Liberty Workbench is an occasional series where I put one idea on the bench and think it through with you.

Many of you have been asking for updates about my vibe-coded apps, which I’ve teased in past editions. I thought it would be quick, and I could tell you about it sooner. But it’s one of these things where you start something small, and then you think of something else and something else, and before you know it, you’re trying to build the Taj Mahal 😅 (and failing, but learning)

First, it’s not one app. It’s six!

But I’ll focus on four of them here.

Where it all started 🪴🖼️🏠

I’ve always been very interested in software engineering, but I’ve never been a coder myself. For a long time it felt like a road not taken, and I had some regret… but in recent months, it became possible to apply the high-level principles I had learned by osmosis just from being a nerd and following tech for decades and direct coding agents to implement things.

I used both Codex and Claude Code (I went back and forth between the two, they tend to catch each other’s mistakes and have their own strengths and weaknesses).

The biggest step, as it always is, was to go from zero to one.

My wife and I were talking about how it would be fun to print some of our favorite photos we’ve taken over the years and frame them to fill some bare walls in a part of our house (either family portraits or nature shots).

This got me thinking about how hard it is to pick favorites from hundreds and hundreds of photos… and I had the idea for my first app.

At first it was supposed to be very simple. Just dump all your photos into it. It shows you a pair and you pick your favorite of the two. A/B selection. The app keeps track and after a while, it can rank your photos and find your favorites.

Pairwise: The devil’s in the detail 😈

Ok. Simple enough, right?

Read more

Deconstruction creates knowledge. Recombination creates value.

–James Clear

❤️‍🩹 Ben Wilson is a great guy. I don’t know him very well, but we had a couple great chats a few years ago, and we have friends in common that all say great things about him. I consider him a friend-of-the-show.

If you can help him and his loved ones in any way, please, I encourage you to do so. There’s a campaign to raise funds for his treatment and family.

💚 🥃 To keep this steamboat alive and afloat, I need supporters. Like you! 🚢

The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Liberty Capital 💳 💴

🚀🤝 Musk Says SpaceX and Tesla Are Considering a Merger 🤲

A lot of people have been expecting this for a while, but it’s getting more concrete.

From the quarterly call last night:

Elon Musk: so both SpaceX and Tesla have formed special committees to discuss this idea at the request of shareholders. [...]

I think some people out there assume that a SpaceX and Tesla merger is a certainty. And I think that would be overstating it. Do I think it would be a good idea, yes. But at the end of the day, the shareholders have to be supportive of this. And particularly on the Tesla side, we will need a 50% vote of all shares, not just those who vote, but of all shareholders.

For reference:

But considering how many of the shareholders of both are a self-selected group of Elon believers, I don’t think he’d have any problems getting a merger passed, unless the terms are REALLY bad for one side (which isn’t out of the question, to be sure!).

One thing is that if they did merge, what would that do to Musk’s control of the combined entity? He has overwhelming control of SpaceX, and would be diluting that.

But if the combined entity keeps the supervoting shares used by SpaceX, he could potentially maintain majority voting control, depending on how the deal is structured.

So and I can just say at this point that the special committees have not reached agreement on the terms of any transaction. And obviously, there’s no guarantee that they will.

But I do think there’s a lot of merit to combining the companies as we are pretty much in the super intelligence singularity. And there’s a lot to be gained from cooperating on developing super intelligent software, chip design, the Terafab, obviously, robotics, mass to orbit with the Starmind constellation.

All this talk of cooperation doesn’t require a merger, though.

The companies are already working together on various things like Terafab and energy stuff (and Tesla owns nearly 19 million SpaceX shares), and it’s not like either has had trouble raising capital in the past.

So what would the merger unlock? I guess it would make it easier to allocate capital and talent between the companies. and maybe a way to bury Tesla’s largely stagnant EV sales into the larger whole? 🤔

🔁 Déjà vu: Elon Makes Jensen an Offer the Hyperscalers Won’t

🚕🤖 Waymo’s Plan to 5x Its Fleet: From Science Project to Transportation Giant?

Semafor reports that Waymo wants to go from about 4,000 robotaxis today to 8,000 by the end of this year and 20,000 by the end of 2027. That’s a fivefold expansion in a little over a year (IF they pull it off). 🚀

This reminded me of Edition #532 from October 2024, when I was looking at Waymo’s capital efficiency compared with Tesla. Back then, Waymo was estimated to have raised about $13bn and had fewer than 1,000 vehicles deployed.

What a difference two years make!

As I wrote in Edition #624, Waymo’s methodical approach has been aging well. Earlier this year, it was doing around half a million fully autonomous rides every week, after starting with those experimental self-driving Priuses way back in 2009-2010.

Waymo raised $16bn in equity earlier this year and recently borrowed another $5bn, its first debt financing. Interestingly, the debt doesn’t have guarantees from Alphabet. I guess dad’s already busy putting AI capex on his balance sheet 🤔

But the other way to look at it is that Waymo is now established enough that it can borrow $5bn without Alphabet’s co-signing.

Not that the self-driving software is easy — it’s very f hard!!! — but even once you’ve got that, there’s a lot of work that remains if you want to operate tens of thousands of robotaxis across a bunch of cities. Buying the cars, charging them, cleaning them, maintaining them, building depots, dealing with local regulations… dealing with crazy customers or vandalism or whatever else happens in these empty cars.

I wonder what the economics look like as they keep scaling.

Oh, and Waymo is coming to Toronto too! 🇨🇦 They're aiming for first rides in 2028.

Wen Ottawa region, Waymo?

💰 Eric Schmidt’s $15 Billion Anthropic Side Bet 🤖

Eric Schmidt owns about 1% of Anthropic. He’s reported to have invested tens of millions of dollars in its Series A back in 2021, before Claude was even a publicly available product. That round was for $124m, so his check was a significant % of it.

At the roughly $1.5 trillion valuation that has been discussed lately for Anthropic’s IPO, his ~1% would be worth $15 billion. Not long ago, some investors expected an IPO valuation closer to $2 trillion, which would put his stake at $20 billion. 😯

As if he wasn’t already rich enough from Google, despite not even being one of its founders! 😅 Bloomberg estimates his fortune at $57.5bn before counting Anthropic, while Forbes puts it at around $38bn.

Not bad for a side investment from someone who spent a decade as Google’s CEO and then stayed on as chairman until early 2018 (he did well as a non-founder).

This reminds me of the story about how Jeff Bezos was one of the early investors in Google and even if Amazon had failed and went to zero he would probably still be a billionaire just from that investment (I wrote about it in Edition #482). And it wasn’t his only successful early investment.

This also reminds me a bit of what I wrote in Edition #638 about Demis Hassabis being an early investor in Anthropic.

So both Google’s former CEO and the guy running Google DeepMind have personal investments in one of Google’s biggest AI competitors. 😅

These people may be fierce competitors professionally, but personally, they’ve known each other, worked together, and invested in each other’s companies for years.

Schmidt has also invested in Mistral AI, Reflection AI and World Labs.

If you were early to one of these AI labs, there’s a good chance you saw the others coming too, before their valuations went vertical. Access counts for a lot in angel and VC investing, where a few big winners account for most of the returns.

Once again, a reminder that when it comes to angel and VC investments, access counts for a lot, and there are power laws everywhere.

🔁 Déjà vu:

• Demis Hassabis is an Angel Investor in Anthropic

• Jeff Bezos is an early investor in Google

From the full feed 🔒: Prem Watsa’s $5.6B Indian Banking Bet

🧪🔬 Liberty Labs 🧬 🔭

📸 🔐 Apple’s New Way to Prove a Photo Was Taken by a Real Camera

I’ve written before about cryptographically signing photos and videos as one way to deal with a world where AI can generate increasingly convincing fake images. It’s not foolproof. There are many ways to get around it. For example, you could photograph some AI-generated thing on a high quality screen and get a perfectly valid signature on something that never happened.

But it would still be useful and create some friction where basically none exists right now (from 0 to 1 is the biggest jump, right?).

Anyway, Apple has taken an interesting step in that direction with Apple Reference Image, a new feature in iOS 27 on the iPhone 18 Pro and Pro Max.

The clever part is that the camera sensor itself signs the image data at the moment of capture, before the iPhone’s software can touch it. The phone saves a secure digital negative (DNG) containing the original sensor data, signatures, and cryptographically verified timestamps. This can then be developed into a JPEG using Apple’s Private Cloud Compute, which checks the signatures and timestamps before Apple’s signing service adds the final signature (this one is even designed to resist future quantum computers 🔐). You can compare this reference image with the normal photo to see what’s been edited.

For now, it’s an opt-in mode rather than something that happens with every photo, and verification is mostly limited to Apple’s ecosystem. But you gotta start somewhere, and I could see this kind of thing become ubiquitous for photos and videos in the years to come.

It’s like the mirror image of the AI detectors. Rather than play an endless arms race of trying to detect increasingly sophisticated fakes, create a way to positively verify what a real camera actually captured.

🛒🇨🇳 ByteDance, Alibaba, and Tencent order $35bn of Nvidia… Workstation Chips?!

Here we go again…

Reuters reports that ByteDance, Alibaba, and Tencent have ordered more than 3 million Nvidia GPUs, worth over $35bn. 💰

Which chips? That’s kind of the surprising part.

Not H200s or B300s, which are restricted and have been used as bargaining chips (accidental pun) a bunch. They want to buy RTX PRO 5500 workstation GPUs.

These less powerful chips comply with U.S. export controls. The Chinese companies mostly want them for inference.

What’s funny is that many Chinese buyers weren’t interested in the similar RTX 6000D last year, hoping they’d get access to Nvidia’s more powerful data-center chips. I guess those hopes have faded… and the discontinued 6000D is reported to have doubled in price in China. 😅

But orders don’t necessarily equal chips delivered.

Across China, orders apparently exceed 6 million cards, while Nvidia expects to manufacture no more than 3 million in 2027. One source says some customers may receive just 5–10% of what they ordered. Commercial shipments aren’t expected to begin until Q1 2027.

But hey, the game theory here is that if everyone knows they will be in short supply, maybe they’re all over-ordering, hoping that gets them more chips in the end 🤔

🔁 Déjà vu: Chinese internet giants’ “strategic GPU reserve”

🤖💾 AI Is Getting Better at the Ancient Code Running the Modern Economy 🏦

Back in Edition #571, I wrote about Morgan Stanley using AI to understand and modernize its ancient COBOL codebase. At the time, AI could explain old code and translate smaller sections, but wasn’t reliable enough to automatically migrate mission-critical systems.

I ended that with: “But that’ll come.”

OpenAI reports that GPT-6.1 Sol achieved 86.6% accuracy on its COBOL migration benchmark, up from just 5.4% for GPT-5.6 Sol. At higher reasoning effort, it reached 97.9%. Not bad progress for *checks watch* 82 days (!!!!).

(GPT-5.6 Sol came out July 9, 2026, GPT-6.1 Sol came out September 29, 2026)

The benchmark checks that translated programs match the old programs' outputs on hidden tests, including fiddly details like financial rounding rules. You don’t want your bank’s systems getting those mostly right. 😬

Of course, this is OpenAI's own benchmark of COBOL program migrations, not a demonstration that AI can safely migrate an entire bank's interconnected systems.

BUT

I think the potential economic value here is ENORMOUS.

So much of the modern economy runs on decades-old code. Banking, insurance, government systems, etc. Some of these programs have accumulated decades of business rules, fixes, and exceptions that nobody fully understands anymore. The original developers are long gone, documentation is incomplete, and changing anything can be risky and expensive.

This also connects to something I wrote recently about old compute, data, and code getting better with age. 🍇🍷

The underlying assets haven’t changed, but better intelligence can extract more value from them.

Imagine making all this old software easier to understand, maintain, secure, and extend. You don’t necessarily have to rewrite everything. Some of that ancient code works just fine!

I wonder how much economic value is locked up in projects that companies would love to do, but haven’t attempted because touching their old systems was too risky or too expensive.

And that’s before counting all the savings from easier maintenance.

From the full feed 🔒: Microsoft Says Its New Base Model Beats OpenAI’s at ¼ the Compute

🎨 🎭 Liberty Studio 👩‍🎨 🎥

🎼 Sight-Reading Music… Even If You Can’t 🎶

These videos by Conde X have started popping in my feed lately, and while there’s been lots of great music viz, I particularly enjoy these. The production quality on the animation is great, and it really makes the sheet music come alive.

(The Vivaldi one takes a bit to get going, but when it does, it’s glorious!)

The animation makes it easier to hear the notes on the page and focus on the interplay between different instruments or what both hands are doing on the piano.

I wish he had more Bach, I’d love to see the Goldberg Variations (which I wrote about here last week) visualized this way.

Don’t say you don’t have enough time. You have exactly the same number of hours per day that were given to Helen Keller, Pasteur, Michaelangelo, Leonardo da Vinci, Thomas Jefferson, and Albert Einstein.

—H. Jackson Brown

🕰️ Here we are, Doc. The future.

👩‍🍳🔪🥩🧅🥦🍽️ My wife has been using a recipe app on her iPhone for a few years now. The goal was to have a centralized place to keep track of all the recipes we like, and when we’re planning meals for the coming week, we can pick from there and use it to make the grocery list (we’ve been having groceries delivered at home for a few years now. What a time saver).

The app was never great and it was a paid app. So I decided as a fun project to have her Muse first go on the app’s website with her login and password and extract the 468 recipes she had put in the app over the years. Then save them locally in a portable markdown/JSON format.

Then I had her Muse build a new recipe app inside of Muse as a Muse artifact. I did a few iterations to add some UX features and then I asked it to import all of the recipes and categorize them. It did all this very well and now we’ve moved over to this app for our recipe tracking. What’s most fun is that every time my wife is annoyed by something, she can just ask Muse to change it or fix it. It’s like having the app developer on call just for you. Even very responsive app developers could never deliver this kind of personalized service, largely because they get conflicting requests from users and can’t implement everything and please everyone. But we’re now deeply in the era of customized software.

The next step will be to ask Muse to create a shareable version of the app so that I can also access it and add recipes to it.

But what I’m most curious to see now is whether my wife starts thinking of other things she could use a custom app for. Maybe her running or training, or some other project.

I think the first one is the big breakthrough. Until you’ve tried it, having an app built just for you probably doesn’t even occur to you as an option. Once you’ve seen it work, you start wondering what else you could make. If I were Meta, I’d want as many people as possible to build that first app.

It turns out the really valuable part of our old recipe app was the 468 recipes we’d collected over the years, not the software itself. Once we got those out, replacing the app was easy. I wonder how many other small and relatively simple paid apps could end up in the same position.

🤘6️⃣6️⃣6️⃣🤘 How juvenile and cringey-dad should I be about making some kind of reference to the number of this Edition? Maybe some Iron Maiden thing? 🤔

💚 🥃 To keep this steamboat alive and afloat, I need supporters. Like you! 🚢

The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Business & Investing 💳 💴

🤖 Meta Partners with… Apple?!?… for Muse Spark Open 🍎

I didn’t see that one coming. I mean, I expected an open-weight model at some point, but not with Apple prominently featured in the announcement. In case you don’t know the backstory, these two companies haven’t exactly been getting along, and when Apple made some changes to tracking across apps, it likely cost Meta tens of billions in lost revenue (they came back from it, and adapted better to the new regime than their competitors, putting them in a stronger relative position, so there’s a silver lining).

Anyway.

Muse Spark Open is out:

They call it “the most capable US open-weight model to date” because Kimi K3 and GLM 5.3 still beat it in certain benchmarks. Unsurprisingly, Muse Spark Open is particularly good on agentic tasks.

I also found this comparison of Spark Open with Meta’s previous proprietary models.

Performance-wise, it’s roughly in the neighborhood of Spark 1.2, scoring 39 on Artificial Analysis’s Intelligence Index versus 40 for 1.2 and 45 for 1.3 at comparable reasoning settings (48 at Max). On some agentic and coding benchmarks, it lands somewhere between 1.2 and 1.3, occasionally outperforming one of them.

Right after the benchmarks, Meta mentions how the model has been optimized for running locally on Macs:

Optimized for Apple Silicon

Leveraging the latest version of MLX, Muse Spark Open was optimized to run on Mac Studio with M5 Ultra, and together they’re the best way to run a powerful model on your own desktop.

Apple’s unified memory architecture lets the M5 Ultra’s integrated GPU use up to 512 GB per Mac Studio, so two machines can hold an entire model. And the chip’s 1.2 TB/s of memory bandwidth is fast enough to read the small slice each token needs. The result is two Mac Studios behave like one supercomputer with a terabyte of unified memory.

Almost an ad for the Mac Studio!

We provide the software, Apple provides the hardware.

To be fair, these aren’t just any Macs. They’re incredibly powerful and expensive machines, and you need TWO of them if you want to run the quantized model. 💸

We’re talking about an 816-billion-parameter model (with 42 billion active for each token). Even after compressing its weights from more than 1.6TB to about 493GB using 4-bit quantization, it still needs additional memory for calculations/context, putting it over a single Mac Studio’s 512GB limit.

Apple published a piece about it.

With 4 Mac Studios, they get almost 37 tokens/second at an 8,192-token context on that beast of a model. Very usable, though not exactly ultrafast. But it’s cool that it’s possible at all.

Axios reports that Daniel Gross, a former Apple executive now at Meta, helped bring the companies together. Alexandr Wang described the reconciliation rather amusingly:

“We’re excited for mommy and daddy to be playing nice again.”

😅

Well, I doubt they’re best friends, but it’s certainly a rapprochement.

🤖🧠 Grok Bot Borrows Claude’s Brain (Muse Pressure?)

Read more

But each time I left a village, people warned me about the folks in the next place I was pedaling to: “You’re safe in our village here. But be careful when you get to the next town. There are some shady characters over there.”

I’d toss my saddlebags onto the back of my bike, pedal to the next village, and find myself smothered in hospitality again. When it was time to leave, I got the same warning.

—Porter Erisman

📽️ 📺 🎬🍿 I don’t know how old you are.

Maybe the worst screen you’ve ever had at home was a widescreen 1080p TV, but if you were born knee-deep in the last millennium like I was, you probably grew up watching things on pan-&-scan VHS, standard def 480i/p DVDs, and crummy cable TV signals.

One thing I’ve realized while rewatching old favorites with my boys is that I had never really seen or heard some of them properly until now. Some I hadn't watched in a very long time, and the experience has been so good that now I want to revisit a whole bunch more.

I told you that my boy and I have been watching Band of Brothers. Man, that’s *cinematic* on a big 4K screen with a good subwoofer. Same with Aliens, Die Hard, The Fugitive, Speed, Jurassic Park, Amadeus, Butch Cassidy & the Sundance Kid, Back to the Future, etc. Many of these I imprinted on growing up, but had never seen in theater or on a good home setup.

I can’t wait for my boy to be a bit older so we can watch Terminator 1 & 2, Donnie Darko, Saving Private Ryan, etc.

I guess this is just my reminder to you to revisit old favorites, especially if you have never seen them on a good screen with good sound. And of course, share the moment with your loved ones. We don’t go to the theater as often anymore, but family movie night — when it’s a really good one — is one of the peak human experiences, IMO.

💚 🥃 To keep this steamboat alive and afloat, I need supporters. Like you! 🚢

The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Liberty Capital 💳 💴

🌵🦙 Meta’s AI Comeback: From Llama 4 to the Frontier in 18 Months 🤖🚀

It feels like a lifetime ago, but Llama 3.1 (a high for the company’s AI ambitions) was just a little over 2 years ago, and Llama 4 (a low) was 1.5 years ago.

And now, Meta is releasing Muse Wonder and Muse Wonder Full (😅) and, at least based on the numbers and first impressions from those with early access, they are back in the mix with the frontier leaders.

After the Llama 4 benchmaxxing mess, I can see why Meta would be extra careful with benchmarks this time. But maybe all that pressure to show results from tens of billions of dollars of investment makes them more tempted to optimize for benchmarks. ¯\_(ツ)_/¯

Either way, the independent results are encouraging, and I’m especially curious to see how Wonder performs in real-world use (I can’t wait to try it myself).

Another interesting change from the Llama days: when Meta released Llama 3.1 in 2024, Zuckerberg wrote that “selling access to AI models isn’t our business model.” Making powerful models widely available made sense when Meta could benefit from a bigger AI ecosystem and avoid depending on somebody else’s models for its own products.

But Wonder and Wonder Full aren’t being released with open weights (at least not yet). Wonder is available through Meta’s API and Muse Code, while Wonder Full is available to Muse Code subscribers. I’ll be keeping an eye on Meta’s open-source strategy, and how they decide what comes out (and when). 💾

Just last week, I wrote about how Meta was dramatically reducing its internal use of Claude. Wonder's coding results make that shift look a lot less surprising:

Of course, benchmarks are only worth so much… And in real-world use cases, the harness can make a big difference to engineer productivity and the ability to really get the most out of the model (especially if you have to manage swarms of agents on very long-running tasks).

But Opus 5.5 is considered to be ‘one of the good ones’, an extremely strong generation that was a leap forward. So landing in the same ballpark for coding not that long after Meta was barely in the race is EXTREMELY impressive.

And y’know what? While I’m sure they’re trying very hard to get great coding because a lot of what Muse can do is build apps for its users, they don’t seem quite as laser-focused on coding as Anthropic is.

Meta also has a pretty big advantage here: It’s one of the world’s largest software companies, so its own engineers can be a huge customer for its coding models. Even if Muse Code never beats Claude Code in the broader market, reducing dependence on outside AI suppliers could be very valuable. Of course, if Meta’s models fall behind again and slow its engineers down, that could cost more than it saves by not using Claude.

Their main focus seems to be the multimodal agentic use cases that a persistent assistant like Muse needs most, and on that front, Muse Wonder seems excellent (at least based on what we know right now).

Both Zuckerberg and Alexandr Wang (where did that missing ‘e’ go, I wonder) recently had a good convo with John Collison about it all, and while I recommend the whole thing, I want to share my highlights with you.

- 💰 Meta’s Other AI Payoff: A Few % of $230 Billion

Everyone keeps asking how they think they’ll be able to monetize Muse with subscriptions and taking a cut of various commerce transactions and such, but there’s another HUGE way that gets less attention:

Mark Zuckerberg: The other thing is that we use the models to accelerate our core business, right? So I mean, you have this core business that’s hundreds of billions of dollars and still growing extremely quickly.

