453: Anthropic, Microsoft Athena AI Chip, Nvidia Strikes Back, Antitrust Catch-22, China vs Wind Power, Nobel Immigrants, EV, and JS Bach

"we’re more limited by vision and red tape than by capability"

Praise is not the cure for insecurity. Insecure people don't believe praise.

—Stoic Emperor

📰 🇮🇱 I have nothing insightful to say about the situation in Israel. All I feel is overwhelming sadness for the victims and their loved ones.

To many, it’s “useful” to abstract war into “group A vs group B” or “ideology X vs ideology Y”, but in reality, the suffering always falls on individuals and families, and it echoes through time long after the bullets have stopped flying.

I was offline most of the weekend, celebrating the Canadian version of Thanksgiving with my family, but what I saw when I came back was horrific.

The only potentially useful thing I can mention is to be careful of psyops and disinformation. Whenever such tragedies occur, there’s a spike in activity from those trying to use these terrible events to gain some selfish advantage.

When confusion is at an apex, it is much faster and easier to make things up to shape the narrative to support whatever you already believed going in than it is to wait for people on the ground to report facts.

I’m sure that Putin’s online armies have been trying to spin things in a way that would benefit him, and many other groups will likely attempt to do the same. For the first time, these groups may be using LLMs at scale to make the astroturfing seem more convincing.

Don’t fall for it, remain skeptical! And remember that when you hear talk of geopolitics, in the end, it all comes down to individuals and families.

Update: To make it extra clear, I’m trying to warn against those who would use high-level geopolitical excuses to dehumanize the situation that this is a human horror that is absolutely indefensible. I’m trying to point out the tool that those who try to minimize things use to try to justify what CAN’T be justified. I fear I may not have been as clear as I intended above because words are inadequate and, to be honest, it’s hard to even think about such things, or imagine anyone making excuses for it. Such dark days…

👨‍🦯🫣🤔 I wonder what my biggest irrational blind spot is?

These can be so obvious in others, but very hard to figure out when it comes to ourselves (by definition — if they were easy to see, they wouldn’t be blind spots).

It doesn’t mean that you shouldn’t try, though!

Ideally, our blind spots should change over time because we did a good job of finding and fixing the old ones…

Two strategies to have a fightin’ chance:

1. Don’t get attached to beliefs. Everything is probabilistic, everything is a best guess until proven otherwise. New information should make you update, even if just a little. It’s easier to have blind spots when you deal in absolutes and are inflexible.

2. Keep sending the signal to the people around you that you welcome constructive criticism and feedback, that it’s okay for them to push back when you are wrong, that you’ll thank them for it and won’t take it personally.

🔌🚘🔋 Back in Edition #435, I wrote about how our next vehicle was likely to be a Tesla Model Y.

Well…

After almost 9 months on the waitlist, someone canceled an EV6 purchase for a vehicle that matches our criteria, so we’ll be getting it around the end of the month.

I’ll write a full review when I’ve had a chance to kick the tires for a while.

There are some things I like better about the Model Y and some things I like better about the EV6, but with the Y being $20k CAD more expensive here, I feel like the EV6 meets my family’s needs while being a better value.

This week, an electrician is coming over to install our home charging station.

I went with a FLO G5 (30 amps). It’s made by a Canadian company, handles extreme cold very well, and has top-notch build quality, according to this incredibly in-depth review:

👹 Mr Beast replied to one of my tweets, and I learned a couple things:

1. A mere reply from him can drive almost a million views to a tweet.

2. His replies are almost *entirely* people begging him for money 😬

💚 🥃🐇 The price of a couple of coffees or one alcoholic drink per month isn't a bad deal for supporting this newsletter and getting exclusive members-only editions!

The entertainment has to be worth something on its own, but for those who care most about the bottom line, there’s optionality:

If you make just one good decision that makes your life better 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).

As Bezos would say of Prime, you’d be downright irresponsible not to be a member, it takes 12 seconds (3 seconds on mobile with Apple/Google Pay):

---

🏦 💰 Liberty Capital 💳 💴

Yes, this is how I imagine Anthropic 💚 🥃

🤖 Antropic’s Path Forward 💰⚠️🚧🚗 🤔

Anthropic is reportedly looking to raise an additional $2bn from Google (and others) just days after a deal for $1.25bn from Amazon (which can increase to $4bn over time).

It’s clear that they are very very ambitious. I think Claude 2 is very good, but what will be their vector for distribution at scale?

Let’s think out loud:

• Microsoft has picked a horse and it is OpenAI (and I wouldn’t be surprised to see more internally-developed models also start to emerge and replace OpenAI’s models for smaller/simpler tasks).

• Google will favor its internal stuff (even if they own a good chunk of Anthropic — they probably see it as a hedge and a way to boost GCP).

• So it seems like Amazon is the most obvious choice.

• I think the real dark horse is Apple. They should probably try to acquire Anthropic before that becomes impossible... But I'm not sure if Anthropic would be interested since independence is very important to their mission 🤔 (but hey, everyone has a price, and Apple can outbid anyone if they decide that it is truly in their strategic advantage to get a top AI lab)

In the end, I would guess the more likely outcome is probably a closer partnership with Amazon, or maybe a partnership with Apple if they can somehow make it work with the tangled web of cross-investments and deals for compute usage that they already have with Amazon and Google.

