term investors care about how much capital a business can invest at a high rate and for how long.

Few things are perfectly stable — and even if they were, the world around them isn’t, so their relative position will still change.

Most things are either improving or deteriorating, and this applies to businesses and their competitive advantages:

At the 2022 Graham & Dodd Annual Breakfast Todd Combs recalled the first time he met Charlie Munger. Munger asked what percentage of S&P 500 businesses would be a “better business” in five years. Combs said less than 5%. Munger thought less than 2%.
By their math 10-25 companies in the S&P 500 are getting better and 475-490 are getting worse. Technology has increased the rate of “creative destruction” which has made widening moats an endangered species.

I don’t know about those specific numbers — you can make your own guesses. But I doubt that most businesses in the S&P500 are getting better…

The challenge for a business analyst is to figure out if a moat is being widened — if crocodiles and spikes are being added to it — or if it’s shrinking and becoming easier to breach because competitors have just invented a new kind of portable bridge or siege technology.

What can make it hard to figure out is that sometimes an improvement in competitive advantage shows up clearly in the numbers while other times the numbers may look ugly even as things are improving.

For a long time, critics described Amazon as “a charity for consumers funded by investors,” failing to see that the overall losses masked a portfolio of diverse businesses in varying growth phases. Profits from well-established ones were reinvested into newer internal startups, laying the foundation for robust competitive advantages built on scale.

Understanding the direction of a moat requires a deep understanding of the business. Investors that stick with stocks for decades focus on KPIs that indicate whether the moat is widening or narrowing. [...]
A widening moat implies a long-lived moat. Time is their friend. Investors in widening-moat stocks tend to stick with their pick for decades. Sleep did with Amazon, Munger with Costco, and Buffett with GEICO.

I don’t think it’s so much about identifying widening moats that nobody else knows about. That’s unlikely. However, understanding the business well enough to give you the confidence to hold it for a long time, through volatility and the inevitable hiccups along the way, can help an investor exploit one of the few remaining edges, which is having a long time horizon (time arbitrage).

‘Ford to scale back plans for $3.5 billion Michigan battery plant’ 🔌🛻🔋

After coming out very hot when first announced a few years ago, things are slowing down for Ford’s EVs:

Ford is scaling back plans for a $3.5 billion battery plant in Michigan as consumers shift to electric vehicles more slowly than expected, labor costs rise and the company moves to cut costs.
The company said Tuesday it is cutting production capacity by roughly 43% to 20 gigawatt hours per year and reducing expected employment from 2,500 jobs to 1,700 jobs. (Source)

The company’s recent deal with its workers’ union is estimated by the CFO to add $850-900 in labor costs to each vehicle.

Napkin math to adjust scale based on the reduced capacity would make it still around a $2 billion investment. That’s not nothing!

Microsoft’s GPU Investments & What OpenAI has access to 🐜🐜🐜🐜🐜

Friend-of-the-show Dylan Patel has a good recap of recent events that includes the Microsoft infrastructure on which both OpenAI and Microsoft rely:

Compute Is King
Microsoft had previously placed huge bets on OpenAI, with plans for >$50B annual datacenter spend to race to AGI and deploy their GPT-4 based copilot products. Our data shows one of OpenAI’s next training supercomputers in Arizona was going to have more than 75,000+ GPUs in a singular site by the middle of next year.
Our data also shows us that Microsoft is directly buying more than 400,000 GPUs next year for both training and copilot/API inference. Furthermore, Microsoft also has tens of thousands of GPUs coming in via cloud deals with CoreWeave, Lambda, and Oracle.

It’ll be interesting to see how this week’s drama will reshuffle the card for other AI labs. It was reported that Anthropic got “more than 100 OpenAI customers” contacting them over the weekend (probably more than that by now).

There are also reports of OpenAI customers wanting to shift from the OpenAI APIs to the Azure version of those same APIs, giving Microsoft even more leverage in the relationship…

🇩🇪 Germany off-budget SNAFU 💰💰💰💰💰💰💰

You know something’s wrong when off-budget items are bigger than the government’s official budget. The courts are not liking it:

Chancellor Olaf Scholz’s governing coalition has been scrambling to work out the implications of the Constitutional Court judgment, which calls into question hundreds of billions of euros of financing in special funds that are not part of the regular federal budget. [...]
Germany has 29 such off-budget funds worth around €870 billion, though a €100 billion pot for military spending is not expected to be affected as it’s written into the constitution. (Source)

Note that the “climate” and “energy subsidies” add up to around 400bn Euros. That’s a lot of geld! Especially considering the more than half-a-trillion Euros already invested in renewables, only to have to restart coal plants and import more LNG.