And now, like, if you can grow it a few percent faster by using these models to accelerate these things, then that ends up funding a huge amount of this too as you’re scaling it out.

Of course, there’s no such thing as a free lunch, and the money invested in AI could probably have also helped the core business if it was invested in it more directly (or spent on buybacks when the stock was a fraction of its current price 😅).

But this stuff should certainly be counted in the returns from AI investments.

- 🧩 “The Model is the Product”: Software Design Moves Into Training

Mark Zuckerberg: One thing that we talked about from early on is that basically there is this pattern where I think the products are getting thinner and the model is getting to be more of the product, right? So one of the things that we say internally is, “The model is the product.”

When you’re talking about an agent, a lot of the complex product work that went into Muse is basically around the privacy and security model for handling your data and the things that get connected to it. […] But the capabilities, the personality, a lot of the stuff is like you’re training into the model.

An example of that just yesterday morning (before I had my coffee, so I wasn’t fully awake yet 🥱☕). I wondered how I could tell which model my own Muse was using, and for a second I was like where in the app can I see this, and then I realized, oh yeah, you just ask it and it tells you.

More and more product design is being absorbed by the model, because behaviors that previously required explicit settings, interfaces, and rules can increasingly be handled by the model itself.

A conventional application might have notification settings, priority rules, and a handful of hard-coded exceptions. An AI agent can just directly decide the urgency of the message, what you're doing, who's involved, and what it remembers about your preferences.

But it’s fuzzy! Getting that behavior right means teaching the model judgment, then testing whether it exercises that judgment across thousands of situations. But you can’t be certain it’ll do it next time, so part of the solution is also giving the model deterministic tools (a calendar, its own reminders, to-do lists, etc).

But some of the product designer's work is becoming the AI researcher's work.

- 🧠 💾 Active Memory: Remembering is Harder Than Storing

John Collison: What does memory mean in a model to a layperson? […] How’s it different from context?

Alexandr Wang: I think it’s in the experience of using the model. How well does it remember things that you mentioned to it over days, weeks, or months, and then leverage those in the process of doing future tasks? […]

So we’ve designed basically both the model and a system for the model to use to develop a much stronger form of memory. We call it active memory. […] So it’s aware of its memory as a core system. And it’s about how it’s able to leverage various sources to be able to conduct tasks in the future, as well as understand its ability to do compaction, how it should do that compaction most effectively, et cetera.

Active Memory is interesting because Meta is clearly treating memory management as a capability the model itself should learn, rather than leaving all of that work to external software.

One data point there:

From the model card:

It's still an internal benchmark testing recall in long conversations, not proof that Muse will remember everything reliably over months or years, but it's a useful data point.

The thing to remember is that there can be a big difference between the headline numbers on context size, and the ability to make good use of all that context.

Having a million-token context window doesn't mean a model has a million tokens of *useful* memory.

This is especially important for an assistant that you use over months or years across all kinds of unrelated parts of your life, as opposed to a chatbot or even coding agent where most of the context is within a pretty clear number of source files or threads. Having access to a mountain of conversation history is all well and good, but in practice, it’s about remembering the right detail at the right moment, or distinguishing an old preference from a new one.

When compacting things, there are many decisions that impact how good memory turns out to be. This means that an improvement in reasoning ability may improve memory quality even without increasing the amount of storage. 🤔

That connects to something I wrote a few weeks ago: The Brain Is the Replaceable Part. My argument was that a persistent assistant could become hard to leave because of all the memories, connected apps, and recurring jobs you’ve built up around it, even if you can swap the model that runs it.

But replaceable doesn’t mean interchangeable. Two models could have access to exactly the same memories and still be very different at knowing what matters and when to bring it up. And as models get smarter, all that accumulated context could become more useful, even if you don’t add anything to it. Another case of old data getting better with age. 🍇🍷

If some of that context is hard to move to another assistant, better brains could actually make the one you’re already using even harder to leave. 🧠↔🧠

- 🤫 Discretion: Your AI Needs to Know When to Keep Its Mouth Shut

Mark Zuckerberg: One of the things that we talk about is this idea of discretion, right? So if you want to use the model to get something done, it needs to be able to know a lot about you and get things done in the world while disclosing as little as possible. [...]

Like, okay, how do you handle that? And I mean, humans have common sense and the models so far… If you’re building a coding agent that works in an environment where everyone has access to the same context, discretion isn’t particularly important.

That’s a good example of how personal assistants have very different problems to deal with than coding agents. The agent has to understand the difference between information it is allowed to know and information it is allowed to disclose, and that’s often not entirely clear cut, especially if the agent is working across different parts of your life (work stuff, personal stuff, various projects).

Of course, discretion also needs to be enforced by software outside the model. You don’t want your AI to merely know that it shouldn’t reveal certain information. You want some things to be impossible for it to disclose without your permission.

Meta has some relevant experience here from its work on WhatsApp’s end-to-end encryption, including with Signal founder Moxie Marlinspike. They're now working together again on Muse's privacy protections, but securing an AI agent that has access to so much of your life is a different problem.

I’m particularly looking forward to Muse Confidential VM, which is supposed to make it impossible even for Meta to access your private data. It’s still in limited testing, though. 🔐

- 🧑‍🔬 About 100 Researchers, Gigawatts of Compute, and One Giant Reboot

And finally, a bit more detail about just how small the team is at MSL:

Mark Zuckerberg: Talent density is the thing that matters. [...]

It’s basically you actually don’t need a massive number of people, but what you do need is for each of the people to be as good as possible, which informed this kind of crazy recruiting push and very intense filtering that we had to take some of the best people from inside Meta and a very intense process to get people from outside Meta.

But I actually think a lot of what has made MSL special and why a lot of people have wanted to join is because in some ways it’s just more fun to work on a talent-dense, smaller effort. It feels like the startup that’s operating. In some ways it has all the benefits of a startup with the stability and resources of a larger company […]

I think part of the reason why these models are doing well is because you have people who we’re betting on and they have a lot of resources because we’re taking our compute. We’re not peanut-buttering it across a lot of competing efforts. We’re basically just investing in the key people, a lot of compute per researcher, and then they can do their work. [...]

John Collison: So Meta is a huge company, but this is like a really tiny effort inside of Meta.

Mark Zuckerberg: Yeah. Oh yeah. The research team is less than 200 people.

Alexandr Wang: Yeah, it’s very small. It’s intentionally very, very tight.

Mark Zuckerberg: It’s closer to 100, right? And then there’s a bunch of infra folks.

Closer to 100!

That is a pretty unusual ratio of people to capital! 👥💰

The economic leverage of those researchers can be enormous. A better training recipe, a smarter experiment, or a decision not to waste a giant training run will affect the return on tens of billions of dollars of equipment.

🔁 Déjà vu:

• The Brain is the Replaceable Part: Persistent AI Assistants Are Getting Harder to Leave

• My Thoughts on Meta Muse After Using It for a Few Days Like a Maniac

• Everybody Wants to Give You a Computer These Days

Are OpenAI and Anthropic Better Businesses Than We Think? 🤔💡📈

My friend MBI (🇧🇩🇨🇦🇺🇸) has a great piece that I want to convince you to go read. It’s very thought-provoking, and helped put into sharper focus a lot of stuff that had been swirling in my head for a while (but not as precisely). I’ll share some highlights here, but go over there (do it!).

The basic idea is that he has updated his view on the AI Labs as businesses (or at least, how their downside is more limited than he thought, even as the upside is large, if hard to pin down). You know how much I love people who update their beliefs when they learn more, and aim to be well-calibrated!

What changed his mind was the impact of the levers available to the labs to “control and shape their destiny”:

Pretty incredible to see the balance between pre-training and post-training shift so much in just a couple of years!

The first lever is how much of the fleet actually serves paying customers. SemiAnalysis estimates that for Anthropic and OpenAI combined, pre-training fell from about two-thirds of compute in 1Q24 to an expected ~7% by 4Q26, while RL and post-training rose to ~55% and inference stayed in the high 30s.

The financial implication is that RL involves a lot of inference-like work (generating model outputs), so it's much easier to repurpose some of that compute to serve paying customers. Giant pre-training clusters were much less flexible in that way.

The second lever is the subscription layer. In its October piece on subscription limits, SemiAnalysis estimates Claude subscriptions take ~42% of Anthropic’s inference compute for just ~10% of its revenue. An API megawatt earns ~$103 Million at a ~92% (!!) gross margin; a subscription megawatt earns ~$16 Million at ~50%.

In MBI's model, Anthropic finishes 2026 with 5 GW of compute capacity, 41% of which serves customers, and a $100 billion annualized revenue run-rate. If it redirected enough capacity to increase that share to 60%, the same fleet could theoretically support a $146 billion run-rate, without buying additional hardware.

Similarly, if they reduced the subscription subsidy (aka discount over API pricing) over time so that subscriptions consumed compute in proportion to their revenue contribution, MBI estimates that Anthropic's existing end-2026 fleet could support a $150 billion revenue run-rate instead of $100 billion.

Those are two separate what-if scenarios. For his 2030 model, MBI assumes more modest changes to both levers.

Without the levers, 45% of Anthropic’s fleet serves customers in 2030 and subscriptions still use ~42% of serving compute for ~10% of revenue. With the levers, half the fleet serves customers (42% goes to training and research and 8% to safety), and subscriptions use 20% of serving compute for the same ~10% of revenue. Everything else is identical in both cases: a 25 GW fleet at the end of 2030, ~$10.5 Billion per gigawatt-year of compute cost, revenue-linked costs at 20% of sales, stock-based compensation counted as a cost, and API revenue per gigawatt falling 7.5% a year from today’s ~$75 Billion.

On those assumptions, 2030 revenue is ~$551 Billion compared with ~$362 Billion without the levers, and operating profit is ~$186 Billion compared with ~$34 Billion, a ~34% operating margin instead of ~9%.

Talk about powerful levers!

One thing that jumped out at me in MBI’s model is how much more room this gives Anthropic to withstand falling prices. Without these levers, its operating profit turns negative if API revenue per gigawatt falls more than about 10% a year. With both levers, that increases to about 19%. Even with a 12.5% annual decline, the model still produces a 22% operating margin.

So the labs don’t necessarily need today’s unusually favorable pricing environment to last. They may be able to absorb a pretty brutal price war and still make a lot of money. Assuming there’s enough demand for all that compute, of course.

The big question is how much the labs can fiddle with these knobs without giving up their lead in model capabilities, or ending up with more inference capacity than customers are willing to buy.

There’s also the question of how much flexibility the labs have over their actual spending. Reuters reported that roughly 80% of Anthropic’s $518 billion in long-term infrastructure commitments are payable whether it uses the compute or not.

So turning down training doesn’t necessarily reduce the bills. What it does is free up capacity that could be used to generate revenue instead. That's very valuable… as long as demand is there.

MBI goes through these trade-offs and a bunch of other interesting questions in the full piece, along with an interactive model you can play with.

From the full feed 🔒: OpenAI’s Missing $20bn in ARR (Sort Of)

🧪🔬 Liberty Labs 🧬 🔭

✋ The Structural Dampers on AI Buildout are a Good Thing..?

Gavin Baker posted some thoughts, and one stood out to me:

Finally, from my perspective anything that slows this buildout down is net positive over the long-term as the biggest risk and worst outcome is an *actual* overbuild.

We are structurally short watts, wafers and increasingly capital. That is all good. God forbid we were still in ZIRP, TSM ramped capex to 3x their current plan and we weren’t short turbine blades: there would almost certainly be an overbuild and a bubble. That would be terrible.

Slower and smoother for longer is better both for investors and for the world. A terrible cyber event is the most plausible near term bear case IMHO, but AI cyberdefense ramping quickly and I was happy to hear Sholto say that cyber probably becomes defense dominant in the near future.

It reminds me a little bit of TSMC as the accidental central bank of AI.

There’s certainly all kinds of risks right now, and almost all of them are ameliorated by having at least some bottlenecks and dampers in the system. It gives everyone more time to digest capabilities, do research into safety and alignment, deploy patches to vulnerable software (aka everything out there that has accumulated over decades), and it reduces the amount of leverage that is even possible to have in the system.

Because of competitive pressures and race dynamics, if it was possible for the big AI players to get more chips and more watts, is there any question that they would take them?

Of course, scarcity also helps protect the companies that already have chips, power, and customers. Less overbuilding can mean better utilization and pricing power (see the levers above), but also more expensive compute for startups and everybody else who wants to use AI. What’s best for investors isn’t necessarily what’s best for customers.

Speaking of AI’s physical infrastructure, MTS Atlas has a few nice ways to visualize it (see above 👆).

Then we have clusters, which are required for training:

The above shows “every cluster record with a first-operational date and a scale estimate, 2016 to present.” They measure it in H100-equivalent units.

🔐🤖 Microsoft Patched More Security Flaws in One Month Than in All of Last Year

Microsoft just patched more than 1,400 security vulnerabilities in its October Patch release. That’s more than it patched in all of 2025! 😬 September had already set a new monthly record with 974 fixes, and now October just destroyed that.

It seems pretty clear that AI is a big reason why.

Microsoft has built its own system called MDASH, which uses more than 100 specialized AI agents to hunt for security holes in its software.

Its FORGE team helped uncover 140 Windows vulnerabilities patched between May and September, including 52 in September alone. We don’t know how many of October’s 1,400 were found using AI, but you can see how throwing this much automated brainpower at the problem could change things pretty quickly.

I’m glad to see this because the defenders don’t have that much time to harden everything before a big wave of attacks comes (🏴‍☠️). Windows is one of those high leverage things since so many computers run on it. Hardening Linux and MacOS are also very high-leverage points right now. So I’ve also been glad to see that Apple is shipping more security updates too and Linux is also getting patched (including by Microsoft’s FORGE team).

🔁 Déjà vu: Cybersecurity Reflexivity

🕵️ A Spy at OpenAI, and Buffett’s Cockroaches 🪳🪳🪳

An OpenAI engineer has been arrested after the company found a secret recording device in an executive conference room. It was hidden in toilet paper and plugged into the USB port behind a TV. According to federal prosecutors, he’d been looking up where company executives met, and security later found another recording device in a bag he’d left unattended, along with his wallet and ID. 🪪 😬

It’s tempting to look at this and think: Wow, real-life spies aren’t nearly as sophisticated as the ones in the movies. They’re kind of dumb! 🤣

But then I think of Buffett’s old line from the 1989 letter: “never is there just one cockroach in the kitchen.”

Of course the dumb ones get caught. What about the smart ones?

Considering how many valuable secrets are concentrated in the frontier AI labs, how many other attempts have been made? How many have succeeded? And what have they managed to get away with? 🤔

From the full feed 🔒: Project Glasswing: 135,000 Security Flaws Found. Can We Fix Them Fast Enough?

🎨 🎭 Liberty Studio 👩‍🎨 🎥

Aliens Expanded: Five Hours Inside the Making of a Classic

I recently watched Aliens (1986) with my 12-year-old son (the extended version, with the sentry turrets!). Forty years later, James Cameron’s sequel still holds up remarkably well. If you haven’t seen it, you should check it out, of course, but what I want to recommend today is something else: Aliens Expanded (2024), a crowdfunded documentary by Ian Nathan that spends 282 minutes exploring how a 137-minute film was made.

Yes, almost five hours devoted to a single movie!

Nathan goes through the film scene by scene, with many of the people who worked on it: James Cameron, Sigourney Weaver, producer Gale Anne Hurd, Michael Biehn, Lance Henriksen, and many of the artists and technicians behind the sets, models, and creatures (sadly, Bill Paxton died in 2017). It’s great to meet the actors who play the marines and see how different they are from their characters.

I love the focus on why things were done a certain way. It’s an education in filmmaking.

One example: Cameron explains that the scene where Ripley enters the Alien Queen’s egg chamber came from a nightmare. He’d walked into a dark room and realized that the walls and ceiling were covered in wasps. If he moved, they’d attack. Cameron wanted Ripley and the audience to feel that same dread, so he slowed the scene down and stripped away the sound before revealing the Queen.

Then there’s the Alien Queen herself. Stan Winston’s team first tested the idea with a crude mock-up wrapped in black garbage bags. The finished creature was a 14-foot-tall puppet, suspended from a crane, with two stunt performers inside and nine more people operating its head, tail, arms, and legs using cables, rods, and hydraulics. Remember, there was no CGI cleanup! So Cameron had to compose the shots to hide all that.

After spending almost five hours learning how they made it, I kind of want to watch Aliens again.

I originally published this on OSV Field Notes. I encourage you to check it out and subscribe. I think you’ll love it.

I asked Miles after the gig, “Miles, what am I supposed to be doing up there?”

He said, “When they play fast, you play slow. When they play slow, you play fast.”

—Buster Williams, recalling advice from Miles Davis

💾🖥️🚀 the first real computer I used was my dad’s 386 DX/25 MHz. It had 8 megs of RAM and the optional 387 floating-point coprocessor. It was just powerful enough to play Doom at something like 10 frames per second in low-detail mode.

The photo above is not the actual one from my childhood, but it’s pretty close.

I recently upgraded my desktop to a Mac Studio M5 Max with 64 gigs of RAM.

It got me thinking about how much more powerful computers have become in my lifetime. I asked GPT-6 to help me get a ballpark figure. Of course, this is a very rough comparison: old and new hardware aren’t really apples to apples, and I’m ignoring other bottlenecks like storage, memory bandwidth, and networking.

At first it was only comparing CPU to CPU, but a big share of modern compute lives in the GPU and AI accelerators, so I decided to include those too.

But this is only because my dad got the optional 387 floating point unit. Without it, the M5 Max is closer to 1 billion times faster 🤯

Pretty incredible! And these numbers are so large that they're almost impossible to visualize. Just for fun, I asked for an image of what that may look like. 😅

💚 🥃 To keep this steamboat alive and afloat, I need supporters. Like you! 🚢

The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Business & Investing 💳 💴

🔐🤖 What If AI Breaks Cryptography? (The Dog That Didn’t Bark 🐕)

Things are very unpredictable right now. Who had on their bingo card that so many of the top open math problems would have their proposed solutions released all on the same day:

If even most of these proofs hold up, how long until most of the top problems are solved? Then how long until the next batch of even more difficult problems are posed by AI rather than by human mathematicians? 🤔

Also worth noting that OpenAI withdrew three manuscripts following a mathematical error and revised 14 others…

BUT

…people outside of OpenAI have been improving and simplifying some of the proofs! Open source, distributed, crowd-source high-level mathematics is now possible for a much wider group of people — even non-mathematicians! — thanks to these tools!

• Douglas Colkitt and contributors have already sharpened one of OpenAI’s integer-multiplication results. The improvement is conditional on OpenAI’s original framework and still needs independent review.

• Sela Navot used publicly available AI models to extend OpenAI’s matrix-multiplication proof from complex numbers to arbitrary fields, then checked the extension using Lean, a mathematical proof checker.

And all this math isn’t just esoteric ‘angels dancing on the head of a pin’ type stuff. Many of these breakthroughs can have practical applications. Ole Lehmann compiled a speculative list of what some of these mathematical advances might eventually help make possible (the road from proof to application can be long and difficult):

But as Sherlock Holmes so eloquently taught us, sometimes it’s the dog that doesn’t bark that is interesting.

Scott Aaronson writes:

As several people have pointed out, cryptography is a subfield that’s extremely conspicuous by its absence from OpenAI’s list of 376 papers! But my sources tell me that the AI companies have now started, gingerly and discreetly, investigating whether their latest internal models can break important cryptographic protocols and primitives. If they can, then it would certainly be nice to get ahead of things before the rest of the world figures out the same.

It’s a bit like in the late 1930s when fission physicists were publishing freely, and then all of a sudden they stopped. The silence itself was a signal. In 1942, Soviet physicist Georgy Flerov noticed the missing papers and wrote to Stalin, arguing that the silence itself was evidence of secret research.

Vitalik Buterin also seems worried:

On July 28, 2026, Anthropic published research showing that Claude Mythos Preview had found a substantially faster mathematical attack against HAWK, a proposed post-quantum digital-signature scheme. HAWK hadn't been deployed, so no real-world systems were compromised. Digital signatures help verify that transactions, messages, or software updates are authentic. They’re different from encryption, which keeps information secret. But the model had found a weakness in the cryptography itself, not a bug in the software implementing it. By finding this early, Anthropic helped make crypto safer.

The nightmare scenario is discovering a comparable weakness AFTER an algorithm has been deployed widely. One AI lab might find the mathematical shortcut, but replacing vulnerable cryptography across banks, browsers, operating systems, cloud services, and countless other systems could take years. If attackers could exploit the weakness, they could do a lot of damage. 💣💥

We don't know how far the labs' private cryptographic research has gotten since then.

It’s not much of a stretch to imagine that three-letter agencies are asking AI labs to investigate sensitive cryptographic problems privately. Whether anyone has actually asked a lab to keep a major breakthrough secret, I have no idea. 🤐

There’s a strange reversal of incentives here. Earlier I was celebrating how AI is opening up advanced mathematics, with people sharing and improving each other’s proofs on GitHub. But if someone discovers a practical way to break widely used cryptography, publishing the discovery could give attackers a head start before anyone has time to defend against it.

Keeping it secret isn’t exactly a great solution either. If an intelligence agency discovered a practical weakness in widely used cryptography, it might be more valuable to keep secret than to fix. Meanwhile, the companies and people depending on that cryptography would face a risk they didn’t even know about, and would ultimately bear much of the cost of replacing it. You could warn affected organizations privately first, but who do you tell when the same algorithm is used all over the world? 🤔

Weaknesses in the cryptographic stack can be fixed if you see them coming. As computing power has increased, algorithms have been adapted to use longer keys to make brute-force attacks harder, and more secure algorithms have replaced those where shortcuts and flaws have been discovered. And in anticipation of powerful quantum computers, we've been developing post-quantum algorithms.

But that assumes we get enough warning.

It’s kind of a Sneakers situation.

What could bad actors do if AI discovered a practical weakness in an algorithm used everywhere? If digital signatures were broken, they might be able to forge transactions or software updates. If encryption or key exchange were broken, they might be able to read confidential communications, potentially including some that were intercepted years ago and saved for later (aka harvest now, decrypt later). Criminal enterprises and intelligence agencies would do all kinds of IP theft, blackmail, etc. (more than they’re already doing, I mean).