The latest financial numbers I’ve seen on Anthropic show that they expect to be at an annualized run rate of $200m in revenues by the end of the year.

This is a lot for a young startup, but not much compared to the amount of capital they are raising — they are definitely aiming to keep up with the big model front runners.

Will they succeed? ¯\_(ツ)_/¯

💨✂️🏭 Siemens Evaluating the Closure of Wind Factories + China's Mastery in Creating Worldwide Dependencies 🇨🇳

China has been working diligently to make the rest of the world dependent on it when it comes to solar, wind, and lithium supply chains.

It’s the kind of thing that works fine until it doesn’t (ask Germany how becoming dependent on Russian gas turned out for them).

On the wind power side, even the largest Western giants like GE, Vestas, and Siemens can’t compete and are being forced to pull back:

Globally, Siemens Gamesa, the world's largest maker of offshore wind turbines, operates 79 sites, including sales and service offices, R&D centres as well as 15 factories to produce blades and nacelles.

Some of those could be closed or put under temporary hibernation as Siemens Gamesa aims to rid itself of production of parts its suppliers can make more cheaply [...]

The measures, which are likely to cause fresh layoffs, are aimed at providing long-term relief for Siemens Gamesa by outsourcing production of some key components, such as blades, in order to raise margins (Source)

The more parts that end up being made at lower-cost suppliers — read: in China — the less control anyone whose grid is dependent on wind power has over their crucial infrastructure.

If Chinese suppliers became inaccessible for any reason, Western manufacturers could figure out how to make these parts, but it would take time to regain the expertise and build manufacturing capacity back up, and costs would likely be a lot higher.

That’s a small problem if a few % of your grids use wind (the same applies to solar), but it becomes a *big* problem when a large percentage of the grid uses those sources and you have to constantly replenish your installed base because wind turbines and solar panels have a useful life of 20-30 years. So *everything* that was installed up to now will need to be replaced over the coming decades, on top of whatever new capacity we want to add.

The rest of the world shouldn’t want to be in a position where if China cuts off supply, our grids start decaying and it’s a race to build back the manufacturing expertise and capacity to keep up with replacement rates, all of this at much higher costs than were initially expected.

🤖 Microsoft is almost ready to unveil its Athena AI Accelerator Chip (Competing with Nvidia) 🐜

I wrote about this in Edition #411:

The software giant has been developing the chip, internally code-named Athena, since as early as 2019 [...]

It’s reported that at least 300 people are working on this piece of custom silicon.

At first it was thought that Marvell was Microsoft’s partner, but “a source with direct knowledge” says that MS owns the IP.

This is not easy! It takes *a lot* of time and expertise.

Although Nvidia still has the best chips, interconnects, and software ecosystem, the benefits of going in-house are pretty big for hyperscalers. They don’t have to pay Jensen’s margins and don’t have to wait for an allocation of the latest chips (as long as they can get foundry space).

Google is leading the way with their TPU, which they began using internally in 2015 and is now starting to roll out V5e, followed by Amazon’s AWS with its Inferentia and Trainium chips (not directly competitive with Nvidia, but cheaper).

The biggest challenge for all these (and I haven’t even mentioned AMD yet!) is that Nvidia’s ecosystem is the standard and its software APIs and frameworks (CUDA, tons of acceleration libraries) are the foundation on which almost everything in the field is built. There are many clever ways to try to abstract it away (Triton, PyTorch), but in practice, Nvidia still dominates.

Even after Athena comes out, Microsoft will likely keep buying tons of Nvidia GPUs for the foreseeable future…

⚔️ The (Nvidia) Empire Strikes Back

Speaking of Nvidia, while hyperscalers and AI labs are trying to reduce their dependency on the neon green company, it is, in turn, attempting to reduce its reliance on the hyperscalers as customers by helping young AI-focused cloud startups (as I’ve written about a few times) AND trying to build its own cloud with a direct relationship with customers.

These efforts started out as sub-leasing equipment owned by hyperscalers and reselling it to enterprise customers branded as Nvidia DGX Cloud, but these efforts are now becoming more ambitious:

[Nvidia] has talked to at least one data center owner about leasing its own space to power the cloud service, according to a person with direct knowledge of the discussions. [...]

The evolution of DGX Cloud, and Nvidia’s recent interest in acquiring data center space, represent an attempt to leverage the company’s dominance in AI server hardware to generate new sources of revenue and gain direct relationships with users of its chips.

Because Nvidia’s DGX Cloud is basically just white-labeling hyperscalers, it hasn’t been very differentiated.

But if it began operating data-centers directly, it could optimize things more and be competitive for the most demanding workloads (there’s probably a bunch of cutting-edge Mellanox gear that hyperscalers aren’t using that could improve performance).