Not to sound like a broken record, but I’d be curious to see how much better things would be in Germany if they hadn’t shut down their nuclear power plants (which together produced 26% of the country’s electricity — remember, not all electricity is the same: this was reliable, predictably-priced, 24/7 baseload, clean power that their now-shrinking industrial base could rely on).

🧪🔬 Liberty Labs 🧬 🔭

You think you know what RAM is, but do you really? 🧐

Great video, insanely detailed and well-made. I’ve been a computer nerd for decades and I learned a bunch of stuff.

👾 What’s Coming to Unreal Engine in 2024 🤯

Some pretty amazing stuff in the pipeline.

The section near the end where they show facial expressions is very impressive, as that’s one of the hardest things to get right (our brains are finely tuned to notice anything wrong with human faces).

☀️🐟 Micronutrients for Longevity: Spotlight on Vitamin D & Omega 3 (with a side of exercise 🏋️‍♂️🚴‍♀️)

Good overview by Rhonda Patrick of low-hanging fruits that everybody should know about (you’re taking this stuff seriously, right? It’s very asymmetric…):

She cites a study that suggests the longevity reduction from Omega 3 deficiency is similar to the impact of smoking. Yikes 😬

I sure am glad I take my 6 capsules of Kirkland super-concentrated omega 3 every day (for a cumulative 4500mg of EPA & DHA). A lot of people take the right things, but in quantities too low to have a noticeable impact.

Magnesium is another easy win for most of us. Personally, I take magnesium L-threonate before bed. I used to take other forms of magnesium (Citrate), but I switched to this one about a year ago.

🤖🔎🩻🔍👨🏻‍⚕️ AI boosts breast and pancreatic cancer detection rates

For years now, one of the most common examples used to describe the potential benefits of “machine learning” (remember when that was the buzzword?) and AI was reading X-rays, CT scans, and MRIs.

Well, it was true:

The new study, published on Thursday in Nature Medicine, highlights how machine learning can help tackle life-threatening diseases by flagging errors or identifying hard-to-read signs overlooked by humans.
AI analysis flagged up to 13 per cent more cases than doctors had identified — a significant chunk of the 20 per cent or more cancers that are estimated to be missed using current non-AI screening.

This tech would be used in conjunction with human doctors, at least at first.

It would act as a safety net, both to catch what bad doctors may miss, creating a kind of detection floor, but also improving even the work of good doctors by making them more productive, allowing them to help more people each day (the study showed that the AI “could save up to 45 per cent of the time spent on breast cancer scan reading times”).

The study used an AI tool, known as Mia, developed by Imperial and Kheiron Medical Technologies, a UK company that specialises in AI medical diagnostics. The paper examined a group of 25,000 women screened for breast cancer in Hungary between 2021 and 2023.
The study comprised three phases, in each of which the radiologists interacted with the AI in a slightly different way. The groups showed improvements in cancer detection rates of 5 per cent, 10 per cent and 13 per cent, compared with the standard reading by at least two radiologists. (Source)

There is also good news when it comes to the very deadly pancreatic cancer (RIP Steve Jobs):

Pancreatic ductal adenocarcinoma (PDAC), the most deadly solid malignancy, is typically detected late and at an inoperable stage. Early or incidental detection is associated with prolonged survival, but screening asymptomatic individuals for PDAC using a single test remains unfeasible due to the low prevalence and potential harms of false positives. [...]
we develop a deep learning approach, pancreatic cancer detection with artificial intelligence (PANDA), that can detect and classify pancreatic lesions with high accuracy via non-contrast CT. PANDA is trained on a dataset of 3,208 patients from a single center. [...]
[PANDA] outperforms the mean radiologist performance by 34.1% in sensitivity and 6.3% in specificity for PDAC identification, and achieves a sensitivity of 92.9% and specificity of 99.9% for lesion detection in a real-world multi-scenario validation consisting of 20,530 consecutive patients.

Source. (h/t my friend MBI (🇧🇩🇺🇸))

🎨 🎭 Liberty Studio 👩‍🎨 🎥

🎞️ Claims of “No CGI” vs Reality 🕵️‍♂️

Don’t believe everything you’re told!

I understand that marketing is all about telling an interesting story, but I also don’t want to be lied to (or to need a lawyer to parse the language to make sure there wasn’t a loophole that makes something literally true even if it’s highly misleading).