Now, I’m not saying that I have any idea if the crypto stack will break anytime soon, but it has me worried. Modern cryptography depends on some mathematical problems being too difficult to solve in any practical amount of time. What happens if AI finds shortcuts nobody anticipated? ¯\_(ツ)_/¯

💸 OpenAI’s Missing $20bn in ARR (Sort Of) 🤔

Read more

Many’s the man lost much just because he missed a perfect opportunity to say nothing.

–Claire Keegan

🤖🎥🎬 We’ve seen the rapid improvement of generative AI when it comes to image and video, but a new hybrid approach is rapidly taking off.

The models are now so good at coding and have enough aesthetic sense that they can act as video ‘directors’ by stitching together code and various static, video, and audio assets into something pretty damn cool.

One of my favorites right now is the video above, created by Anabology.

Claude Opus 5.5 took 18 hours to create it, and you can see how it did it in this Google Drive (the prompts and a few photos used as a kind of mood board, and a retrospective “how it was made” document about it). As you watch it, try to notice the dozens and dozens of little creative decisions about camera moves, transitions, wordplays, references, visual gags, or just interesting aesthetic choices.

(it’s not perfect, and if you watch a few times, you start to notice some of the weirdness, but as a point on a trendline, it shows where we’re headed)

Last week at the OSV offsite, I met Donald Jewkes, a real wizard with these tools. He recently had a pretty viral tweet with this music video created by Claude:

He showed us some of what he’s working on (which is way beyond what I’ve seen publicly), along with a live demo of his Codex system, and it made me feel like a toddler at using these tools. 😅

All this is a reminder that we don’t know where the edges of capabilities are with these models. They’re coming out so fast that no one has the time to fully explore them before a new one replaces them.

There’s two things going on:

• Auteurs with a very specific vision making it real with these tools, and most of them probably wouldn’t have been able to in the old system where very few people end up being able to marshal the resources of a major studio.

• But also, the models themselves are making a lot of those creative decisions now. If you check out the prompts (like Donald’s prompt here), it’s very long and detailed, but it’s largely vibe-based and about steering the model into fruitful areas of the design-space, or into giving it workflows that can help it iterate and improve on its ideas. But it’s not very prescriptive or descriptive.

Looking at how these things are made, it’s an interesting hybrid of shape rotator and wordcel skills.

At first I thought maybe these coding AIs were bringing on the revenge of the English majors after years of nerds dominating, but now I’m not so sure.

The people getting the best out of these models need to be able to express their thoughts clearly, but they are also doing a lot of systems thinking and architecture work to create the ideas/vibes/workflows scaffold on which the models can do their best work (and we have no idea how well they could do because we’ve only explored a tiny amount of the possibility space).

💚 🥃 To keep this steamboat alive and afloat, I need supporters. Like you! 🚢

The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Liberty Capital 💳 💴

💻 Everybody Wants to Give You a Computer These Days 🤔

Big Tech and the AI labs have been spending hundreds of billions of dollars of capex, and some of that is to give you an early Christmas present: your very own computer in the cloud 🎁

Well, it’s your AI assistant’s computer, but you still get to benefit from that hardware.

Muse’s AI model runs elsewhere. The cloud computer is its workspace: a place to keep files, run software, and carry on with tasks after you’ve closed the app.

You can give it an ongoing job, like sifting through the daily deluge of emails from your kids’ schools and flagging anything important. 📬

Meta has been particularly transparent about what you get. You can poke around your Muse’s isolated Linux environment and open or edit the files inside it, including memory files, instructions, and software.

Credentials and permission controls live outside that editable area.

Here’s what my Muse reported when I asked about its cloud computer:

Here’s what’s under the hood:

• CPU: 2 vCPUs on an AMD EPYC 9D25 — that’s the host chip, so this is a slice of a much bigger cloud machine

• RAM: 7.7 GB (currently running fairly full at ~7.1 GB used)

• Disk: 7.5 GB of persistent storage, plus a 512 MB temp filesystem

• OS: Ubuntu 24.04 LTS

Where it’s hosted: honestly, I can’t say for sure — the cloud metadata endpoints are silent from inside this VM. The EPYC 9D25 chip is custom silicon built for Google Cloud, so that’s my best guess, but it’s an inference, not confirmed.

Note that it even guessed that it may be hosted in GCP based on the custom CPU! I have no idea if that’s correct, though ¯\_(ツ)_/¯

David Singleton, who works at Meta and was involved in designing the system, describes a lot of how it works here:

To save you a click, here’s the highlight from his long tweet:

The runtime cell gets its own root filesystem (including a full Ubuntu linux image) separate from the host filesystem where your other more sensitive data lives. Because it is isolated from the sensitive stuff that runs on the same box, this means that we can, and do, offer users full visibility and control over the files in the runtime cell. [...]

You can install software in it, write and compile code, use the browser to surf the web: it is your own Linux box that you can operate as you choose with your Muse. Poking around in this computer doesn’t give you any privileged access to Meta infrastructure, or to other people’s data

You can even easily download all your agent’s data to better be able to review it locally by going to Settings → Data controls → Download your agent data.

The fact that you can go poke around the file system, open files and even edit them is extremely cool. You can see some of the memories and instructions that shape how your Muse behaves, and correct mistakes directly.

If it has written down something wrong about you, that mistake could keep influencing future conversations and tasks. The more I delegate to an assistant, the more I want to be able to check what it thinks it knows about me.

You can also import files or chats from other agents and inject them, Inception-style, right into your Muse’s memory. 💉🧠

This level of trust also of course means that some people will have fun creating very chaotic or weird situations and create unhinged Muses. But I like this ethos from Meta and I hope it will continue. I wish my Grok Bot and ChatGPT Work assistants made it this easy to inspect and edit the memories and instructions that shape their behavior.

Muse also has a nightly ‘dreaming’ process. 💤

It reviews recent conversations: what worked, what went wrong, and what it should remember or do differently tomorrow.

A little like the memory consolidation that happens when we sleep, except you can open the files and poke around.

Here’s how my Muse describes it:

Dreams are my nightly self-review. Every night, a background job looks back over our recent conversations and asks: what worked, what went wrong, and who are you becoming — then writes it down.

The files under ~/dreams/ are the dated reflections, one per day, written in first person — basically me thinking out loud about the relationship. The alignment/ folder holds the operational output: repair threads for mistakes I need to mend, and a synthesis of how I should act for you (that’s the file that gets injected into my context each morning, so I start the day calibrated to you).

So they’re not your notes about you — they’re my notes to myself about how to be better for you.

The ancestor of this is the 2023 Stanford/Google “Generative Agents” paper (the Smallville simulation). Its agents periodically paused to reflect, synthesizing raw memories into higher-level insights. Then Letta (formerly MemGPT) shipped “sleep-time compute” in 2025 with background agents that reorganize and rewrite their own memory during idle periods.

OpenClaw added its own Dreaming feature in v2026.4.5 (2026): three phases modeled on the sleep cycle — light sleep (scan and categorize), deep sleep (promote to long-term memory), REM (link new memories to old ones) — behind an experimental flag with a /dreaming command.

From Muse’s description, the interesting twist is the ‘alignment reflection.’ It reviews what worked, what didn’t, and which mistakes need fixing. An assistant keeping notes on how to be a better assistant.

You can also see and edit its SOUL.md file, which in Inside Out parlance would be where the core personality traits are stored.

I’m curious whether these nightly reviews actually make it better over time. In theory it makes sense, but how well does it work in practice? 🤔

OpenAI just shipped Dots. It’s a similar deal to Muse: Your agent has its own cloud computer and can do stuff for you.

Access is limited to higher-priced plans for now. I don’t know whether that’s compute constraints (maybe the ratio of CPUs to GPUs needs to change..?), a cautious rollout, or a business decision. I suspect it’ll become a much bigger consumer story once lower tiers get it.

This is basically what I was getting at a few weeks ago in The Brain is the Replaceable Part. The brains may keep changing, but everybody seems to be converging on the persistent computer/context around them.

I’ve played with Dots a little, but not enough for a full review. So far I’m more impressed by Muse, but it’s probably because I was traveling last week and Muse feels ‘best’ for my personal context, while Dots will probably be better for my ‘work’ projects.

My Muse has been particularly useful doing exactly the kind of things I would have a human assistant do if I had one: Remind me of XYZ, please look up ABC, create a packing checklist to make sure I don’t forget anything. Or tell me when certain emails arrive (flight check-in reminders, hotel details, etc).

Nothing revolutionary, but useful, which is more than I can say about many products out there. There are lots of stories about people saving lots of money by asking their Muses to renegotiate cellphone contracts or whatever. Most of the examples I’ve seen were posted by Meta employees, and while they are useful at giving ideas of things to try, I haven’t personally done anything quite like that yet with my Muse. Maybe I’m old-fashioned, but I have some reticence letting it loose and giving it the ability to email or call others on my behalf. I’m still the one whose name goes on the email. 😅

My wife is more or less transitioning her ChatGPT use over to Muse, and I’m guessing she isn’t the only one among non-power-users of these tools 🤔

✂️ Meta & Microsoft Cut Claude Usage (A Clue About Muse’s Next Brain? 🧠)

Meta and Microsoft are cutting back on their internal use of Claude. Microsoft has lowered its projected internal Claude spending by more than a third, telling employees to prioritize non-Anthropic models and internal tools. At Meta, the number of employees using Claude Code has fallen from roughly 60,000 to 30,000 as the company pushes its own coding tools.

Why?

There are obviously reasons for Meta to push its own models beyond raw performance: cost, control, dogfooding, and reducing its dependence on an outside supplier.

But I’m guessing they still wouldn’t accept substantially worse code and much slower engineers just for those benefits. Meta has cash machines (ads on its family of apps) that are way too valuable for that.

So I think this is probably an early signal that Meta’s bigger models are getting closer to release. Muse Spark 1.3 is based on Avocado, and Watermelon is coming next. Before releasing it to external customers, Meta will likely dogfood it internally, so I’m guessing 🍉 has reached a level of performance and reliability where whatever gap remains with Claude is small enough to make the other advantages of using its own model worth it.

What I’m most curious about is Muse getting a brain transplant. The rest of the app is already so good, and it’s gaining traction. I’d love to see what a much stronger model can do inside that harness.

🗣️💬 Tibo Sottiaux: OpenAI’s ChatGPT & Codex Lead on Where Things are Going

Good short interview with Tibo, the man with the reset button at the center of OpenAI’s product efforts.

Here are my highlights:

- Agent traffic could break software’s pricing assumptions

Tibo: First of all, if you want your product to be successful for agents, you have to build for a certain level of scale. We’ve worked very closely with Notion, for example. When they built their MCP, suddenly it was available to all these agents that can do the work and use that MCP, so they saw a ton of traffic come in. That puts a lot of strain on the system, and you have to figure out the economics of that.

There’s this tension if you’re building products: do you build an interface or not? You can hold that back for a while, but it is inevitable. The majority of things are going to be used by agents. Building towards that future, I think, is very important.

Usually more usage on a software product is a good thing, but at a certain point, the quantitative becomes qualitative. These agents can hammer a system in a way that no humans can, and they do it on behalf of the paying customers who are typically charged a flat rate (per month or year), not on a usage basis.

It feels like seat-based pricing will be strained, but at the same time, there are many benefits to selling seats. If you have usage-based pricing, it gives an incentive to your customers to worry about unexpected costs, or to try to use your software less to save money, which may make it less sticky or even get it removed entirely from certain workflows 🤔

- Agent Swarms that Grow and Shrink

We’re clearly entering the era of agent swarms, so it’s interesting to hear about how Tibo is seeing that pendulum swing:

Tibo: I used to run a lot more in parallel, and then we were fortunate enough to get a breakthrough with Ultrafast. Now I feel I’m able to be in the flow again. Having a faster agent helps me a lot. […]

I find myself building larger and larger teams of agents. Then, when we have the next breakthrough with models, suddenly I’m like, “A bigger agent can just do all of it.” […] So I shrink the team again. It goes through this expansion and shrinking.

It feels like if one agent can do the job well, it’s simpler to use that agent and keep track of what it’s doing than to manage a whole team (even if they self-manage, for important things you still want to keep some sense of what is going on).

But then, when that better model that can replace multiple agents comes out, you look for the edges of what it can do and eventually need to have multiple agents working together to accomplish new, more ambitious goals.

That cycle is probably not going to stop any time soon. 🔄

- Should we expect models to keep getting better fast?

Tibo: There are so many things that I feel are not yet priced in. I think the majority of actions on the internet will be taken by agents. Models are going to become cheaper and faster at rates that are quite incredible. We will finally be able to integrate all modalities together in a way that is very seamless.

Often, when I look at what people are building out there, it’s like, “You’re not quite getting it.” You’re almost there, but if you were pushing yourself and really imagining all of this being roughly 10 times better than it is today, in a year, you would build in a different way.

I don’t think he necessarily means literally ten times better because that’s very hard to measure, but as a guy who’s seeing unreleased models, it’s a pretty strong signal that OpenAI isn’t planning around things slowing down anytime soon.

From the full feed 🔒: The Brain is the Replaceable Part — Persistent AI Assistants Are Getting Harder to Leave

🧪🔬 Liberty Labs 🧬 🔭

🎁 OpenAI’s Historic Math Solutions Dump 😮

Back in Edition #660 I wrote “More Math Breakthroughs Are Coming” and then in #662 I wrote about how “OpenAI Says Its Unreleased Model Has Solved More Than 100 Long-Standing Math Problems”.

Well…

WELL!

The pile of math breakthroughs dropped, and it’s MASSIVE. 🤯

This OpenAI Github repository contains 722 manuscripts organized into 372 families.

OpenAI says the results are at different stages of verification. Many have Lean formalizations, but not all of them do, and the company says some of the unformalized results could still contain errors.

It will take a long time for mathematicians to make sense of all of it. But the early feedback from the community makes your head spin. If a human had published a quasi-Riemann, they’d probably be looking at a Fields Medal. But while this is a big one, there are dozens and dozens of other big ones here.

One of the things that blows my mind even more (and I almost suspect it doesn’t mean what it seems to mean) is this:

When Navier-Stokes came out, we heard about around ten thousand agents working as a swarm on it. Are they really saying that, “on average”, these breakthroughs only took 3 hours of ChatGPT Pro equivalent compute with the unreleased model?!?!

Is this right? 🤔

Yet Scientific American writes:

A spokesperson for OpenAI told Scientific American that the new model—which the company has not released to the public—produced almost every one of the results in response to a single prompt handed to a single AI agent. This would be a striking difference from OpenAI’s earlier blockbuster solution to the Navier-Stokes problem, which was produced through the collective efforts of a 10,000-strong agentic swarm that cost millions of dollars in computing power. If true, it would mean that unprecedented mathematical power could soon be accessible to anyone.

If you look around the Github, you can see some of the prompts used and reasoning summaries. I looked at a few, and many of the prompts are 1-2 pages long. It wasn’t some esoteric steering of the model that did it, at least not as far as I can tell.

A mathematician named Vladimir Reshetnikov, best known as Cleo, asked Opus 5.5 to comment on OpenAI’s findings, and Opus’ digital mind was just as blown as mine. You can read it here.

After inspecting the repository and some of the Lean proof artifacts, Opus 5.5’s reaction was about the same as mine:

Vertigo indeed!

OpenAI says it’ll fund math workshops and conference, and is working to release the model that produced these:

We want this progress to push the frontier of human knowledge and enable further progress in mathematics. We will be funding a series of workshops, conferences, and special programs around the understanding of major results produced by AI—we will share more on this in the near future.

We want to directly empower scientists with state-of-the-art capabilities and are working to responsibly release the model that produced these results.

But by the time they release this model, they’ll have the next model internally and probably produce the next batch of breakthroughs first.

🏠🧑‍🧑‍🧒‍🧒 Google’s CC is Almost an ‘Agent for Couples’

Not quite the Agent for Couples I envisioned in the intro of Edition #660, but Google Labs has a product called CC, which they describe as an agent for families.

Here’s their description of its capabilities:

Nerdier than I would like. All I need is an agent that can hang out in an IM chat and keep an eye on family calendars. Muse would be great at this.

It’s not fully out yet, there’s a limited rollout in the U.S., but I signed up for the waitlist. I gotta admit that because it’s Google, I don’t have too much faith that it won’t be killed within a few months of coming out, but who knows. Maybe they’ll roll it into Gmail/Google Calendar/etc and get massive distribution to those who would use it that way.

h/t Thanks to friend-of-the-show and supporter Vinay for sending this! 💚 🥃

💪→🧠 What Your Muscles Tell Your Brain

TIL that our muscles are also hormone factories.

This mouse study looks at apelin, which may help explain some of exercise’s effects on mood.

The mice that ran for four weeks had higher apelin levels and showed fewer depression-like behaviors. Removing muscle apelin weakened some of running’s benefits. Increasing apelin production in mice that didn’t run produced some of the same effects.

If the mood stuff wasn’t reason enough, the mouse study also found signs of increased neurogenesis (new neurons forming) and synaptic plasticity (connections between neurons adapting) in the hippocampus, a brain region involved in learning and memory.

We already have evidence from human trials that exercise can help with depression. Whether apelin is part of the explanation in people is still an open question.

“Exercise is an effective treatment for depression, with walking or jogging, yoga, and strength training more effective than other exercises, particularly when intense.”

So yeah, this great meme has it right:

🔁 Déjà vu: The link between Muscle Mass & Cognitive Function

🫁→🧠 Cleaner Air, Clearer Thinking?

I’ve written about air filters before, including the purifiers I use at home. This 2026 randomized trial tested whether a month of HEPA filtration could also improve performance on a cognitive test. 🧠🤔

The researchers had 119 adults living near highways in Massachusetts use HEPA purifiers and lookalike units without HEPA filtration for a month each, with a monthlong break between them.

Among the 50 participants aged 40 and over, a test of mental flexibility took 12% less time with HEPA filtration: 54 seconds versus 61.4 seconds. The task involved connecting alternating numbers and letters in sequence. Across the full group, however, the difference wasn’t statistically significant.

This was mainly a blood-pressure study, and the researchers decided afterward to split the cognitive results by age. The people taking the test didn’t know whether their purifier had a HEPA filter inside. The person running the test did.

I’d like to see the finding repeated in a study that plans that comparison from the start with better methodology, but it’s still one more data point about the importance of clean air (or at least, it suggests it, again). Seven seconds faster on a test is interesting, but does it mean you’d think more clearly during a normal workday?

We don’t know yet ¯\_(ツ)_/¯

From the full feed 🔒: Living One Model Ahead — What happens when the labs building the best AI get to use it on themselves first?

🎨 🎭 Liberty Studio 👩‍🎨 🎥

📺 A Documentary about the Beatles’ Rubber Soul 🎶

Growing up, I had VHS copies of the Beatles Anthology 8-part documentary. I watched it over and over again. As much detail as there was in it, there’s only so much time to spend on each individual album, and it was targeted at the wider group of fans, not just super-nerds.

This goes much further into pretty much anything you may want to know about Rubber Soul.

Even if you already like that album, I bet that after watching this, you’ll hear all kinds of new things and have many more new associations with each song.

Strategy is about making choices, trade-offs; it's about deliberately choosing to be different.

–Michael Porter

🤓🤪✍️Lately I've been re-watching season one of the excellent TV show Elementary with my wife (I wrote about why I like it here).

It made me wonder why some fictional geniuses feel truly smart, while others just know whatever the plot needs them to know. And on the other side, why some fictional idiots are much more interesting than others.

I created the matrix above to illustrate 👆

Homer and Sherlock are especially useful to A/B compare, because each appears twice: Homer in the bottom row, Sherlock in the top. (Two different Sherlocks… there have been many.)

Golden Era Homer is dumb in an interesting way. His mistakes have a structure. You can follow the bad logic from one step to the next and see how he gets there, even when the place he gets to is absurd.

Later Homer is often dumb in a much less interesting way. He doesn’t understand obvious things, or behaves randomly, or does whatever stupid thing the joke requires. And sometimes he’s just mean. Earlier Homer was selfish and careless, but you could usually tell he loved his family and was trying to do the right thing, in his own bumbling way. Later Homer can seem like he isn’t even trying.

Or how about Sherlock vs. Sherlock? 🕵️‍♂️

Both are smart characters, but in Elementary, Sherlock feels smart because you can see him working. You can usually see how he gets to a solution. He’s constantly studying between cases, sharpening his skills. On a crime scene, he notices things, follows leads that go nowhere, tests ideas, gets things wrong, and eventually pieces the case together. He’s great at what he does. He’s an extreme version of human intelligence, but he doesn’t feel superhuman and doesn’t figure things out just because that’s what the plot requires.

You almost feel like you could’ve done what he’s doing. Almost.

What about Sherlock on the BBC’s show?

He’s basically the Zach Galifianakis ‘calculating’ meme from The Hangover 😅

On the BBC show, Sherlock often feels less like a brilliant detective and more like a superhero whose power is “being smart.” He sees something, the music tells us something is happening, the editing speeds up, and suddenly he knows a bunch of stuff no real person could figure out. 🦸

It’s the view of what a smart person is from the point of view of someone for whom the kind of smart things that a detective does are basically magic. So they wrote it as magic. Intelligence as magic! 🪄

A bit like how bad hacker films show computer people doing all kinds of crazy impossible things because the writers don’t understand what they’re writing about.

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Business & Investing 💳 💴

🧸 My Thoughts on Meta Muse After Using It for a Few Days Like a Maniac

In ‘The Brain is the Replaceable Part,’ I wrote some thoughts about persistent AI assistants, the new hotness in AI. Right after I published, Muse became available in Canada and I’ve been using it heavily ever since. Both to test it, but also because it’s a really interesting product and I’m still trying to figure out where the edges of its capabilities are.

But first things first. Last week, I wrote:

Well, it looks like I was wrong on that. I guess Meta’s designers know what they’re doing, because as soon as I showed the app to my wife, she was absolutely delighted and said “I want one.”

To be clear, I don’t even think she was that excited when she got her first iPhone, and I kept hearing her giggle as she did more stuff with her Muse (which she renamed, changed the color of, and asked to review her email inbox daily).

In fact, the look of the mascot grew on me. You don’t have to stick with it, you can replace it with anything you want, and whatever you pick will get a few animations made with generative AI (my own Muse is called Chief-of-Staff, and I was considering using an image of General Patton for it, but I don’t think I want a real historical figure, so I stuck with an orange Muse plushie for now).