Of course, all this is a trade-off between the capital-light and capital-heavy approaches. We’ll see how much appetite Jensen has for this, and whether he gets caught pushing the gas pedal to the metal as another semiconductor cycle turns downward…

⚖️ Big Tech Antitrust Catch-22 🏛️

Friend-of-the-show and Extra-Deluxe supporter (💚💚💚💚💚 🥃) Byrne Hobart posted this great line:

Ronald Coase [1991 Nobel Prize in economics] said he had gotten tired of antitrust because when the prices went up the judges said it was monopoly, when the prices went down they said it was predatory pricing, and when they stayed the same they said it was tacit collusion.

I’m worried and skeptical about the amount of power — financial and otherwise — that Big Tech is amassing.

BUT

The way regulators are attacking Big Tech for ideological/political reasons and/or using them as piggy banks feels unsavory.

I wish there was a substantive debate about how to best deal with this new phenomenon and which trade-offs we can live with rather than pretending that it all fits the old supply-control paradigms from the Robber Barons era (as an aside, how effective is the “Robber Baron” branding? It really stuck to them).

The longer we wait until we do the hard work of creating new frameworks, the less credibility will remain for regulators, and the more entrenched the incumbents will be as a result of misguided regulation that hurts Big Tech but hurts small tech and new entrants way more.

---

🧪🔬 Liberty Labs 🧬 🔭

Visualizing the Faces of History: Using Generative AI on Statues 🏛️🔍🤖

There are so many statues and portraits that could be brought to life using this technology. I kind of want to do a series of my faves from various eras and places 🤔

It’s like the extreme version of taking grainy old black-and-white films and colorizing and stabilizing them to have a better idea of what things really looked like at the time.

Via One_Sherbert_6904

🔌🚙🇬🇧 How energy-efficient are electric vehicles? How much extra electricity would we need to charge them?

Hannah Ritchie did napkin math to calculate how efficient EVs are compared to internal combustion engine (ICE) vehicles and how much power we’d need to electrify the UK’s transportation fleet.

Most people don’t know that the vast majority of the energy contained in a gallon of fuel is wasted as heat and doesn’t do anything to move the vehicle:

Note how regenerative braking contributes! It’s such an elegant part of the design — rather than dissipate kinetic energy as heat on brake pads, use the electric motors as generators to brake the car and dump that power back into the battery.

Because of this, we don’t need a 1:1 ratio of electricity to hydrocarbon energy to electrify transportation (a common mistake):

Electricity transport significantly reduces primary energy demand

If we were to electrify transport, we’d need much less energy overall.

When we look at the amount of energy that countries currently use, it can seem like an overwhelming challenge to replace all of this with low-carbon sources.

But, we don’t need to produce the equivalent amount of low-carbon energy. Our current energy stack includes the energy that we actually need, but it also includes all of the wasted energy in converting fossil fuels to something useful. When we switch to electric cars or low-carbon electricity, a lot of that disappears.

We can demonstrate that with transport.

Great Britain currently uses the equivalent of 445 TWh from petrol and diesel road vehicles. As we’ve just calculated, if this was electric we’d need just 118 TWh. Almost 4 times less. Our energy demand has shrunk dramatically.

Hannah estimates that for the UK, electrifying all cars would increase electricity demand by approx 25%. If all other road transport in included, it would be 40%.

It’s a lot, but it’s doable if we plan for it smartly. We’ve done bigger things very quickly in the past — we’re more limited by vision and red tape than by capability.

🏅 Nobel Immigrants 🏅

Adam Ozimek writes:

Thanks to immigrants, the US won 75% of Nobel prizes in 2023. Without them, we only would have won 25%. 

High skilled immigration is a policy lever for more innovation unlike anything else. Oh and its free.

Since 1901, immigrants have represented 34% of US Nobel Prize winners.

Also note that 75% of all prize winners in 2023 are immigrants, even outside the US. The most able are disproportionately most mobile.

Any country that is lucky enough to be where the smartest and most driven people in the world want to go live should figure out a way to let them…

---

🎨 🎭 Liberty Studio 👩‍🎨 🎥

AI-generated portrait of the great man 💚 🥃

‘Is Bach the greatest achiever of all time?’ 🤔

Interesting argument by Tyler Cowen.

Clearly, this is not a question that can be decisively answered, but I find thinking about it very fruitful.

Some achievers were far ahead of their peers and made the world progress rapidly (Albert Einstein, Claude Shannon), but they didn’t necessarily keep performing at this high level decade after decade the way that someone like Bach did (not that this takes anything away from them, but if we’re looking at consistency as one of the criteria, it can matter for this exercise).

Who would be your pick and why?

Interview: Marc Andreessen, on Writing ✍️

Friend-of-the-show David Perell (📝) had a great conversation with Andreessen, mostly about writing, but also about a lot of interesting topics like how the internet is changing our lives.

The section about the role he played in creating some of the arbitrary conventions of the web (ie. hyperlinks are blue) and some things he wanted to implement in the early days but never had a chance to is great (every webpage almost had a built-in Wiki/comment section!).

He also made me very curious to check out author James Ellroy.

I haven’t read any of his fiction, but I liked L.A. Confidential and his book about Los Angeles in the weeks after Pearl Harbor, when they expected to be next, sounds quite intriguing.

I recommend the pod:

• 🎧 Marc Andreessen: It’s Time To Write