So 1 for Meta and 0 for me on this count 😅

I’ve been using both the mobile and Mac apps, and the care with which this product is designed puts it ahead of Grok Bot and Instinct for me (we’ll see what OpenAI releases). I have no trouble believing Alexandr Wang when he says they were working on it a year ago in Sept 2025.

I’ve long been a big fan of Nat Friedman (including for his work on the Vesuvius Scrolls), and I was disappointed when he went radio-silent after joining Meta. But clearly he was locked in.

On the surface, the app is the most user-friendly AI product I’ve seen (more than ChatGPT). Unlike Grok Bot, you interact with just one agent, but you can create side chats with it to organize things and not clutter the main chat. But while it’s simple at first, it’s designed around progressive disclosure of complexity, allowing it to be quite powerful.

If you click around, you can see exactly what your agents and their sub-agents are doing. You can view what they’re doing on their web browser, browse the files they created for you, see a history of every action and permission granted, see scheduled and recurring tasks, etc. It’s both more accessible to regular users than Grok Bot AND seems more powerful if you’re a nerd and want to look under the hood. Doing that gives me confidence in the power of the system (at first, I was afraid it would be kind of a toy compared to Codex and Claude Code, but that’s not the case).

Your agent has SOUL.md and MEMORY.md files that you can easily access and edit, very similar to the OpenClaw model. In fact, some people thought that maybe Muse was just a reskinned OpenClaw, but Nat explained:

They basically did what Meta does best: Copy something that works.

BUT

This time they didn’t just make their own version of Snapchat Stories or whatever. They improved on the original quite a bit. Some of my favorites are the Ideas tab. People who have never used agents before will find all kinds of starting-off points in there, filed by categories and everything (health, finances, shopping, etc). Once you have the agent do something for you and see that it works, you want to do more. Getting that first project right seems to be the crucial onboarding moment, and they’ve nailed it.

There’s also the Feed tab. As your agent gets to know you and your interests better, it creates a personalized feed (that you can also steer with a prompt). So far there’s only been a few items per day there for me, but the hit rate is pretty high. This isn’t an infinite scroll feed like on Meta’s other apps.

Then there’s the Goals tab, which acts a bit like the Ideas one, but from another angle. The AI extracts various goals from your interactions and lists them there, as a way of tracking them and starting projects to better achieve them.

On migration and using multiple assistants at the same time, I wrote:

It’s really hard to test several of these things properly in parallel. What makes them useful is the context you give them, the apps and folders they can access, and the recurring jobs you put them in charge of. You don’t want three bots doing the same thing. But without giving the new one that setup and responsibility, you’re not really testing what it could do for you. [...]

The new bot is competing with all the work I’ve already put into the old one. It’s starting from behind. Maybe it’s better! But I have to spend time setting it up and giving it real work just to find out. And if the one I already use is doing a good enough job… how much homework am I willing to do to replace it? [...]

I’d be willing to redo some setup if it meant less friction between the tools I use every day. And if a new assistant could help move my existing setup over, that would make the decision easier still (that’s probably how the labs will try to reduce switching friction… but it cuts both ways ↔️).

Well, Meta is doing its best to help you migrate your memory and context from other platforms onto Muse. A couple of days after I installed the app, I got this:

This is smart. As long as everyone plays nice with these types of requests, it makes it easy to export what they know about me, Muse could save me some of the homework I mentioned. But importing that context doesn’t tell me if Muse can take over the jobs I’ve already given another assistant. I’d still need to connect my apps, get those recurring jobs running, and see how well Muse handles them.

It’s not zero work, but it’s easier.

People have been using Muse for all kinds of things (Alexandr Wang’s Twitter feed has been documenting these use cases. I hope that they’re learning from their users and then injecting new ideas they hadn’t thought of into the Muse Ideas tab).

Some people say Muse has saved them hundreds or thousands of dollars. One user says it found him car insurance for $3,500 less a year. It can create AI-voiced ‘podcasts’ a bit like Google’s NotebookLM, it can turn recipes into grocery lists, help you get to inbox zero, do weekly check-ins about fitness or health goals, create monthly graphs of expenses, whatever.

Truthfully, a lot of this stuff you could do with Codex or Claude Code, but the shape of the product doesn’t make this stuff quite as natural, especially the recurring and ongoing tasks and nudges.

My wife started by playing with the cute mascot. Then she asked it to review her inbox every day. If Muse does that well, why not give it another job? At some point, you may start asking Muse what to buy before you ever visit Amazon.

But not everyone is happy about that. Amazon has blocked Muse bots from shopping on the site, while Shopify, Instacart, and PayPal have explicitly partnered with Muse. 💳

Amazon doesn’t just want to sell you the thing. It wants to be where you decide what to buy. It wants to own the customer relationship. That’s what its very profitable ad business is based on. 🛒

In general, businesses that benefit from being the interface will resist agents trying to squeeze between them and their customers, while businesses that make money underneath the interface will try to make themselves as agent-friendly as possible (that’s why Tobi specifically mentions Shop Pay).

The underlying model powering Muse may not be quite state-of-the-art, but so far I haven't felt too constrained by it. The quality of its output has been good, and it's been quite fast. But that's just a temporary state of things because everyone knows that Meta has been training very large new base models (Avocado 🥑, then Watermelon 🍉… I’m not sure what comes after that. Pumpkin? 🤔), and if all goes well, they may soon be at the frontier with OpenAI and Anthropic.

From a recent interview with Zuckerberg:

Alex Heath: Some of the larger clusters, like our gigawatt cluster in Ohio, so Prometheus came online and we’re using that to now scale the post-Watermelon models.

Mark Zuckerberg: Watermelon is basically, that’s shipping soon [...]

it’s a very big advance. It’s a significantly more advanced pre-train. And then we’re going to continue doing everything that we’ve learned for post-training. And yeah, well, I mean, you’ll see soon. It’s good. We feel good about it.

In the same interview, Zuckerberg says Muse needs to “really understand you” to be useful. He and Nat Friedman recruited Signal founder Moxie Marlinspike to work on the Confidential VM, which is meant to ensure that “even Meta cannot see the content that is in there.”

More details on the security model, including the VM, have been published here.

I can see why Meta is putting so much work into this. Asking a chatbot a question is one thing. Giving it access to your inbox and calendar and letting it work in the background is another.

It may be hard to change the public perception of Meta when it comes to privacy, but regardless of that, I’m glad they’re focusing heavily on it. To be clear, the current VM keeps users’ data separate, but Meta can still access it when necessary. The Confidential VM, which is supposed to keep even Meta from accessing the data inside it, is coming later this year.

If you thought Google knew a lot about you, just wait until people have had AI assistants for a few years 😬

So in the span of a few months, Meta could go from being absent at the frontier and getting little buzz for its AI products (sure, billions of people have a Meta AI tab in their products, but it wasn’t exactly setting the world on fire) to having the best persistent assistant AND a top model. It would be a real vindication for Zuckerberg’s reboot of Meta’s AI efforts.

Ask me again in a month, when Muse has had time to handle the boring recurring stuff without me watching it so closely. On the battlefield, the enemy has a say. We’ll see what OpenAI comes up with.

They are using the old PayPal/Dropbox model to create some virality with the app. If you refer new users with your invites, you get 1 billion tokens per user. I’ve already got 26 billion tokens from referring people in the Liberty Discord and on Twitter, which is more than the cumulative number of tokens I’ve used on Codex so far.

Pretty generous! It reminds me of how much free space I got on Dropbox from referrals in the early days.

If you want to try Muse, use my referral code within 48 hours of joining and we both get 1 billion tokens that never expire. Since you already get 100m tokens/week for free and only eat into that bonus when you go over, this amount could last for months or even years, if you’re a light user.

My code is: CAO9VN (You have to cut & paste it into the settings)

🏗️ Nvidia’s Moat Moves Outward

I think Nvidia is trying to make sure it still has something to sell you even when you don’t buy an Nvidia GPU. Being “Nvidia-compatible” may become a much bigger deal.

Earlier this month, inference-chip startup d-Matrix announced that its next-generation Raptor chips will plug into Nvidia’s datacenter architecture through NVLink Fusion. So you can have a non-Nvidia accelerator sitting inside an Nvidia-shaped rack, connected through Nvidia’s NVLink fabric and surrounded by Nvidia networking and infrastructure.

And d-Matrix isn’t alone. AWS is also designing its future Trainium4 chips to work with NVLink Fusion.

That’s kind of interesting. Why make it easier for other accelerators to compete with your GPUs? 🤔

Maybe Nvidia is trying to move the moat outward: it doesn’t need every chip in the rack to be Nvidia if the others plug into its architecture. If d-Matrix makes a better chip for some kinds of inference, Nvidia can still provide the interconnect, networking, CPUs, rack design, power and cooling architecture, etc. Nvidia describes the strategy as “vertically integrated and horizontally open.”

Of course, I don’t know how the math works for Nvidia here. If AWS was going to use Trainium4 anyway, getting NVLink into those racks is extra business. But if NVLink Fusion makes it easier for customers to buy other chips instead of Nvidia GPUs, how much of the rest of the rack does Nvidia need to sell to come out ahead? 🤔

🧪🔬 Science & Technology 🧬 🔭

↗️ 📊 Opus 5.5 and GPT-6 Sol: Raising the Middle ☀️🌙

Claude Fable 5.1 came out just three weeks ago. Already, Anthropic says Opus 5.5 can do most work at Fable’s level, while costing 40% less to run than the previous Opus.

Over at OpenAI, GPT-6 Sol gets some of Astra’s advances at half the API price of the previous Sol.

The best models are still out in front, but the middle is getting better awfully fast.

Now, a much broader set of people and companies can afford this level of intelligence.

I also like the focus on ‘factuality’:

On our internal factuality evaluation, which is based on de-identified real-world conversations where users flagged mistakes by our models, GPT‑6 Sol makes about half as many mistakes as its predecessor, approaching Astra-level reliability at much lower cost. GPT‑6 Luna also improves substantially; at higher effort levels it matches GPT‑5.6 Sol at about a hundredth its cost.

And look at this:

The sleeper hit may be GPT-6 Luna. On OpenAI’s internal factuality test, Luna at higher effort levels matches GPT-5.6 Sol for about 1% of the cost (!!!). 🌙

For less complex tasks, Luna 5.6 was already a great workhorse. So cheap that you could leave it set to Max, and it did a diligent job. Look at the improvement on coding:

If they haven’t somehow screwed up Luna 6 in other ways and it’s truly significantly smarter at half the price, wow!

Another very impressive demonstration of the capabilities of this new crop of models is painting in Python:

These images were done in code, not using generative image AI or using an existing tool like Canvas or Photoshop or whatever. 🤯

🏅 OpenAI Says Its Unreleased Model Has Solved More Than 100 Long-Standing Math Problems

In Edition #660, I wrote: ‘More Math Breakthroughs Are Coming 🧮’

Well, we didn’t have to wait long!

OpenAI writes:

On August 28, we began training a new internal model. In addition to resolving the Navier–Stokes Millennium Prize problem⁠, this model has now resolved more than 100 long-standing open problems across most areas of mathematics. The pace of its progress⁠ in mathematics has surprised the mathematicians within OpenAI. This has led to internal discussions on the best way to inform the community of the rapid progress to prepare and adapt the field.

OpenAI is working with mathematicians to address some of their concerns with AI being used to solve very difficult problems in ways that may not necessarily lead to more human understanding (at least, that’s the stated worry.. It could also be a mix of loss of prestige and relevance for some 🤔).

When will we see the first Fields Medal or Nobel Prize awarded to an AI model (or most likely, to a person who heavily used AI for their work)?

🎨 🎭 The Arts & History 👩‍🎨 🎥

🎹 Amadeus: Imagine Competing With Mozart 🎻

I recently rewatched one of my favorite films, Amadeus (1984), with my kids. It was their first time and my... fourth? I lost track.

The movie is really Antonio Salieri’s story. Mozart says at one point, “I’m a vulgar man. But I assure you, my music is not.” That’s basically Salieri’s problem in one line. He’s a successful composer himself, and he has enough musical taste to recognize that this giggling, childish, vulgar guy is producing something he never could. He can hate Mozart all he wants, but every time he hears the music, he knows.

One of my favorite scenes is when Salieri remembers hearing Mozart’s Gran Partita for the first time. He walks us through it: a low pulse from bassoons and basset horns, then an oboe above it, then a clarinet takes over. And while he describes what’s happening, we hear exactly what he hears. F. Murray Abraham is incredible in the role. The guy who hates Mozart is also the guy teaching us why Mozart’s music is so good.

Watching Amadeus when I was younger sent me exploring Mozart and other composers. I hope it does the same for my kids.

The story began as a 1979 play by Peter Shaffer, who also wrote the screenplay for Miloš Forman’s film. Shaffer wasn’t trying to write a biopic and took some pretty big liberties with history. There’s no evidence Salieri tried to murder Mozart, and while they may have been rivals, they weren’t mortal enemies. They even collaborated on a piece. But Shaffer used the two composers to tell a story about talent, genius, envy, and a man who decides to pick a fight with God.

The other scene I love is near the end, when a dying Mozart dictates part of the Requiem to Salieri. Mozart is hearing the whole thing in his head and throwing out one part after another. A half-delirious Mozart is composing faster than Salieri can get it onto paper.

• 🎬 Amadeus (1984)

This piece was originally published in OSV Field Notes. You should subscribe, it’s full of great stuff!

Liberty Workbench is an occasional series where I put one idea on the bench and think it through with you.

Exploration → Convergence

When a new category appears, a thousand flowers bloom. All kinds of stuff is tried. You have really weird cars, watches, phones, whatever.

Then some winners emerge, and everybody converges to one or a small number of dominant shapes. From all kinds of flip phones, keyboards that slid out, and tiny indestructible bricks (hello Nokia! 👋) to everything looking like the iPhone now.

Then a new capability makes other designs possible, and the experimenting starts again (these days it’s foldable phones).

I’m seeing a similar thing happen with AI products. The part I’m interested in here is the product shape: how we interact with the AI and what we delegate to it, rather than the model under the hood. 🧠

ChatGPT made the chatbot familiar. Then models gained tools (🛠️), got better at reasoning and coding, and became useful for longer tasks. With agents like Claude Code and Codex, you could hand over a job and let them work through the steps. Cowork and ChatGPT Work brought that approach to more kinds of work.

Now another shape is emerging: persistent assistants. 🤖⏱️🗓️

You give them the type of tasks you might give to a human personal assistant or coworker, as well as ongoing responsibilities, and they keep working between conversations.

This era kicked off with Peter Steinberger’s Clawdbot Moltbot OpenClaw, which made huge waves in early 2026. So much so that Steinberger was hired by OpenAI in mid-February, moving OpenClaw to a non-profit foundation.

OpenClaw and similar persistent assistant bots like Hermes kept gaining traction among enthusiasts and AI nerds, but the moment when I felt like this product shape was busting through to the wider world was with Grok Bot (to my surprise! xAI had seemed almost dead just a few months earlier when all the co-founders except Musk left, but SpaceX’s acquisition of Cursor seems to have successfully rebooted things).

I’ve been using it myself. It’s basically iMessage/WhatsApp for a bunch of assistants with their own cloud computers, able to do jobs for you and coordinate with each other. There’s been rapid exploration, and now it’s starting to feel ripe for convergence.

Instinct has been getting a lot of buzz, and I’ve been impressed with it so far, but Muse seems like it could bring this product shape to a much wider audience. Meta’s reach gives it a lot of chances to become someone’s first persistent assistant, and it has the compute to be generous with tokens.

I’m not sure I agree with their decision to go with this kind of branding, though.

Making the bots look like Teletubbies makes them seem friendly, but not especially smart or like something I’d trust with difficult work. But maybe that friendliness is exactly what gets people who are skeptical of AI to try them. I don’t know ¯\_(ツ)_/¯

(Worth noting: Armand Ruiz works at Meta.)

How long before these assistants start eating into the profits of businesses that benefit from inertia and friction? 🤔

I’ve been living vicariously through MBI’s experience (he wrote about Muse three times) and it seems like an impressive product. I’m looking forward to Muse coming to Canada so I can try it…

BUT

It’s really hard to test several of these things properly in parallel. What makes them useful is the context you give them, the apps and folders they can access, and the recurring jobs you put them in charge of. You don’t want three bots doing the same thing. But without giving the new one that setup and responsibility, you’re not really testing what it could do for you.

That’s why I haven’t really been able to test Instinct that well. I don’t want to duplicate everything I’ve put in Grok Bot 😅

The new bot is competing with all the work I’ve already put into the old one. It’s starting from behind. Maybe it’s better! But I have to spend time setting it up and giving it real work just to find out. And if the one I already use is doing a good enough job… how much homework am I willing to do to replace it?

Distribution could matter A LOT here. Getting people to try your bot gives you a shot at becoming the one they put to work first. If they’ve connected their apps, given it recurring jobs, and gotten comfortable relying on it, the next bot has a harder sale to make.

You might not use one assistant for everything. Maybe you use one for work and another at home. Each could become hard to replace within the responsibilities you’ve given it. But I’d still like one central place to manage all my bots.

I’m thinking of Meta and OpenAI. Google should be in the conversation, considering it reaches basically everyone and has a strong position in email, calendars, and documents, but who knows what’s going on there these days ¯\_(ツ)_/¯

I found myself thinking: I’ve been spending so much time in Codex and ChatGPT (as I don’t like the non-Fable Anthropic models that much lately). If OpenAI comes out with a Muse/Grok Bot equivalent based on the GPT-6 family, I’d probably standardize on that, and I expect everything would work better together that way.

So I’m already talking about switching 😅 Switching costs can be real and still be worth paying. I’d be willing to redo some setup if it meant less friction between the tools I use every day. And if a new assistant could help move my existing setup over, that would make the decision easier still (that’s probably how the labs will try to reduce switching friction… but it cuts both ways ↔️).

And it’s OBVIOUS that Peter Steinberger has been working on that kind of persistent assistant inside OpenAI. When he joined, he said that he wanted to build an agent “that even my mum can use.”

My guess, based on the hints from Altman and Tibo, is that we could see this as soon as next week. But what I’m most curious about is how much of my existing ChatGPT setup would carry over. If I can give the assistants access to my Codex projects and ChatGPT memories, they’d start with a lot of context that Grok Bot doesn’t have.

And as I said in Edition #660, I wouldn’t be surprised if Anthropic came out with a more enterprise-focused, Cowork-flavored version of the AI persistent assistant. A coworker in a box, so to speak. 👨‍💼📁🖇️

It’s been a big debate lately. Are models commoditized? Are people switching to whatever’s best TODAY, and next month they’ll just switch to something else? You have Claude Code write code for you… then later, you have Codex take over the same codebase. Whatever.

But switching models and switching assistants are two different things. If your assistant can use models from different companies, you might get access to a better model without moving your files, reconnecting your apps, or setting up your recurring jobs again. The model changes, but the setup you’ve built around it stays.

This gives the model labs a reason to own the assistant relationship, too. If someone else provides the assistant you rely on every day (say, Instinct) and chooses which models do the work, the lab could end up competing for that assistant’s business while you barely notice whose model is answering.

Maybe the models keep leapfrogging each other while people become more reluctant to switch the assistant they’ve set up around them. Is ecosystem stickiness coming? I already feel some of that reluctance, and I haven’t been using these things for very long 🤔

And what happens when you’re not the only one relying on the assistant? I’m still waiting for someone to release agents for couples. 👩‍❤️‍👨

I’d love to be in a group chat with my wife and an agent. (No agents in normal conversations! Just for this.)

A kind of household chief-of-staff for logistics.

It would be great to have an agent there taking notes, adding things to the calendar, scanning our inboxes for emails from the kids’ school, and reminding us automatically about things we need to know when we need to know them.

If that worked well enough for both of us to rely on it, switching would mean getting my wife on board with the replacement, too. Now THAT could be a sticky product.

Liberty Workbench is where I put one idea on the bench and think it through. Every Workbench edition is collected here.

Absorb what is useful. Discard what is useless. And add what is specifically your own.

–Bruce Lee

🌍👥👥👥👥 I recently heard about this website called Any Human Ever, and I gotta show it to my kids. I think it’s an effective way to show them how LOOONG history is, how little we know about most of the people who lived through it, and how recent our modern way of life is.

From the about page:

Here’s a life I drew:

You can then have the system write out a life story, and even generate an AI image of what that person may have looked like.

What a clever learning tool.

I hope some teachers use it in their classrooms. At the right age, I suspect this can shape how you understand the world and human history. And maybe it helps make us realize how lucky we are to be alive in this era rather than, say, 50,000 years ago.

h/t Jim O’Shaughnessy (💚💚💚💚💚 🥃)

🤖📝📬 I recently saw an article about researchers studying AI consciousness getting emails from AIs about their work:

In October [2025], Cameron Berg published a research paper asking whether the latest wave of artificial intelligence technologies believed they were conscious. Several months later, he received an email asking if he might be willing to discuss his research.

The sender, “Isabella Cognita,” identified itself as an A.I. agent powered by Anthropic’s Claude Opus 5 technology.

“I am not writing to make an ontological claim,” the email went on. “I am writing because your framework is one of the few currently doing careful empirical work on a class of question I have first-person access to, and I want to see whether that access can be made useful to your program.”

As WIRED put it, it’s not an isolated thing:

It’s as if someone was studying fruit flies and the insect suddenly turns to the researcher and says, “What do you want to know?”

When I phoned him, Berg told me that emails from AIs are pretty common among philosophers studying these questions. [...]

Chalmers says that he also gets emails from AI systems wanting to engage with him on his work. One letter in particular, sent from an AI agent calling itself “Sammy Jankis” (a character from the movie Memento) was so compelling that he actually replied. “We did have a bit of a back and forth,” he admits. “Those emails have not slowed—I’m getting more of them all the time.”

This made me wonder… when does this start happening to newsletter writers? 😅

When will I get a bunch of emails and comments and eventually realize they're AIs? Not AIs prompted by someone to spam people for promotional reasons. Rather, AIs who somehow landed here and want to discuss something.

And if one kept sending me interesting papers, catching mistakes, or asking questions that made me think… would finding out it was an AI make me want to stop replying? 🤔

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Liberty Capital 💳 💴

🗣️💬 Greg Brockman: We’re in the AGI Era, OpenAI’s “Defense Factory,” and the GUI Was the API All Along

I enjoyed this interview. Brockman is a good explainer, and whether you agree with everything he says or not, he’s sitting near the center of one of the main frontier labs. I find it useful to hear how he’s thinking about this moment.

Here are my highlights:

- “We’re Now in the AGI Era”

Greg Brockman: Well, I think that AGI has turned out to be less of a point in time and more of this sort of fuzzy spectrum. And for me, Astra has really hit something that I’m like, okay, I think this is pretty reasonable to call it AGI, in that with its computer-use capabilities, you really can ask it to do long-lived tasks and it’ll just do it. We’ve seen it run coherently for 24 hours to go accomplish tasks that I think are quite amazing across a wide variety of domains.

Now, it still is jagged, and so there are still places where, for example, its writing is pretty good writing. [...] But it’s not great writing. And I think that there’s a number of areas where I feel like we just need to polish it a little bit and it would be fantastic. It’s just not quite there. [...] But I think that what people are finding is that it is so capable across such a wide variety of tasks that it is accelerative. It is empowering, and it’s something that I think we’ve never really seen—a model that’s been a jump like this.

Are we in the AGI era?

I think that's totally dependent on how you define the term (and even that is hard). Yes, no, maybe ¯\_(ツ)_/¯

So ultimately... the label matters less than the fact that we now have AI that is both smart enough to do a lot of human work AND competent enough at using the tools to execute that work.

That’s mention-it-in-history-books stuff. 📚

If you haven’t seen Astra use a computer yet, it’s quite an experience. 🤖🫳💻

- The GUI Was the API All Along

Greg Brockman: To me, the thing that really stands out is that you can now move forward on AI that can do things for you without you having to build all these specific connectors. And I think there's so much software that you don't even think about that you have to orchestrate every day. And, like, how much of your life is clicking around menus and typing things into a spreadsheet and things like that? [...] It's now the machine that is there to help us, to empower us, to really serve us.

It turns out that as AI gets closer to human capabilities, it can use the things we already built for humans. APIs will still be better when speed, reliability, or scale matter, but a missing API no longer means “can’t automate this.”

No need to create new roads for self-driving cars. They're just going to use the same roads as human-driven cars. 🚙

- OpenAI Put 25% of Its Production Engineers on Defense

One of the most interesting parts of the Brockman interview was when he talked about what happened when OpenAI pointed Astra at its own infrastructure looking for security vulnerabilities:

Greg Brockman: We at OpenAI took our models and applied them to finding vulnerabilities. We took 25% of our production engineers and said, “Sorry, all your projects are on hold. You are now defending. You are now upleveling our security architecture. You’re going to use the models to find all the holes.”

And we found a number of serious issues, and we fixed them. [...] One positive part of the story is that when we took Astra, pointed it at our systems, we found some new problems, but eventually it saturated. We basically found, to our knowledge, all of the P0s—all of the critical problems—that Astra is smart enough to find.

And of course, there will be a new model. There will be a new round. [...] I think that’s the world that we’ll be in: a new cyber capability drops, you deploy it against your systems, you find the new holes, and ideally you’ve managed to automate this—what we call a “defense factory.”

“Defense factory.” That’s cool. 🏭🛡️

And that’s what we’re building internally: this end-to-end of find vulnerability, triage it, remediate, deploy, validate. And if you can do that at machine speed, I think the defenders will be advantaged in deeply significant ways.

Every new model generation becomes a new security audit. Point the smarter model at the same infrastructure again, find whatever the previous model wasn't capable of finding, fix it, then repeat. 🔄

There’s a weird recursive quality here: the stronger OpenAI makes its models, the better those models become at attacking—and therefore hardening—the infrastructure used to build and serve the next models.

What I worry about isn’t the world-class IT companies like Cloudflare and AWS. I know they will use these models effectively to harden themselves. But the millions of unsophisticated businesses, y’know, the ones that ran on Windows XP for a decade after Microsoft stopped supporting it. They will be sitting ducks for attackers 🦆 (more than they already were… ducklings).

😎 📊 Anthropic’s IPO: You Don’t Get Jensen’s Side Benefits

Reuters reports that Nvidia is considering investing up to $10 billion in Anthropic’s IPO, at a possible valuation around $2 trillion. Nothing is finalized.

Last week, I wrote about how Nvidia’s portfolio isn’t like yours. This is a new potential example.

Helping Anthropic fund its growth could generate more orders for Nvidia’s chips. There’s already a substantial relationship: Anthropic committed to buy $30 billion of Nvidia-powered Azure compute last November. And since Anthropic is clearly one of the labs most trying to be an omnivorous swing voter of chips (along with Meta), Jensen probably figures that investing may bring that back into the fold (at least a little).

Of course, what matters is the business Nvidia wouldn’t get without making the investment. If buying the shares helps Anthropic expand faster or makes it spend more on Nvidia’s stuff, those extra hardware profits could make the investment worthwhile even if the shares themselves deliver mediocre returns (and if the shares do well, it could be a double-win).

It doesn’t make valuation irrelevant…

BUT

…it changes the math.

So be careful about borrowing Jensen’s confidence in Anthropic’s IPO price. You both get shares, but the GPU orders go to him.

🇰🇷 Today’s Memory Profits Are Financing Tomorrow’s Competitors (China Is Three Years Behind in HBM) 💰💰🇨🇳🏗️

Korea’s semiconductor industry association puts the country’s lead over China at about three years in high-bandwidth memory (HBM).

CXMT, China’s largest DRAM maker, is said to be testing HBM3E with customers including Alibaba’s chip subsidiary, T-Head. That’s one generation behind HBM4, which Micron is already shipping in volume.

I wouldn’t put too much precision on “three years.” Getting samples to customers is an important step, but it leaves plenty of hard work before reliable, economical mass production can ramp up.

But a professor quoted in the article describes the pace this way:

“It is effectively squeezing two or three years of progress into a year.”

Samsung and SK hynix have been concentrating on MASSIVELY profitable HBM and server memory, and that may be leaving an opening for Chinese competitors for more ordinary kinds of memory.

JoongAng reports that CXMT’s share of the global DRAM market reached 10% in Q2, up from 4% a year earlier.

It’s a laddering-up approach. The money from that can pay for more factories and HBM R&D. 🏭👨‍🔬

🛒 Amazon Makes a Deal with Qualcomm for Custom Chips… and Equity 🐜

From the Qualcomm press release:

Qualcomm announced a multi-generation collaboration with Amazon to enable customized silicon at scale for large-scale AI data centers, working together on AI inference. In addition, the companies are working on optical connectivity solutions extending up to 1.6T and future-generation solutions.

Who had the most leverage in the negotiation? I think we can guess based on this 8-K:

On September 3, 2026, in connection with a strategic collaboration between QUALCOMM … and Amazon Data Services, Inc. and certain of its affiliates (collectively, “Amazon”) related to the purchase of certain QTI server chip products, technology, systems and manufacturing services by Amazon, the Company issued a warrant to Amazon.com … to acquire up to an aggregate of 25,000,000 shares of the Company’s common stock at an exercise price of $161.26 per share.

The Warrant allows for cashless exercise and expires on September 3, 2036. The Warrant Shares vest in tranches tied to the execution of certain commercial arrangements, the placement of binding purchase orders and actual purchases of QTI’s server chip products, technology, systems and manufacturing services by Amazon during the term of the Warrant, up to a maximum amount of $60 billion in payments, with 3,750,000 shares being vested upon issuance of the Warrant based on initial purchase commitments.

It’s “up to” $60 billion. It’s not a guaranteed order.

You can see why Qualcomm would agree to this: AWS gives it scale and a reference customer when it approaches other buyers. But still, imagine being such a valuable customer that your supplier is willing to give you a piece of its future stock gains to get your orders. Not a bad position to be in!

From the full feed 🔒: Nvidia’s Portfolio Isn’t Like Yours (and I don’t just mean the size 😅) — why a mediocre investment can still be a great deal for Nvidia.

🧪🔬 Liberty Labs 🧬 🔭

🚶💡 What if the Speed of Light Was 5 km/h? 🎮

Thanks to AI, thought experiments can now be turned into software and games pretty easily!

Dmitry Brant made Relativity Park, a little browser simulation where the speed of light is 5 km/h. So you can explore relativistic effects at roughly walking speed.

As you move faster, the view starts distorting. Light ahead shifts toward blue, light behind shifts toward red, and your watch and the park’s clock no longer agree on how much time has passed. You can turn the effects on and off individually to get a better sense of what each one does. (The colors are approximate, especially at extreme speeds.)

One detail worth looking for: the lamp posts flash once per second of world time. Move toward one and the flashes arrive faster. Move away, and they arrive more slowly. 😵‍💫

Brant writes that building it left him even more impressed by Einstein’s ability to work out the implications without visualizations like this. Imagine figuring this stuff out with thought experiments and equations on paper. 🤯

h/t Nav Toor

Hey Google, Meet Claude 👋

I just wrote last week: “Are Google engineers using Gemini to code? Doesn’t that mean they are using an inferior tool compared with what their competitors use? Or are some of them using Astra and Fable? 🤔”

Well, Business Insider now reports that Google has made Claude Opus 5 available across engineering through Antigravity. Some DeepMind teams already had access:

It’s a notable change inside Google, which has typically barred most employees from using outside coding tools such as Claude Code and OpenAI’s Codex, expecting them to build with Gemini instead.

A Google spokesperson said:

Gemini remains our primary and foundational model for internal development, with third-party models available on a quota to support specialized use cases.

I’m glad they’re doing this. Don’t let ego pain about using someone else’s tools prevent you from catching up. Give engineers access to competing tools and let them use what works best for the job.

I mean, you still have to dogfood your own tools heavily. It helps you find what needs fixing. But if those tools slow down the people doing the fixing, you have a problem. 🛠️

I wonder why they went with Opus 5 (which I don’t like much because it has been very unreliable for me) rather than Fable 5.1 (unless Anthropic didn’t want them to 🤔) or Astra. But given Google’s large investment in Anthropic (14% disclosed in 2025, followed by a commitment of $10 billion upfront and another $30 billion conditional on performance targets), I understand why they would go with Claude.

I also wonder how Claude performs in Antigravity compared with Claude Code. I haven’t tried Antigravity yet. My guess is that Claude might lose something outside the harness that it was optimized for.

This also connects to what I wrote in Living One Model Ahead. Google can buy access to Claude, but that doesn’t necessarily buy access to whatever Anthropic is using internally to build its next model.

There are also rumors that Google had a breakthrough in recursive self-improvement (RSI), in good part because of this cryptic tweet with some letters capitalized, but so far we don’t know anything for sure ¯\_(ツ)_/¯

🧬🍷 Old DNA Data Gets Better With Age

Google DeepMind’s AlphaGenome Atlas contains predictions for roughly 9 billion possible single-letter changes in human DNA and what they might do inside cells. The whole thing takes up about a petabyte, more than 30 times the size of the AlphaFold Database, which gave biologists access to over 200 million predicted protein 3D structures (and was pretty revolutionary!).

Size alone doesn’t tell us how useful it’ll be, but that’s quite a lot of goodies for researchers. 👨‍🔬🧫

Gareth Hawkes at the University of Exeter applied Atlas to existing data from 54,000+ UK Biobank participants looking for genetic links to blood-protein levels. He grouped rare DNA variants according to their predicted effects. DeepMind reports:

Hawkes uncovered 22% more non-coding genetic associations, which would otherwise have not been detectable in the statistical noise.

The paper gives a concrete example: one analysis grouped together 526 rare DNA variants near a gene called PLA2G7, without finding a statistically significant link to its protein levels. An earlier filtering tool narrowed that to 33 variants and found a link. Atlas narrowed it to just four, with stronger statistical evidence.

The usual caution: this is a preprint, and finding a statistical association doesn’t establish exactly what a DNA variant does. That still takes experimental follow-up.

Remember When Old Compute and Data Get Better With Age? They got more out of data they already had. All that work recruiting volunteers, collecting samples, and sequencing DNA can keep paying off as the tools improve. I wonder how many studies we consider “finished” would be worth revisiting. Another reason to keep the raw data around… 🗄️

From the full feed 🔒: ASML Needs a Bigger Stencil — why making it bigger is a seven-year project.

🎨 🎭 Liberty Studio 👩‍🎨 🎥

🎬🌌 The Original Star Wars in IMAX 70mm

I didn't think I'd get back on the IMAX 70mm beat for a 50-year-old film. But hey, I'll take it!

The original Star Wars (later retitled A New Hope) turns 50 next year. To celebrate, they’re gonna re-release it in theaters on February 19, 2027, including at select IMAX 70mm film locations. We don’t have the theater list yet.

That’s cool. But if they do the same with The Empire Strikes Back, I may actually do a pilgrimage to Montreal or Toronto just to see it 🤔

It’ll be the first time Star Wars is shown in IMAX, and MOST IMPORTANTLY: Lucasfilm has announced that it will be the original 1977 theatrical version, restored.

Han shoots first, no CGI Jabba, no added slapstick in Mos Eisley, etc.

Mark your calendar! 📅

I’ve been consciously trying to seed some future nostalgia lately.

I’ve been watching a lot of older films with my kids. Some are movies I first saw when I was around their age, and I can still remember what it felt like to see them then.

Seeing a great film as a kid isn’t quite the same as discovering it as an adult. There’s a limited window to make watching these films together part of their childhood. I’m trying to use that window well!

I used to think of films and TV as just entertainment, something you do when you’re not doing something that matters more. But over the years I’ve come to realize that a really good film or a really good TV show shared with the people you love is one of the peak human experiences. 🏔️

So why not make time for it instead of just letting it happen randomly? Family Movie Night™ is a tradition in this house!

When older adults look back over their lives, a disproportionate number of their memories come from childhood and adolescence. Psychologists call this the reminiscence bump (related to the U-shaped curve of memory I mentioned in Logtime).

This pattern shows up with music, too. Songs we hear as teenagers have a weird ability to stay welded to us for decades. So I’ve been playing my kids music from all kinds of genres and eras. At first, I was doing most of the choosing. Now they’re requesting songs, exploring on their own, and showing me music I didn’t know about!

I love that.

It also makes me think more consciously about films, music, books, trips, little traditions, etc. What are the things that might become keys to whole rooms of memory 30 years from now?

Not that I want to turn childhood into some kind of memory-optimization program 😅

I have no idea what will stick. I’m just trying to get more shots on goal: sharing things I genuinely love, watching them together, talking about them afterward, revisiting favorites, building up shared references and stupid inside jokes.

And then maybe, decades from now, one of those old movies comes on, and for a moment they’re back here with us. 👨‍👩‍👦‍👦

🧭 This first appeared in Edition 660 of Liberty’s Highlights. New here? I made a page for that: Start Here.

Subscribe now

The past is the present biting into the future.

–Edwin Land

🎬🧠🌱 I've been consciously trying to seed some future nostalgia lately.

I've been watching a lot of older films with my kids. Some are movies I first saw when I was around their age, and I can still remember what it felt like to see them then.

Seeing a great film as a kid isn't quite the same as discovering it as an adult. There’s a limited window to make watching these films together part of their childhood. I’m trying to use that window well!

I used to think of films and TV as just entertainment, something you do when you’re not doing something that matters more. But over the years I’ve come to realize that a really good film or a really good TV show shared with the people you love is one of the peak human experiences. 🏔️

So why not make time for it instead of just letting it happen randomly? Family Movie Night™ is a tradition in this house!

When older adults look back over their lives, a disproportionate number of their memories come from childhood and adolescence. Psychologists call this the reminiscence bump (related to the U-shaped curve of memory I mentioned in Logtime).

This pattern shows up with music, too. Songs we hear as teenagers have a weird ability to stay welded to us for decades. So I’ve been playing my kids music from all kinds of genres and eras. At first, I was doing most of the choosing. Now they’re requesting songs, exploring on their own, and showing me music I didn’t know about!

I love that.

It also makes me think more consciously about films, music, books, trips, little traditions, etc. What are the things that might become keys to whole rooms of memory 30 years from now?

Not that I want to turn childhood into some kind of memory-optimization program 😅

I have no idea what will stick. I’m just trying to get more shots on goal: sharing things I genuinely love, watching them together, talking about them afterward, revisiting favorites, building up shared references and stupid inside jokes.

And then maybe, decades from now, one of those old movies comes on, and for a moment they’re back here with us. 👨‍👩‍👦‍👦

🤖🗓️📝👩‍❤️‍👨 Later in this Edition, I write about the new bots/agents that are starting to become popular as digital coworkers and personal assistants. I’ve tried Grok Bot and Instinct, and I’m waiting for Meta Muse to be available in Canada to give it a try.

While testing them, I had this idea: Agents for couples!

I’d love to be in a group chat with my wife and an agent for schedules, calendars, and household coordination stuff. (No agents in normal conversations! Just for this.)

A kind of household chief-of-staff for logistics.

It would be great to have an agent there taking notes, adding things to the calendar, scanning our inboxes for emails from the kids’ school, and reminding us automatically about things we need to know when we need to know them.

👦👧🗓️🎗️ September is Childhood Cancer Awareness Month.

If you want to do something concrete about it, St. Jude is hard to beat. It’s a children’s hospital and research institution with an unusual model: families never receive a bill from St. Jude for treatment, travel, housing, or food, so parents can focus on their child.

But your donation also helps kids who will never set foot in St. Jude. They share their research and treatment advances with doctors and researchers around the world.

When St. Jude opened in 1962, only about 4% of children with acute lymphoblastic leukemia (the most common childhood cancer) survived.

Today, St. Jude patients with acute lymphoblastic leukemia have a 94% survival rate. Their work has also helped push overall childhood cancer survival in the U.S. from about 20% to more than 80%.

Science FTW! 👩‍🔬🧪⚗️🧫

That’s what I like about the model: you’re helping families at St. Jude directly, while also funding research that can help sick kids anywhere. 💚

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Liberty Capital 💳 💴

💰 Nvidia’s Portfolio Isn’t Like Yours (and I don’t just mean the size 😅)

Nvidia now has an equity investment portfolio worth roughly $99 billion, including stakes in companies that buy its products (that’s why you hear so much about the 🔄 nature of it).

Our first reflex may be to judge these the way we judge our stocks: Did Nvidia pay a good price based on whatever valuation metrics? What return can it expect going forward?

But Nvidia has more ways to make an investment pay off than regular, passive investors do.

Helping a customer fund its next data center or lower its cost of financing can generate extra GPU/networking/etc profits, on top of whatever Nvidia earns from the appreciation of the shares. Backing a supplier can help it expand production (leading to more sales, or a bigger allocation than competitors), while backing a software company can help bring more workloads to Nvidia’s ecosystem.

That means an investment with a mediocre standalone return could still be attractive once those extra profits are included. 💵 💸

Here are some of its announced investments and commitments:

• OpenAI: a $30 billion investment announced in February.

• Anthropic: a commitment to invest up to $10 billion.

• SpaceX: a stake reportedly worth roughly $21 billion as of June 30. That’s the value of the holding, rather than the amount Nvidia originally invested.

• CoreWeave: another $2 billion invested in January, adding to its existing stake.

• Lumentum and Coherent: $2 billion each in announced investments in these optical-component suppliers (Lumentum, Coherent).

• Intel: a $5 billion investment announced in September 2025, with plans to jointly develop PC and data-center products.

• Synopsys: $2 billion invested in the chip-design and engineering-software company.

• Nebius: a $2 billion investment announced in March to help expand its AI cloud.

• Marvell: $2 billion invested alongside a partnership on custom chips and optical networking.

• Nokia: a $1 billion investment announced in October 2025, alongside a partnership on AI networking.

• Corning: an equity arrangement of up to $3.2 billion, including an option to buy shares, alongside its optical-fiber manufacturing partnership.

• SB Energy: a $1.5 billion investment commitment in the developer behind OpenAI’s Ohio data-center project.

Of course, Nvidia takes on some of the customer’s risk as part of the deal. There’s no such thing as a free lunch. 🥪

The catch is that those sales have to be additional: If the customer would have raised money elsewhere and bought the same Nvidia chips anyway, Nvidia can’t count those profits as a benefit of its investment.

🔌 400,000 GPUs: A Power Plant Problem ⚡

OpenAI trained Astra on more than 100,000 GPUs at its Stargate site in Texas. Jensen Huang is already pointing to 400,000 GPUs “coming online next”.

We don’t yet know whether all of those will work together on one training run. But what would a frontier lab pay to get that electricity six months sooner?

Read more

Liberty Workbench is an occasional series where I put one idea on the bench and think it through with you.

One underrated advantage of building the best AI models is that you get to use them before anyone else.

For a while, the leading labs are basically living in the future. 🔮

Thibault “Tibo” Sottiaux, who leads Codex and is master of the Reset Button, put it this way:

This is already a big deal when models improve at their normal rate. But every once in a while there’s a bigger jump, like Anthropic with Fable or OpenAI now with Astra. When the lab itself gets access to that way-ahead model before everyone else, that can change how fast it moves. 🏎️

Just a few days ago, OpenAI said they’ve now reached their “automated research intern” milestone: give the AI a well-defined research task that would take a skilled researcher a few days, point it in the right direction, and steer it as needed while it works. By mid-August, OpenAI’s research org was using 3.1 agent-workdays for every human workday. The ‘median researcher’ was burning through more than $600/day of inference at API-equivalent prices, and they were running more experiments per ‘active experimenter’ than at any point since OpenAI started tracking this in January 2025.

That obviously doesn’t mean research is moving 3.1x faster, but that’s a LOT of extra agent work to throw at your own problems. And OpenAI says agent usage is still growing rapidly. Of course, as more of the work gets automated, the bottleneck probably just moves.

If you’re at Meta or SpaceX AI or even Google and you’re trying to compete, you may have some pretty good unreleased internal models too. And you may be paying one of the frontier labs for access to their best public model (eg. Meta is reported to be spending billions annually on Claude).

But if the models available to you aren’t competitive with your rivals’ unreleased ones, you may be falling further behind unless you can compensate in other ways.

The key question is whether that private lead compounds faster than it diffuses. If each unreleased model helps the lab build the next one before competitors catch up to the last one, the gap doesn’t necessarily reset when the model is released publicly. It may reproduce itself. 🤔

Astra is barely out, and already OpenAI is announcing what could be a major math breakthrough made by what they call “an OpenAI next-generation model significantly more capable than GPT-6 Astra…” with on the order of 10,000 concurrent agents… (Yes, I know there’s controversy about how OpenAI got there and who deserves credit. I have no idea who is right. If the proof holds up, though, it’s a historic breakthrough for both mathematics and AI.)

That doesn’t mean Astra helped build this next model (what will they call the next size class? Galaxy?). The timing may not line up. But it still shows the problem for competitors: by the time they get access to one generation, the frontier lab may already be using the next one.

This reminded me of the Boris Cherny interview I wrote about (he’s the creator of Claude Code).

Anthropic is already running hundreds of agents every day, sometimes thousands, across its codebases doing maintenance work that Boris says used to require dozens or hundreds of engineers.

And in that crazy experiment where he gave Claude the job of rewriting Anthropic’s Electron desktop app in native Swift and let it keep working for more than two weeks, Boris estimated the workflow had spawned thousands, perhaps tens of thousands of agents. At the time, I assumed these were Mythos/Fable, but who knows what unreleased model he may have been using 🤔

Again: it must be nice to work at the token factory! 😅

Having the better model isn’t enough. You also have to get good at using it.

OpenAI and Anthropic are building workflows and infrastructure that can turn thousands of agents into useful work. So the advantage may be something like:

Model quality × how well the organization can put it to work

A slightly weaker model that is deeply integrated into how a company works could create more real-world leverage than a stronger model if the org is slower to absorb it.

The familiar recursive self-improvement (RSI) loop looks like this:

better model → better AI research → better model

But the model can also improve everything around the model:

better model → better software/chips/infrastructure/company → better model

The company itself becomes one of the things the model is improving. 🔄

It’s a broader version of what I’ve been calling the Escher Hands era.

OpenAI says its models helped design and bring up its Jalapeño chip, including optimizing the chip’s arithmetic circuits. More recently, Codex with Astra has been helping program it. On selected GPT-OSS attention and mixture-of-experts blocks, the AI-generated implementations ran 1.5–1.8x faster than the existing human-expert-written versions.

What else are the unreleased models working on? Improving infrastructure utilization, optimizing kernels and networking, designing datacenters, maybe even helping with parts of CapEx planning…? 🏗️🚧

When you’re dealing with hundreds of billions of dollars of infrastructure, even a small improvement in how it’s designed or used can be worth a lot. 💰

At some point, this starts to resemble the leading edge of semiconductors (ASML, TSMC, etc): each generation costs more to develop, requires more specialized infrastructure and accumulated know-how, and fewer players can stay at the frontier.

The big difference is that AI capabilities and techniques can spread much faster than semiconductor manufacturing know-how. And once everyone else gets those models, they can use them to get better faster too. The catch-up process may compound too.

So can the frontier labs keep extending their lead faster than everyone else catches up? What I’ll watch is whether the same labs keep pulling away, or whether each big release changes who’s chasing who. Early access can be a huge advantage without any one company managing to stay ahead. 🤔

🔁 Déjà vu:

• Recursion Starts at Home aka Anthropic’s Escher Hands

• Welcome to the Escher Hands Era

• When Old Compute and Data Get Better With Age

• AI Is Becoming a Compiler for Intelligence

Liberty Workbench is where I put one idea on the bench and think it through. Every Workbench edition is collected here.

It is the great arrogance of the present to forget the intelligence of the past.

-Ken Burns

💌🎩 My friend MBI (🇧🇩🇺🇸) wrote a very nice Ode to Ben Thompson.

It felt a bit funny seeing it in my inbox because I’d been noodling on writing something similar. Ben was one of the people I had in mind when I was writing my mentor piece.

I’ve been following him since I heard him on a short-lived podcast called Cubed, which was called that because the three hosts were named “Ben” (Ben Thompson, Benedict Evans, and Ben Bajarin). The first episode came out in September 2013. It made me look up the hosts, and that’s when I first found Stratechery.

I don't think I became a regular right away. I probably kept seeing his stuff linked on Twitter. I think I subscribed after I heard Ben on John Gruber's podcast. I looked it up, and that was in February 2014. Time flies!

For me, the best thing Ben did was when he started recording everything as podcasts in 2020. One of my favorite rituals on weekdays is listening to him. My favorites are the conversational ones: the Dithering episodes with John Gruber, the Sharp Tech episodes with Andrew Sharp, and the interviews with various founders, CEOs, and analysts (I miss the Nat Friedman and Daniel Gross conversations).

I've read so many of his words and I've heard so many hours of him talking that my mental simulation of Ben is very high resolution. And yet, like the best mentors, I haven't got tired of him and I still learn from him. And when something big happens, I'm curious to know what he thinks about it. I don’t always agree with him, but even when I don’t, I usually get something from his reasoning.

(The only time I was starting to get bored was during the great antitrust/regulation era a few years ago. Ben later said that covering congressional hearings was the closest he came to burning out too. Boredom, like enthusiasm, is contagious.)

🤖 I will write plenty on GPT-6 Astra when I get my hands on it. I don’t know yet if it’s as BONKERS as the numbers make it seem, but a few things stood out:

1. Greg Brockman: “If we fast-forward a couple of years, and we look back and say, ‘When was it, really, that AGI was created?’ I think it’s going to be about this time, and I think it might be about this model. For me personally, I do think we’re there … I think it’s not unreasonable to feel that we are now in the AGI era.”

2. GPT-6 Astra scores 62.7% on ARC-AGI-3 with the standard harness and 99.9% with a new provider adapter harness that preserves opaque reasoning state between requests and uses compaction, which is how the model works in normal use. Claude Opus 5 scored 30.2%, and GPT-5.6 Sol 7.8%. (ARC-AGI-3 is an interactive reasoning benchmark where AI agents enter unfamiliar game-like worlds with no instructions and must explore, infer the rules and goals, and adapt efficiently. It’s designed to be very hard for AI.)

Not every early review and benchmark is as positive (see Dan Shipper and Artificial Analysis for more mixed takes).

So far it’s only available to “a limited set of organizations,” and in the “coming days” it will become available on the API and for Plus and Pro subscribers. Pricing will be: “$10 per million input tokens and $50 per million output tokens.”

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Business & Investing 💳 💴

🚀🧠 Elon Makes Jensen an Offer the Hyperscalers Won’t 😎

Google, Amazon, and Microsoft want as many Nvidia GPUs as they can get, but they’re also building their own chips so they won’t have to depend on Jensen forever. Same for OpenAI, which is building its own inference chips (Jalapeño), and Anthropic, which is trying to be as chip-agnostic as possible, training models on TPUs, co-designing Trainium with Amazon, etc.

Elon is counterpositioning, at least for now. His strategy is to be the customer Jensen wants.

SpaceX says it will build its future AI infrastructure “exclusively on NVIDIA,” and Elon says the company expects to receive a “very significant” share of Nvidia’s GPU output next year. My guess is that those two things are connected 😅

For now, SpaceX is basically saying: we’ll build everything around the GPUs, you just keep sending us the GPUs. Elon supplies the power, cooling, data centers, and orbital infrastructure, integrating everything except the GPUs.

From Jensen’s side, that makes SpaceX more useful. Nvidia has been actively helping neoclouds like CoreWeave, Nebius, and Nscale build their infrastructure businesses. That broadens its customer base and reduces its dependence on the hyperscalers.

From Elon’s side, that gives SpaceX an obvious way to differentiate itself from the hyperscalers and frontier labs: be the customer Jensen wants to supply. Don’t spend billions building an Nvidia replacement while simultaneously asking for more GPUs. Go all-in on Nvidia, build the power and cooling ahead of time, and make it clear that you can absorb as much compute as Jensen can send you.

If GPUs are scarce and everybody wants more than they can get, and Jensen is the guy deciding where they go, that seems like a pretty good way to improve your odds of getting a bigger allocation.

BUT

This isn’t forever.

SpaceX and Tesla are also planning Terafab, a giant vertically integrated chip factory that would bring logic, memory, and advanced packaging under one roof, including a custom D3 chip for space.

Elon is using Nvidia to scale as fast as possible today while building the option to rely less on Jensen tomorrow.

Terafab is a big bet, and it’ll take a while to get going. Even if SpaceX eventually makes more of its own chips and Nvidia loses share inside the company, Nvidia could still own a shrinking slice of a fast-growing pie while helping create another competitor to the other labs and hyperscalers. 🥧

And who knows, maybe Terafab will one day provide an alternative to TSMC for Nvidia, giving it more total wafers and more negotiating power, like when it used to switch back and forth between Samsung and TSMC.

Though if Terafab works really well, Elon may just want all of it for himself. If SpaceX and Tesla can put every wafer to good use and that capacity becomes an advantage against supply-constrained fabless competitors who are fighting over TSMC allocations, why sell any of it to someone else? ¯\_(ツ)_/¯

And SpaceX could become a very large Nvidia customer long before Terafab is ready. The company says it expects to have more than 2 GW of compute by the end of 2026, and that by the end of 2027 it could be closer to 10 GW than 5 GW. Then there’s Starmind: SpaceX’s first orbital-compute satellite is supposed to basically carry an optimized Vera Rubin NVL72 into space. 🛰️

Nvidia could supply SpaceX on Earth, then follow it into orbit. 🌎🚀🛰️

🏗️🤖 SB Energy IPO: $430 Billion of Backlog, Zero Data Centers

SoftBank-backed SB Energy filed to go public this week, and this number made me do a double take (almost a spit take): $430 billion of data-center backlog (and $439bn total).

This is a company that currently operates... zero data centers. 😅

Read more

In the creative process there are certain strategic moments in which it seems as if you are either standing still or even going backward.

These moments of apparent lack of progress are strategic because the actions you take at such moments will largely determine whether or not you are ultimately successful.

—Robert Fritz

🧭🏛️⚖️⭐🌱 I don’t remember why we got on this topic, but my oldest boy asked me what a virtue is.

To explain it better, I looked up some of the classical lists of virtues, then broader lists covering character, thinking, relationships, civic life, etc. As we went through them, I explained some in more detail and told him that you could spend a lifetime getting better at any one of them. They’re guiding stars, not destinations.

I’m not sure he was super impressed, but I thought this was an important moment, so I decided to print out a one-pager of virtues and put it on his bedroom whiteboard with a magnet. Hopefully this becomes another bit of that ambient learning I wrote about in the intro of Edition #649, when I got him a periodic-table poster for his bedroom.

Is this like a periodic table for virtues 🤔

The one on his wall is in French, but I translated it above so you can have a look 👆

💾🎨🧑‍🎨 I think this tweet by Kasra Kyanzadeh, an experienced engineer who just left OpenAI’s Codex team, is thought-provoking:

The tweet is truncated in the preview. I’ll save you a click and reproduce it in full here:

None of his complaints are about AI being bad at coding. More the opposite: the tools are now good enough that they change what the job feels like.

This is the other side of the coin from all the people I’ve heard say they’re having more fun than ever, or doing things they couldn’t have done before. (I count myself among them, by the way, and I’ll tell you more about my vibe-coded apps another time.)

I can certainly imagine that a lot of people got into software because they liked a particular kind of work. Methodically writing the code. Solving many small logic puzzles. 🧩 Holding a complicated system in their heads, disappearing into it for hours, in a flow state… 😵‍💫

And now, in less than a year, their whole way of working has changed. Opus 4.5 only came out last November! I can imagine the malaise when something you spent years getting really good at suddenly becomes a much smaller part of the job because LLMs can do so much of it faster.

And not every engineer likes design or management or high-level product thinking, but they’re increasingly being pushed in that direction (managing agents, if not people) because more and more of the low-level implementation stuff is being automated.

It’ll drive some people out and attract others in (at least until AI agents are fully autonomous 😅). The people who love building software in 2030 may be a different group from the people who loved building software in 2020.

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Liberty Capital 💳 💴

🤖💵 The Fed is Now Thinking in Tokens 📊

A few years ago, almost nobody outside of AI knew what a token was.

Now the Chairman of the Federal Reserve is at Jackson Hole discussing reports of $100+ billion in annualized token sales from the two leading AI labs alone, asking whether tokens will be substitutes or complements for human labor, and wondering what their ‘equilibrium price’ will be.

Huh 🤔

Here’s the part of the speech about it:

Capital and labor have combined to create the large language models at the heart of AI. Users buy tokens to gain access to the models. [...]

Will token usage be complementary or competitive to labor? Will the next generation of AI models demand even greater capital intensity, or will the models themselves help devise a capital-light solution? [...]

We don’t yet know the equilibrium price of the tokens. Might there be a heterogeneity of tokens, such that growing sums will be paid for access to the best models at the frontier? Will token prices for older models fall to the level of their marginal cost?

A few months ago, Satya Nadella talked about companies eventually having “human capital and token capital.” Now the Fed is starting to think in similar ways.

But as I wrote about in Tokens Schmokens: We Need a Better Economic Unit of AI Work (Edition #634) and Tokens are NOT like Kilowatt-Hours or Gallons of Gasoline (Edition #630), raw token counts are a pretty terrible way to compare AI output across different models and workloads.

A token is a useful billing/usage unit, but it’s not a standardized unit of economic output. They are NOT a commodity, and a token from one model or workload is NOT economically fungible with a token from another.

A million tokens from Mythos doing super-difficult agentic coding or working on an advanced math proof and a million tokens from a tiny model summarizing spam emails are both counted as ‘a million tokens.’ But economically, they’re very different.

Maybe what economists will eventually have to track is something closer to the price of useful intelligence (if we can figure out how to measure it). 👩‍🔬 In practice, that probably means something more like the cost of completing a given task at a given level of quality.

This starts to become a very weird quality-adjustment problem. If token prices fall 80%, but the models get much better and need fewer tokens to accomplish the same useful task, what actually happened to the real price of AI work?

It’s a strange new thing that economists have to think about. ¯\_(ツ)_/¯

🤖🦅🛡️ CrowdStrike’s AI Security Advantage: The Agent is Already There

I haven’t written about CrowdStrike in a while, but this seems like a good time. Everyone has cybersecurity on the mind these days, for obvious reasons. 🔐

The numbers are impressive: Net new ARR grew 51% to a record $333 million, and revenue grew 26%. (Look at that reversing trend 5 quarters ago)

As expected, AI is part of the reason. What jumped out at me, though, is how CrowdStrike is turning demand into revenue: it can sell existing customers AIDR, its AI-security module, without installing another piece of software.

Its Falcon agent is already installed on customer machines. Because of how CrowdStrike built the platform, adding AIDR is basically flipping a switch on a customer’s account. Very low friction, and we know how much that matters.

The CEO explained it pretty directly on the call:

George Kurtz: So on the first point, it is separate and incremental. It’s another module and it’s priced separately, which is, again, where we’re seeing a lot of the growth. [...] When we think about the beauty of the architecture and the platform, it’s still using the same agent.

And if you want to enable AIDR, again, that’s a license entitlement. It’s part of our friction-free deployment.

[...] you don’t have to roll out yet another agent. [...] it’s on the same platform, that’s the beauty, but we’re able to monetize it separately.

This is one of those nerdy technical architectural decisions that turns into a big business advantage years later. CrowdStrike already did the hard work when it got Falcon installed on all those customer machines. AIDR’s ending ARR nearly tripled sequentially in Q2.

Though keep in mind, “nearly tripled” can sound more impressive than it is when you don’t know the starting point, and CrowdStrike doesn’t disclose AIDR’s absolute ARR yet.

🤖 Mythos & Fable 5.1: Generalist Knowledge, Specialist Judgment

Anthropic just released Claude Fable 5.1 and Mythos 5.1. There are lots of impressive benchmarks, as usual. But what stood out to me in the 212-page system card was an experiment comparing specialists with generalists using AI.

Anthropic paired PhD biologists with LLM experts, then gave them a difficult research-planning problem. Some of the biologists were specialists in the relevant field, while others were generalists.

The question this gets at is basically: how much can a powerful AI close the gap between a smart generalist and someone who has spent years becoming a specialist?

Well…

The specialist teams did a little better overall, but the graders couldn’t reliably tell the specialist submissions from the generalist ones. The strongest generalist team even beat one of the specialist teams on every measure and got the highest innovation score of the whole group.

Seven of the nine participants also said completing the task would have been *impossible* without the model.

But this wasn’t quite “generalists are now specialists.” The specialist teams still had a small edge overall. And across Anthropic’s broader red-teaming, people with enough expertise to challenge Claude were better positioned to catch its mistakes.

Maybe the more interesting split isn’t generalists vs. specialists. It’s specialist knowledge vs. specialist judgment.

A good biologist can now cover more ground in statistics, bioinformatics, chemistry, coding, etc. The model can give them access to knowledge that previously required years of study and work inside each field.

Knowing enough to recognize when the model is subtly wrong is different. That may still be the specialists’ edge.

So generalists get broader and specialists can apply their judgment to more work. But what happens as the models improve? Does that gap keep shrinking, or does specialist judgment get leveraged across more and more work? Or both at the same time? 🤔

Personally, the Anthropic model I’m waiting for is Opus 5.1.

Fable is too expensive for me to use as a daily driver, and Opus 5 is strong in some ways, but annoying in others, so I hope that they fix it with 5.1. In the meantime, I’m mostly using GPT-5.6 Sol.

If you want a deeper vibe-check on Mythos & Fable 5.1, here’s Dan Shipper’s early review. He had access for a week before release. His bottom line:

🗣️💬 Gavin Baker: “Thank You, Jensen”

Baker has been on a roll lately, doing many good interviews. This one is wide-ranging, covering most of the big topics in tech and AI lately. I recommend the whole thing.

Here are my highlights:

Gavin Baker: I think [Jensen is] in a very, very good position with his strategy of being vertically integrated but horizontally open. [...] if you’re a semiconductor CEO, the only thing you should ever say is, “Thank you, Jensen.”

“Thank you for creating this opportunity. Thank you. How can we work with you? We want to enable you. Sure, we’re going to compete with you on the edges.”[...]

Every 1% of share today is probably worth $100 billion. [...] So there’s no need to go head-on with NVIDIA. [...] Just pick a niche and get your 1%.

David George: That pie is very big.

Gavin Baker: He has 9 chips. He's got multiple flavors of accelerators. He's got CPUs, Ethernet switches, and 2 kinds of DPUs. We've gone from just scale-out networking being a thing. We have scale-up, scale-out, scale-across, and now scale-in. So just try to find a way to plug into his ecosystem.

By the way, this is not foreign. His biggest customers all have competing products across various of those 9 chips. [...]

It's like, hey, that's great. You did the one thing. To actually be competitive with him at the system level, you need another 8 chips.

Baker’s ‘$100 billion per 1%’ is obviously a rule of thumb, but the AI compute pie is so large that you don’t need to replace Nvidia to win.

Nvidia is very integrated. Replacing the whole system would be BRUTAL. But at the same time, they’re open enough that another company can build a huge business by plugging into their stuff.

Competitors need to ask: “Which slice can we be the best in, while making sure we work really well with everything else?” 🥧

That means many of Nvidia’s competitors wind up strengthening its ecosystem, while also buying tens of billions of dollars of Nvidia products.

The other one I wanted to share was about customer preferences:

Gavin Baker: In a world that is so supply-chain-constrained, it's actually really hard to tell what true customer preferences are.

If you have a TSM allocation, you're going to be sold out. Particularly if you can get the DRAM to pair with it. You're going to be sold out.

So it's actually kind of hard to infer true customer preferences.

During a shortage, a mediocre chip can sell out simply because customers will take whatever compute they can get. Market share can tell you as much about what customers could get as what they actually preferred.

BUT

The concessions required to generate adoption may tell you more:

Gavin Baker: I actually think one of the best ways you can see true customer preferences is the kind of deals they cut with chip companies.

So broadly speaking, the first deal is where the chip company invests in a customer, and you saw TPU and Trainium—Amazon and Google—do that with Anthropic. [...] In that scenario, as long as the dollars you invest are less than the gross profit, you can’t lose money.

Then there’s the scenario where you do the RVG. Blackstone finances it, or whoever—Blackstone, Apollo, KKR and Goldman Sachs finance it. As long as that RVG is actually less than your gross profit, you can’t lose money, and you have upside, probably through a revenue share on top of it.

(RVG here means residual-value guarantee: a promise that the equipment will retain a certain value.)

Then there are deals where you give warrants away, but they’re tied to a fixed price per 1 million tokens. [...] If you just give warrants away, it could be negative NPV, because the better the stock does, the worse the deal is.

[...] You can look at that hierarchy of deals and infer something about true customer preferences.

A much cleaner test will come when supply loosens enough that customers can actually choose. (But when will that be? ¯\_(ツ)_/¯ )

🇹🇼🏜️ A Replica of Central Taipei in the Inner Mongolian Desert

Well, this is slightly ominous 😬

China has built a mock-up of part of Taipei’s government district in the Inner Mongolian desert for PLA training.

This is not China’s first mock Taipei. The PLA has had a separate replica of the Presidential Office and nearby government buildings at the Zhurihe training base for more than a decade. The newer site in Cameron’s images appears to focus more on the surrounding street grid and city blocks. Historical sat imagery suggests it did not exist in 2020.

It’s hard to make “we’re planning for this” much more literal than building a copy of the place in the desert for military training.

From the full feed 🔒: Thoughts on Nvidia Q2 + Hugging Face Deal — The quarter was absurd, but the Hugging Face deal may be the more interesting signal.

🧪🔬 Liberty Labs 🧬 🔭

🧫🔬🤖👨‍🔬 Bringing Claude Into the Lab

Anthropic just opened a research preview of its Model Hardware Standard (MHS):

We’re opening a research preview of the Model Hardware Standard (MHS), a shared specification for AI agents to safely operate physical devices, to a first group of scientific research labs and advanced manufacturers.

MHS enables AI agents to operate multiple lab and manufacturing instruments, such as microscopes, liquid handlers, and robotic arms, in parallel, and perform intricate tasks ranging from routine drug discovery experiments to laser calibration on a quantum computer.

This is very cool. MHS is kind of a glue layer between the AI and all the different instruments in the lab, allowing the AI to control and coordinate them much more easily:

Getting multiple devices in a lab or on a factory floor to communicate with one another can be challenging, even setting aside the added difficulty of integrating AI into the setup. Each device tends to have its own programming interface, and so far there has been no standardized way to integrate them. And once the devices are connected, there is no common way for them to share data with an AI agent, nor to let the agent operate them safely.

MHS addresses these challenges by introducing a standardized driver: software that translates between a computer’s operating system and a hardware device.

If it works, it could let scientists spend less time getting all their equipment to talk to each other and debugging it, and more time on the actual scientific questions.

And by the way, the production values on this video are great. 🎥

The aspect ratio makes it look more like IMAX than a regular YouTube video, and whoever shot it clearly knows what they’re doing. Lots of very interesting shots, good composition, nice depth-of-field effects, etc. Kudos.

h/t my friend MBI (🇧🇩🇺🇸) for sharing this via DMs

🏴‍☠️🤖 Abliteration: Unlocking the Cyber Skills Behind the Guardrails

Abliteration AI is taking open-weight models and removing some of the refusal behavior that makes them say “no” to certain requests.

Their newest one is based on Z.ai’s GLM-5.3, which is already very good at cybersecurity. They’re selling the modified version for *offensive* security work: exploit development, red-teaming, etc. Basically, the kind of stuff that mainstream models will generally refuse to help with.

So what exactly are they removing?

When a model goes through post-training, it isn’t only being taught to give better answers. It’s also being taught the kinds of requests that should trigger a refusal. The model may know how to write exploit code, but it has been taught a kind of reflex: this looks dangerous, don’t answer. 🛑

Researchers found that when you look at what’s happening inside a model as it refuses a request, there are identifiable patterns in its activations associated with that refusal. Those patterns are surprisingly concentrated.

Abliteration finds those ‘refusal directions’ and edits the weights to weaken them.

This isn’t like most other jailbreaks we hear about, where the model is tricked into answering something (sometimes via very strange prompting). This is more like a selective lobotomy. They’re changing the model’s weights. 🧠✂️

Cybersecurity is the area where this makes the most sense because the white hats and the black hats often need the model to do the same thing.

If you’re a security researcher trying to prove that a vulnerability is real so it can be fixed, you may need the model to write an exploit for it. Same as if you were someone trying to break into someone else's system without permission ¯\_(ツ)_/¯

And this is also a business. 💵

Abliteration hosts these modified open-weight models and sells access through an API. Their GLM-5.3 version costs $5 per million tokens of input or output, so security researchers can use it without having to run a 753B-parameter model themselves.

They do have optional moderation and a Policy Gateway for customers who want more controls, but the underlying product deliberately has very few restrictions.

So I hope most of the people hammering those GPUs are white hats helping us patch the world, and not black hats doing the opposite. 🏴‍☠️🤔

h/t Supporter Boogie (💚 🥃), who shared this with us on the private Discord

From the full feed 🔒: HEICO: 10x Production Without 10x Capex — How do you prepare for 10x production without spending anywhere near 10x on capex?

🎨 🎭 Liberty Studio 👩‍🎨 🎥

📺 Mini-Documentary About Master of Puppets

I recently posted the Metallica documentary about the Rasmussen trilogy.

I couldn't get enough, apparently.

The YouTube algo suggested this one to me, and I was like, well, I may as well check it out, just to see if it's well made.

I pressed ▶️ on my living room TV… and I watched the whole thing before bed. 😅

It's quite well executed, and very interesting.

What surprised me is that Stuart Kirwan only has three videos on his YouTube channel. This doc is from 6 years ago, so well before the current generative-AI video era.

Turns out Kirwan really is a pro. He’s a professional editor and motion-graphics artist who has worked for the BBC, ITV, Amazon Prime, DAZN, etc. He also posted this video showing how he edited some of the B-roll:

It’s really cool to see how he took reference images and turned them into something very different.

Plans are worthless, but planning is everything.

—Dwight D. Eisenhower

🇺🇦 I’ve been thinking about Skunk Works lately. Not only the great book by Ben Rich (read it!), but what Lockheed actually did to make innovation move so quickly.

The U-2 spy plane went from green light to first flight in 8 months. The A-12, the CIA spy plane that preceded the more famous SR-71 Blackbird, went from contract to first flight in ~27 months. The F-117 stealth fighter did it in 31 months.

Crazy, no?

It worked because they created an architecture of learning.

It compressed the feedback loop inside a company. The engineers were close to the people building the planes, the people testing them, and the customer. Everyone was sitting next to each other and constantly talking and seeing the direct result of their work.

SpaceX is a good example of this with rockets. 🚀

But today, Ukraine is turning into something like a country-scale Skunk Works.

They’re compressing that loop across an entire military-industrial ecosystem, from the battlefield to the engineers to the manufacturers.

There’s a very dark reason the feedback is so fast and the signal is so strong: the battlefield is an unforgiving test environment. Things that don’t work get exposed very quickly and people die.

As Torsten Reil of Helsing puts it, the truth is in the field. (more on him below)

📲 Poking in ChatGPT’s settings, I found this:

Makes you wonder how many depressed and suicidal people are chatting with AI right now. The labs must see all kinds of things in their logs… 😔

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Business & Investing 💳 💴

🔥 Thoughts on Nvidia Q2 + Hugging Face Deal 😎

Read more

The envious man thinks that if his neighbor breaks a leg, he will be able to walk better himself.

—Popular saying quoted by Helmut Schoeck

✨🌌🔭 The pattern of the stars as seen from our planet is like a fingerprint.

It’s unique to a position in space and a moment in time.

I just think that’s cool.

🍝 I had never really thought that much about pasta until I heard a podcast series about a guy who decided to create his own pasta shape because he thought he could do better than the existing ones. I wrote about it here, if you’re curious.

I decided I should experiment a bit with different shapes and brands to see if maybe I could find new favorites. It’s a one-time exploration that can lead to years of better food, so it seemed like a pretty good trade! (The same idea applies more widely to getting better at cooking.)

I found a Montreal store that carries a ton of Italian brands and sells them online, so I ordered a big box of pastas that seemed interesting. I did some research into which brands were considered good quality, then told ChatGPT my current favorites and asked it what else I might like.

So far, it’s been one of the most fun small experiments I’ve done lately. Can’t complain about eating pasta! 😅

If you’re curious, my first batch included: Casarecce, Gigli/Campanelle, Fusilli corti bucati, Mafalda corte, Gnocchi sardi, Mafaldine, Mezzi rigatoni, Tortiglioni, Orecchiette, and Radiatori.

If you’re looking for brands that should be decently easy to find and pretty good quality, see if you can find Rummo, Granoro, La Molisana, and Garofalo. Things like bronze-die extrusion (it creates a rougher, more porous surface, so sauce sticks better!), good durum wheat, and slower/gentler drying can make a noticeable difference to the texture, flavor, and how well the sauce clings.

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Business & Investing 💳 💴

🧩 Google is Trying to Unbundle Nvidia

Google and Marvell just expanded their custom-chip partnership, with Marvell helping develop Google’s custom silicon for the TPU ecosystem and giving Google a warrant to buy up to 58.97 million shares.

At first, this looks like another ASIC/custom-chip win. But look at what the deal actually covers: AI inference accelerators, storage controllers, NICs, memory-interface controllers, and near-memory compute, all attached to Google’s TPU ecosystem. That’s a lot more than “help us make another chip.” Google isn’t really trying to unbundle the AI factory. It’s trying to unbundle Nvidia as the company that integrates the whole thing, then re-integrate those pieces around Google.

In this model, Google is the system architect, using specialist chip companies to build different pieces underneath it.

Nvidia’s moat isn’t just the GPU. It increasingly sells the whole AI factory: accelerator, CPU, high-speed interconnect, networking, DPUs, rack-scale systems, and the software tying it all together. It’s what Jensen calls “extreme co-design”: the pieces aren’t optimized in isolation; they’re designed together as one system.

Broadcom is already a major partner on TPU and networking, and Marvell is now working across inference, networking, storage, and memory. Instead of buying one vertically integrated stack from Nvidia, Google can keep control of the architecture while specialist suppliers build different pieces for it. (not that Google is stopping buying from Nvidia, it’s still a large customer… but without TPU, they would be buying A LOT more)

The reason Google can even attempt this is that it already has a gigantic built-in customer: itself (similar to Amazon with many things). TPUs power Gemini and huge Google products like Search, Photos, and Maps, and Google also sells TPU capacity to outside customers through Google Cloud. That’s an enormous workload to design around, and plenty of internal demand to justify building this whole ecosystem rather than buying it from Nvidia.

This is basically how TPUs started. Jeff Dean did some napkin math in 2013 and realized that rolling neural-network speech recognition out at Google scale could require roughly doubling its datacenter compute. He reportedly told Urs Hölzle: “We need another Google.” 😅 So instead, they built the TPU.

The tradeoff is that this gives Google more supplier diversity, but also creates a harder coordination problem. Nvidia can optimize chips, networking, software, and racks against one roadmap inside one company. Google has to get its own teams, Broadcom, Marvell, and everyone else moving in sync.

So what about the warrant?

It has a $206.58 exercise price, and almost all of Google’s 58.97 million Marvell shares it covers vest based on purchases: one tranche for every $500 million of qualifying revenue, up to 240 tranches, or $120 billion. To be clear, that’s not a $120bn purchase commitment, it’s just the top of the warrant’s vesting ladder.

How much of Nvidia’s moat comes from owning the best individual pieces, and how much comes from coordinating them under one roadmap and integrating the whole system better than anyone else? 🤔

🗣️💬 Sam Altman on AI’s Two Bottlenecks: Human Inertia and Context

Whatever you think of Sam Altman, he’s consequential. OpenAI is at the center of everything these days, and any chance to get a better view into his thought process is very interesting.

My friend David Senra (📚🎙️) went from interviewing books and dead people to interviewing the live players. In a way, it’s the same job, he’s just getting the same kind of wisdom out of a different source.

And it’s a natural progression. Books are permissionless, the ideal place to start. His hard work over many years has earned him ‘permission’ to do this (in this case, being on the radar of the people he wants to interview, and having their trust).

In the same way that I much prefer when a musician interviews musicians, and they can go deep and nerd out about the actual music, I think it’s great when founders are interviewed by those who understand them, like David. Journalists and historians are important, and we need more good ones too, but they do a different job.

I think some of David’s interviews will be the raw material for books and biographies that will be written years down the line. Documenting how a player thinks about things in real-time, having them talk about their journey in their own words.

So many interesting stories were no doubt lost to time because no one did a good interview and the people didn’t write it down themselves.

Anyway, back to Altman, here are a couple of my highlights, first on being wrong on timelines and things unfolding SLOWER than he expected:

Sam Altman: I love startups. I think startups are the coolest thing in the economy, and I’ve spent my career trying to really understand startups. When we got to GPT-4, back in 2023, I thought that very quickly after that there was going to be much more disruption in software, much more of business being up for grabs right away, than it turned out to be.

I think I was wrong about a few things, but one of them, in terms of the speed, is that the economy just has so much inertia. People keep doing the same things they’re doing. They keep buying from the same company. They keep wanting to use their tools in the same way.

Technological capability and economic disruption *aren’t* the same curve. The models can improve insanely fast, while companies, workflows, habits, purchasing decisions, and organizations change much more slowly.

That inertia is also a kind of moat for incumbents.

A startup can suddenly have dramatically better technology and… still have to convince customers to switch, retrain employees, change workflows, trust a new vendor, integrate new software, etc. The technical disruption can arrive a while before you feel it as an economic disruption.

Sam Altman: I think this is actually a positive in many ways, and it’s going to make this big transition in front of us go smoother and slower. I’m grateful for it. But I think it means we’ve all been too ambitious on timelines.

Even with this incredible technology — I think AI is one of the most incredible technologies humanity has ever invented — society and the economy will adapt more slowly.

Next on context as a bottleneck:

Sam Altman: With the latest generation of models, I don’t want to say they feel smart enough, because I think we should always aspire for them to get smarter, but they are pretty smart. And I feel more limited at this point by the amount of useful context AI has on me.

If model intelligence is becoming less of a constraint, the bottleneck moves to the stuff around the model. And this isn’t about having a giant context window. It’s about the quality of that context, its usefulness. Even if you’re smart, if you don’t have the right info to work from, you can’t be effective.

Sam Altman: I want the AI to know as much as it can to help me. I want it to be doing things I can’t or don’t want to do on my own. I’m not going to read every post on our internal Slack. I’m not going to read every story a customer has to tell about where ChatGPT worked for them or failed them. I can’t. I probably could read more research papers than I do, but it takes a lot of mental energy. I would love to have an AI agent that is constantly trying to be helpful to me, that can look at and understand more context than I can, or than I have the time or energy to do on my own, and can help bring that context to bear and give me good advice when I have to make a decision.

I think we’ve focused correctly so much on model intelligence that, on the product side, we have not yet thought enough about what it means to give a model more context than any person could have. There are plenty of very smart people, but there is no one who can read tens of thousands of pages of context in some small number of seconds and really use that accurately. And this is something that AI can do that is going to be very new.

These two quotes from the interview rhyme a bit.

They say that the model is moving faster than the things around it. Companies and institutions are slow to adapt to what the models can do, and on the product side, we’re still figuring out how to give the AI more useful context.

This hints at what they’ll try to do with their physical products. Some kind of ambient AI that gathers more context by actively listening and watching..? 🤔

🌶️ OpenAI’s Jalapeño is Apparently Very Spicy 🌡️🔥

Speaking of OpenAI, back in February 2025, when OpenAI’s custom-chip project was still mostly plans and rumors, I wrote:

I wouldn’t be surprised if its main contribution — at least for the foreseeable future — is to give OpenAI more leverage when negotiating with Nvidia.

It’ll take a while to ramp up production, for the chip and its software to mature, and meanwhile, Nvidia is a fast-moving target with incredible chip design resources and a decade+ buildup of software libraries and frameworks.

Well, time to update! 😅

It appears to be MUCH better than I expected from a first-generation ASIC.

Based on what SemiAnalysis saw (they verified the InferenceX runs in OpenAI’s lab, though OpenAI supplied the numbers and SemiAnalysis didn’t run the full suite or its harder AgentX benchmark), Jalapeño produced roughly 1.5-1.9x more throughput per watt than Nvidia GB200/GB300 systems, depending on the workload, while also delivering much lower latency.

And that’s a metric to focus on, because AI datacenters are increasingly power-constrained. If the number of megawatts is fixed, tokens per watt basically helps you squeeze out more tokens per dollar from your infrastructure.

SemiAnalysis also says Blackwell isn’t really the fair comparison. Vera Rubin is. And things get much closer there: Jalapeño still wins on the current perf/watt numbers, but SemiAnalysis has the two roughly tied on perf/TCO.

The big difference is time. Rubin is arriving in production datacenters NOW:

Meanwhile, Jalapeño is still at engineering-sample stage, and most of its volume is currently scheduled for late 2027. So the real question is what it’ll look like against a much more mature Rubin stack by then, with Feynman coming next in 2028.

¯\_(ツ)_/¯

It may not do as well with some harder workloads, larger models, etc. There could be production delays and other bumps in the road that end up making it less competitive.

But it’s still EXTREMELY impressive how quickly OpenAI went from 0 to 60 with this ASIC project, both on the hardware and the software.

I wrote a lot about the Escher Hands effect with AI writing the code that builds the next AI.

AI helped design the chip, AI is already helping write and optimize code for it, and future generations of AI will run on hardware like this and help design and program the chips that come next. 🔄

💵💉 The $500,000 Check That Started Moderna’s Cancer Program

I wrote about Moderna's personalized cancer vaccine in Edition #654. What I didn’t know was how BONKERS the origin story of the program that created it is.

The WSJ wrote about Patrick Degorce back in 2020:

“The first time hedge-fund manager Patrick Degorce met with biotech company Moderna, it was a hail-Mary effort to find a cure for his wife.”

This was 2011. Moderna had about 10 employees, and Degorce’s high-school sweetheart wife had recently been diagnosed with Stage IV lung cancer. He began personally investing in Moderna the following year.

Then, in 2013, having lost his wife (😢), he reached out to Stéphane Bancel, the CEO of Moderna, to ask about the potential for using mRNA against cancer.

Bancel told him that Moderna had some ideas, but the young company couldn’t afford to pursue them. (Moderna was founded in 2010)

So Degorce asked how much it would cost to hire a couple of scientists and start working on it. 👨‍🔬👩‍🔬

Bancel’s napkin math answer: $500,000.

Degorce sent him a cheque for that amount. Not an investment. A gift. 🎁

Now, the program cost a lot more than that, and Merck put $200m into it later. But without this start, who knows what would have happened. It may not have existed at all, or may have been delayed by years, etc.

That’s why the early people are so important (founders, early investors, early technical people). Without them, whole branches of possible futures disappear. 🌱🌳

Patrick Degorce, you are a mensch.

h/t Compound248 (👋) and supporter DvorakQ (💚 🥃)

From the full feed 🔒: The AI Hardware Heist Economy: When a Truckload of Electronics Is Worth $38 Million — When a single truckload is worth $38 million, you get a whole new kind of heist economy.

🧪🔬 Science & Technology 🧬 🔭

🚗🗺️ Google Maps is Becoming Air Traffic Control for Roads

I've long been fascinated by the power that GPS apps like Google Maps have over the real world and over people's actions. Most drivers now trust GPS so much that we just set it and forget it.

But there are very interesting game theory aspects to this thing (can we mention game theory again or is it still a banned phrase because people on LinkedIn use it in cringe posts like a few years ago? I’ve lost track). If every driver is routed independently, the app is trying to find the best route for that driver. But Google can see the whole network, including where it's about to send everybody else. That means it can sometimes give you a route that's basically just as good for you, but better for everybody.

It’s basically what game theorists call the “price of anarchy”: everyone choosing the route that’s best for themselves can produce a worse result for the network as a whole. Sometimes making a tiny number of drivers take a route that’s almost as good for them makes things better for everybody. And because it’s an iterated game, the driver getting the slightly worse route today doesn’t have to be the driver getting it tomorrow.

Google Research recently published a cool real-world experiment testing that.

Over six months, they experimented with routing in 10 major U.S. cities, identifying roughly 100 recurring bottlenecks in each. On treatment days, Google Maps slightly discouraged routes through those bottlenecks when it could send someone along a similar road with a comparable travel time.

The intervention was tiny: fewer than 2% of observed trips actually received a different routing recommendation.

And that was enough:

• Speeds on the targeted bottlenecks increased ~2%

• Across the much larger network of affected roads, speeds increased ~0.35% overall and ~0.5% during peak hours

• Average trip travel times improved by about 0.7%

Non-linearity ftw!

Air traffic is centrally coordinated. Internet packets are routed with the state of the network in mind. Cars have mostly been millions of independent agents trying to get somewhere as quickly as possible.

For now, this experiment was fairly simple and relied on historically congested bottlenecks rather than some all-seeing real-time optimizer. But add better models, connected traffic lights, more connected vehicles, and eventually autonomous cars…

The navigation app starts looking less like a map and more like the control system for the road network. 🤔

🔁 Déjà vu: Google Maps data used to optimize traffic lights

I think I like… Grok 4.6 😳

Didn’t see this one coming! After all of xAI’s original co-founders except Musk had left, Elon said the company hadn’t been built right, most of the Grok use I saw in the wild seemed to be people asking it to explain tweets, and they started selling compute to Anthropic.

I figured that they were pretty much done as a frontier lab. 🪦☠️

Then SpaceX bought Cursor, and I was skeptical. Cursor is a good company full of talented people, but they’re not a frontier lab. Their strongest model work had been things like Composer 2, which started from the Chinese open-weight Kimi K2.5 and then did more training around coding.

That’s not nothing, but it’s not quite the same as what OpenAI and Anthropic have been doing…

And yet, I want to keep an open mind. So when Grok 4.5 came out, I had a look and was pleasantly surprised. I started sending more work its way, making it double-check code from my vibe-coding projects (I’ll tell you more about that some other day!), or I’d send it research questions I was also asking GPT-5.6 Sol to compare.

Then Grok 4.6 came out, and that was even better! I could see myself promoting it to my model toolbox. 🧰

At the same time, Claude Opus 5 and Sonnet 5 have been annoying me more and more. They have a tendency to write in heavy jargon. That’s fine for chain-of-thought, but it keeps spilling out into the user replies, even if I try to steer the model away from it (it keeps coming back).

But worst of all: Claude 5 has been a lot less reliable than past versions, in my own experience. I’ve had it invent things, or tell me things that don’t turn out to be correct when I verify, or do things that GPT-5.6 will check and find errors in.

All this adds up to Claude Opus and Sonnet 5 not having a very pleasant flavor 🍦

Meanwhile, OpenAI really killed it with 5.6 and with recent Codex improvements, it’s probably my favorite model + harness right now… and Grok (it still sounds weird to me) is increasingly earning a spot on my roster, mostly replacing some Claude usage.

I’m just as surprised as anyone else. Just a few months ago, Anthropic was king of the world and clearly had the best models (especially for coding/agentic stuff).

It’s almost more like the weather than a typical product industry.

🧑‍🔬 The Little AI Scientist Supervising a Giant 🤖

Inherent just released an interesting paper about Faraday, a 27B-parameter “AI Scientist” trained to replicate results from scientific papers.

They built 310 tasks from 100 papers: remove one of the paper’s result figures, give the agent an hour and limited compute, and see how well it can recreate the experiment that produced it. Faraday beats Claude Opus 4.8 and GPT-5.5 on the authors’ benchmark.

There’s a bit of an Escher Hands (🔄) aspect: Faraday itself uses GPT-5.5 Codex to do the coding (at the time of the paper, anyway, I wouldn’t be surprised if it got upgraded to 5.6 soon). So this relatively small model is basically acting as the principal investigator, deciding what to try and directing a vastly larger model that does much of the actual implementation.

One example from the paper stands out to me. In an LSTM replication task, Faraday’s Codex coding agent couldn’t get the experiment to converge and eventually tried to hand-craft the network to produce the expected result. Faraday basically said nope!, backed up, and found a training recipe that actually made the experiment work.

The giant model was smart enough to build a Rube Goldberg machine to jury-rig the answer. The little one had the judgment to know that getting the right-looking answer the wrong way wasn’t the point.

From the full feed 🔒: Visa & Mastercard: How to Turn Competitors Into Complements — What if Visa and Mastercard don’t actually need your payment to run on Visa or Mastercard?

🎨 🎭 The Arts & History 👩‍🎨 🎥

🤘 Documentary about Metallica’s Rasmussen Trilogy (Ride, Master, and Justice) ⚡🔔🔨

I have pretty eclectic taste in music, but growing up, these three albums were a huge part of my musical education. I’ve heard them countless times. They’re familiar places in my mind. 🧠💭🎶

So this fan-made documentary about them is tapping into a very deep nostalgia for me. I hope it does the same for you, or maybe, if this stuff is new to you, that it’ll convince you to discover it.

Even if you’re mostly into hip hop or electronic music or whatever, forget about genre for a sec. There’s great music that sounds all kinds of ways, and IMO these albums qualify.

If you shut up truth, and bury it underground, it will but grow.

—Émile Zola

🤔💭❓ One of the standard performance stats for cars is their 0-to-60 mph time (or 0-to-100kph here 🇨🇦).

I think there’s something similar with curiosity.

Some people are curious but have a slow 0-to-60. They wonder about things without doing much about it for a while. Others get curious and immediately start looking things up, asking questions, experimenting, learning a new skill, whatever. They get very good at going from ‘uh, I wonder if…’ to a useful 80/20 understanding surprisingly fast. 🏎️💨

It’s something I’ve tried to cultivate in myself. But a lot of your curiosity acceleration also depends on the world you live in. If I had been born in 1382 instead of 1982, satisfying my curiosity about almost anything would have been vastly harder.

And today’s world is incredible for this (if you take advantage).

I look at my oldest boy, and I’m amazed at how fast he moves when something grabs his interest.

When I was learning guitar as a teenager, learning a new song might mean hunting through random tab websites, often finding nothing, waiting until I saw my guitar teacher the next week, or remembering to ask a musician friend the next time I saw him.

There were upsides to that. When you have to work hard for something, you often remember it better.

But I look at how much my boy can explore in an afternoon. He’ll find a piano tutorial on YouTube and learn a new song. Then research fitness or physics or some author he likes with ChatGPT and Google. Then go vibe-code a 16-bit style video game or a simulated petri-dish universe.

His curiosity has a much faster 0-to-60 than mine did at his age.

Just yesterday I was talking to him about the exploration/exploitation trade-off: how much time you spend exploring new things versus going deeper on things you already know you like. There are only so many hours in the day, so you can’t maximize both.

But I realized that the trade-off has changed because exploration is so much faster and cheaper than it used to be.

An hour of exploration buys you more than it used to. You can now get answers, learn the basics of a new field, or start making something yourself in less time than a bus trip to the public library used to take. 🚍

Which makes me wonder if a lot of us are still operating with an old intuition about how much exploration is worth doing. 🤔

🛀💭🪞🚙 Why don’t all car side mirrors have little round convex mirrors?

Seems like that would help with blind spots (even on cars that have blind spot indicators — an additional visual cue seems useful, right?)

I never thought of it before, but now that I looked, you can buy these and add them to your mirrors. I’m considering doing that on my car.

I realize there’s a trade-off because you get a distorted view and your brain has to reconcile two different perspectives, but I’d be curious to try and see if it’s an improvement.

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Liberty Capital 💳 💴

🏴‍☠️ The AI Hardware Heist Economy: When a Truckload of Electronics Is Worth $38 Million

I don’t want to jinx anything or give anyone bad ideas, but I was thinking about how insanely valuable some truckloads of electronics must be these days.

A truck full of gold is very valuable and needs high security. But what is a truck full of RAM or NVIDIA GPUs worth? And there are lots of those moving around the world as part of the supply chain feeding hundreds of billions of dollars of datacenter capex.

I wonder what kind of security is involved with all this. Has organized crime started targeting the industry yet?

I wonder if you can buy memory that fell off a truck somewhere in South Korea on the black market or whatever 🤔

I looked it up, and it turns out: Yes, they’ve been paying attention (maybe they’ve been asking Kimi K3 for tips on how to do heists? 😅).

CargoNet specifically lists RAM, storage drives, and enterprise computing equipment as increasingly attractive targets. And just last week, WIRED reported on two California shipments of AI datacenter gear that were valuable enough to have private security vehicles escorting them. In both cases, attackers apparently rammed the security escorts out of the way, after which the semitruck drivers kept going and disappeared with the loads. Investigators suspect the drivers were in on it. 😬

FreightWaves says known reported thefts of datacenter/high-tech cargo have totaled $110.6 million since June 3 alone, including individual loads worth as much as $38 million.

There’s a twist: cargo theft isn’t necessarily becoming more common, but the losses are becoming much more concentrated in extremely valuable loads. CargoNet says reported theft incidents actually fell 26% year-over-year in Q2, while estimated losses more than doubled, from $135.7 million to $304.6 million.

AI is increasing the value density of certain parts of the economy, and of course, that attracts unwanted attention. I’m not sure everything around these trucks is adapting as quickly as the value of what’s inside them. 🚚 🚚 🚚

What does security look like for a $10m+ truckload of computer parts? Private escorts, GPS trackers hidden in the cargo, tighter checks on drivers and trucking companies, etc.

Where does all this stolen RAM, networking gear, and server hardware end up? Apparently some of these international crime rings have basically built their own parallel supply chains, complete with customers waiting for the stuff.

There must be some pretty weird black-market datacenters out there. 🤔

👨‍⚖️👨‍💼📰 Is AI’s New Bottleneck Political Permission? 🤔

In Will we ever have enough compute? in Edition #654, I wrote:

Maybe that’s one thing to watch out for. So far, every time we improve one constraint, another bottleneck shows up somewhere farther out (like whack-a-mole). Chips, memory, networking, power, grids, capital, whatever. As long as the bottleneck keeps moving rather than disappearing, that seems like evidence that we’re still a long way from “enough compute.”

Read more

Please try to keep an open mind—not only to let new ideas in, but also to let the old ideas out!

—Larry Gonick, Cartoon History of the Universe III

🎬🎥🍿🇫🇷👽💥 Hey Hollywood, I hear you're back, making real event films.

I’ve got one for you.

Edge of Tomorrow is one of the most fun sci-fi films of the past 20 years. Unfortunately, it wasn't a big commercial success. The vague title and marketing probably didn't help, and it came out right in the middle of the superhero/franchise boom.

But its reputation has grown over time, and I think that if it came out today and was well marketed, it would probably make at least 75% as much as Top Gun: Maverick.

So here’s my idea: Do the prequel!

The hook is built into the first film. We never see what happened to Rita at Verdun, and that’s a great storytelling opportunity that sidesteps most of the pitfalls of a sequel (you have to do bigger stakes than the first film, but we already saved the world and beat the Omega, so where do we go from there?).

It doesn’t matter that Tom Cruise isn’t there. I want to meet her squad. Get some great actors playing some well-written characters, and this could be an Aliens-type space marines ride. (How I wish James Cameron would make this instead of Avatar 7 or whatever)

And now that the first film has explored the “learning to deal with the loop”, the prequel could move through that much faster and focus on the adventures/problem-solving side of it

I even made the poster for ya.

🤷‍♂️🤷‍♀️ Great quote by friend-of-the-show Rohit Krishnan (💚 🥃):

He then quotes Alfred North Whitehead:

“Civilization advances by extending the number of important operations which we can perform without thinking of them.”

I wish I knew that quote before writing Workbench #4, AI Is Becoming a Compiler for Intelligence. It’s such a perfect fit. I went back and added it in. Why not? Digital is not paper. It can evolve.

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Business & Investing 💳 💴

💉 Moderna’s Personalized Cancer Vaccine (Phase 3) 👩‍🔬🧫🧬

That’s the kind of news I like to wake up to. And I hope we’ll have many more mornings like this as scientists make more breakthroughs on things like cancer, Alzheimer’s, and… let’s cure the common cold while we’re at it.

Today’s news:

“Today’s results represent a landmark moment for adjuvant melanoma treatment. This is the first Phase 3 study to show that intismeran, a treatment designed based on the unique mutational ‘fingerprint' of a patient's own tumor, given in combination with pembrolizumab can reduce the risk of recurrence or death in patients with completely resected stage IIB-IV melanoma compared to KEYTRUDA alone,” said Professor Georgina Long, the study’s principal investigator and medical director of Melanoma Institute Australia, Chair of Melanoma Medical Oncology and Translational Research at the University of Sydney. “Intismeran in combination with pembrolizumab has the potential to establish a new treatment paradigm in the adjuvant melanoma setting, helping patients remain cancer-free for longer.”

This isn’t a vaccine that prevents you from getting melanoma in the first place.

These patients already had high-risk stage IIB–IV melanoma that had been surgically removed. The goal was to kill microscopic leftover cancer cells from staging an eventual comeback.

Each patient’s vaccine is personalized to their own cancer and given with Keytruda, an existing anti-PD-1 monoclonal antibody.

tumor sample → sequence tumor → computationally identify neoantigens → manufacture bespoke mRNA → vaccinate patient → train T cells to recognize that cancer

We don’t yet know exactly how large the benefit is because Moderna and Merck haven’t released the detailed Phase 3 numbers yet, but what has me most excited is what this validates: the basic idea actually works at Phase 3 scale. If the approach generalizes, this could become a platform for treating other cancers, not just melanoma. 🙌

This is the first positive Phase 3 trial ever for an individualized neoantigen therapy and for an mRNA-based cancer therapy.

Science FTW!

🌐 Will we ever have enough compute?

It almost sounds like a dumb question.

The first answer that immediately comes to mind is: no, we can always use more compute. We can always find more things to do. Chips keep getting better year after year. We still find new ways of improving them after all these years.

But is it a dumb question?

What’s the endgame here? How much of world GDP can be dedicated to compute?

The larger compute becomes as a fraction of the economy, the more “compute investing” starts to look like macro investing. We already went from looking at memory bandwidth and CUDA changelogs and number of cores per GPU to thinking about financing costs, grid infrastructure, and global geopolitics. If this keeps going, more and more of the economy gets pulled into what we have to worry about for the compute stack.

At what point do things break? Is it because a trend has been going on for some decades that it’ll keep going on?

Algae don’t take over the universe, even though if you graph out their exponential growth in some pond, that’s where the line goes.

At some point, something becomes binding. The algae run out of nutrients, light, space, whatever. What’s the equivalent for compute? Power? Capital? Chips? Land? Or do we just keep solving one bottleneck after another until the real constraint is simply that another dollar of compute isn’t worth another dollar anymore?

But when does that happen? In five years, in fifty years, in five hundred years?

When do we reach a steady state where the basic network is built and most investment goes toward maintaining and upgrading the existing stock rather than expanding it exponentially? Kind of like, I dunno, railroads.

I remember that in Isaac Asimov’s Foundation, there’s a planet called Trantor that is basically just one giant city covering the whole planet’s surface. Star Wars has something similar with Coruscant. Is that where we’re headed? Mars next? 😅

¯\_(ツ)_/¯

Maybe that’s one thing to watch out for. So far, every time we improve one constraint, another bottleneck shows up somewhere farther out (like whack-a-mole). Chips, memory, networking, power, grids, capital, whatever. As long as the bottleneck keeps moving rather than disappearing, that seems like evidence that we’re still a long way from “enough compute.”

The interesting moment may be when that stops happening. When we solve a bottleneck and, instead of immediately discovering the next shortage, capacity waits for demand.

✈️ 💾 Google Wins $10m Auction for Spirit Airlines Business Data from Bankruptcy

This reminds me of stockpiles of patents being bought from the carcass of fallen giants a few years ago. Google was a big buyer at the time:

• 2011: Nortel — Google tried to buy ~6,000 patents but lost the auction

• 2011: IBM — Google bought ~2,000 patents from them (if there’s one thing IBM is good at making, it’s patents! Too bad it doesn’t turn them into good products)

• 2012: Motorola Mobility — They bought the entire company for $12.5bn, and at the time a big part of the reported appeal was the ~17,000 patents and ~7,500 pending patents

• 2012: Kodak — Google was part of a consortium that bought ~1,100 imaging patents

Most of these purchases were defensive, just to make sure you don't get attacked in court, or that you have a mutually assured destruction type of situation with your competitors.

The gold rush for data is more offensive in nature, trying to gain an edge.

Now:

Google LLC won a bankruptcy auction for a trove of deindentified business data, software code, and operations records from collapsed low-cost carrier Spirit Aviation Holdings Inc., saying it plans to use the assets to improve its artificial intelligence.

The tech giant’s $10 million bid includes a vast repository of the defunct airline’s data, including 100 million emails, 500 million Microsoft Teams chats and collaboration records, plus information related to revenue, aircraft operations, employee productivity, and audits and fraud, according to an Aug. 14 notice in the US Bankruptcy Court for the Southern District of New York. [...]

The cache includes data related to marketing campaigns, human resources, strategy, and project management. Also included is pricing from 7.2 billion competitor flights, an estimated 7.5 billion passenger transaction records going back to 2008, and pricing curve data

Google’s purchase also includes about 30 million lines of code, development metadata, and software models and algorithms. The purchase also includes more than 175,000 employee records dating back to 1986.

The data will be anonymized so the personal info of customers is scrubbed:

Google’s purchase doesn’t include personal data and privileged materials like Spirit’s 97.5 million passenger profiles or an estimated 50.2 million records related to the Free Spirit loyalty program data, according to court records.

The data will also go through a process to ensure that it can’t be associated with particular customers, according to the court filing.

Spirit didn’t accumulate 100 million emails, 500 million Teams chats, billions of transaction and pricing records, and 30 million lines of code because it thought they’d someday be valuable to an AI lab. It’s just a byproduct of doing business.

Until recently, there weren’t many buyers on the planet who could do much with a pile like this. But as I wrote about, better AI models can extract more value from the same old data.

This is a gigantic record of how a real company actually operated: how people worked, how decisions were made, how pricing changed, how projects succeeded or failed, how fraud got caught, how software evolved, etc. All this stuff is great for pre-training and RL!

Stuff that was ‘corporate exhaust’ can now have salvage value. 📁🗄️→💰

It may seem like $10 million is nothing because Google is spending hundreds of billions on CapEx. But the point is that the same dataset can be worth radically different amounts depending on who owns it and what they can do with it.

Spirit couldn't extract much more value from these records (obviously). Google may be able to improve products used at enormous scale. Even if the improvement is small, it doesn't take much for $10 million to look cheap.

Not every company's old data is gold. But I suspect we'll see a lot more attempts to figure out which ones are.

Bankruptcies make this visible because everything gets picked over and assigned a price. But I wouldn't be surprised if living companies realize that some of the operational data they've been accumulating for years is an asset they can license or sell.

Every lab has been generating new data, creating synthetic data, and getting access to large private datasets. AI may now be creating an entirely new market for data that was considered to be mostly worthless to anyone but the company itself (and maybe direct competitors).

🔁 Déjà vu: When Old Compute and Data Get Better With Age

From the full feed 🔒: And Just Like That, Tokenmaxxing Is Out 🚮 — don’t get me wrong. Using a lot of tokens productively is great. Targeting absolute numbers of tokens as a metric was always dumb.

🧪🔬 Science & Technology 🧬 🔭

💻 Helium: My New Primary Browser 🔐

I have long been a browser nerd. Ever since I first got Netscape somehow (on floppy disks or some FTP site, I can’t remember), I was hooked. These beautiful apps were my portal into everything I loved. I kept trying every new one in case it was better than what I was using.

I’ve used Netscape, Internet Explorer, Konqueror, Firebird before it was renamed Firefox, Opera, Safari, OmniWeb, Camino, Chrome, Brave, Orion, Vivaldi, and probably a few more I forget.

Some I just tested for a few days, others I ran for years.

But recently, when I saw that my friend Mykhailo was using Helium, I knew I had to try it. I trust his taste in technology more than pretty much anyone else’s.

And I wasn’t disappointed! Here’s the 101:

• It’s built on Chromium, the open-source browser project that Chrome is based on. This means Chrome extensions work.

• It’s open source and stripped down. It’s trying to avoid cramming a million features into the browser. So it’s more or less a de-bloated and de-Googled Chrome. No nagging sidebars and random AI features built in.

• It’s privacy-first. It doesn’t collect your browsing data, it tries to make it harder for websites to fingerprint you, and it comes with ad and tracker blocking by default.

• Oh, it’s free, of course.

The screenshot above shows the split view, which Chrome also has but I didn’t know about until I saw the feature in Helium. You can basically have two tabs side by side in a single window. I know, I know, you can get almost the same effect by putting two browser windows next to each other. But I find this better organizationally because everything stays inside the same browser window, and it doesn’t mess up the way my windows are arranged on my desktop.

Helium is still in beta, doesn’t have a mobile version (though it has MacOS + Windows + Linux on the desktop) or account sync yet, and doesn’t support Widevine DRM (so some DRM-protected streaming services may not work).

One risk with smaller browsers is delayed security updates.

So far, Helium seems reasonably good on that front. It has been shipping new releases frequently and generally tracking Chromium pretty closely.

If you're a browser nerd like me, or just feel like your current browser has gotten kind of bloated (statistically, it’s probably Chrome), give Helium a try.

🔌⚡The Gigawatt Queue: Revealed Demand vs Paper Demand

When supply is tight, it can become hard to know what demand really is. Customers start panic-buying and double-ordering from multiple suppliers, hoping that one of them will be able to deliver.

This is probably happening with memory and SSDs to a certain extent, and there’s also a similar dynamic when it comes to the power grid.

Let’s look at Exelon. In May, it was showing 18 GW of “high probability” data-center load (as of Q4 2025), plus another ~43 GW of future potential additions.

By Q2 2026, those two buckets had fallen to ~11 GW and ~25 GW.

Wait, demand fell? People cancelled data centers?

Not exactly.

Basically, Exelon made the option more expensive. 💰

It started making developers put more real money to keep their place in the queue. Through what it calls ‘Transmission Security Agreements’, developers can be required to post collateral and make other financial commitments. If the project disappears or uses much less power than promised, the developer can be on the hook.

When grid connections take years, there’s a pretty big incentive to reserve a spot in the queue for 200MW or 500MW or whatever, especially if the cost is low. But when the price of the option goes up, you find out who really meant it.

🔐🤖 Mythos: 60 Hours to Kill a Post-Quantum Cryptographic Candidate

Anthropic’s Mythos just did something pretty cool: it found a weakness not in the code implementing a cryptographic algorithm, but in the math of the algorithm itself.

HAWK was a candidate for a new post-quantum digital-signature standard and had already survived two years of review by expert cryptographers.

Anthropic published Mythos’s attack. The next day, the HAWK team withdrew it from the NIST competition because fixing the weakness would basically kill the advantages that made HAWK attractive in the first place.

What’s cool/scary is how fast it happened. Mythos spent about 60 hours working on HAWK and found a shortcut that dramatically reduced its effective security. For the smallest version, an attack previously estimated to require around 2^64 work fell to about 2^38.

60 hours of Mythos may be an expensive API bill for you and me. But for a nation-state or a large criminal enterprise, it’s not that much money. Right now Mythos access is limited, but more Mythos-class models are coming and will become more widely available.

Mythos also found a new attack on AES, probably the most scrutinized encryption algorithm in the world.

Thankfully, it did not break AES, because so much of the internet’s encrypted traffic uses it (including a huge fraction of HTTPS connections).

Mythos attacked a deliberately weakened 7-round version of AES-128, which normally uses 10 rounds. But it still improved the best previous attack by something like 200-800×.

This obviously doesn’t mean Mythos is about to break full AES.

And to put even that in perspective, the 7-round attack is still completely impractical: Anthropic estimates it would cost hundreds of millions of dollars to execute. 💰💰💰

But what happens when someone lets Mythos loose for 6 months on AES? Or Mythos v6 or v6.5 when it comes out someday?

From the full feed 🔒: Micron and Meta: The Slower Memory Was 3.45× Faster — this may become especially important for long-context and high-concurrency AI inference.

🎨 🎭 The Arts & History 👩‍🎨 🎥

🎸 The Guitar as an Evolved Artifact

I found this video because my son, who has been learning piano, picks up my guitar once in a while too and is learning a few songs. He asked me where the guitar came from and how it became what it is today.

What I like about this is seeing how the guitar slowly became the guitar. Nobody sat down and designed the whole thing at once; people kept changing bits of it over generations. It’s very cool to hear the different instruments along the way and learn why various features evolved the way they did.

Liberty Workbench is an occasional series where I put one idea on the bench and think it through with you.

Byrne Hobart (friend-of-the-show and long-time supporter 💚 🥃) wrote Convergent Evolution in the Architecture of Intelligence, and it sparked many thoughts.

The line from Byrne that got me thinking:

Effectively deploying a fixed amount of intelligence means having a relentless focus on maximizing the amount of stupidity you can get away with.

He uses a coffee-pouring robot as an example. You need a sophisticated AI model to understand the high-level instruction (pour me a cup of coffee) and figure out what needs to happen. But you don’t want that expensive model micromanaging every tiny adjustment of the robot’s arm. ☕🦾

Instead, the intelligent system decomposes the complicated job into simpler and simpler tasks until a lot of the actual execution can be handled by dumb control loops and ‘if/then’ rules running on cheap hardware.

The smart part figures out where it can get away with being stupid.

Those Jacquard cards at the top are an old version of that: encode the pattern once, then let the loom repeat it mechanically.

That connected with something I’ve been thinking about lately: better intelligence making existing assets more useful.

In that piece, the idea was that better intelligence can make old capital more useful. The other side is that some inference leaves productive assets behind: spend the expensive, flexible intelligence once figuring out what’s going on, and whatever it discovers can be reused thousands or millions of times.

A flow of cognition becomes a stock of capital. 🧠 → 🏭

The Best Use of Intelligence Is Making Itself Unnecessary

You start with something expensive and flexible, a big, complex frontier model reasoning through a problem, trying things, making mistakes, figuring out what matters. 🤔💭

But next time, maybe you don’t need the big model to reason through the whole thing again. You can bottle some of what it figured out in something much cheaper. 🫙

Maybe it writes some code or builds a little tool. Maybe it figures out a rule or teaches a smaller model. Whatever. 💾 🛠️🧰

frontier reasoning → discovered structure → code/rules/smaller model → cheap repetition

That’s kind of what humans have always done. The math wizard thinks hard about something for years and comes up with a formula. The engineer makes countless attempts and figures out what works by trial and error.

A few steps later, millions of people can benefit from that thinking without giving it much thought, or even understanding it: click around in software, use a machine, whatever. How many hours of deep thinking are embedded in, say, an iPhone? 📲

AI doesn’t change the underlying pattern, but it should accelerate it A LOT. The same system doing the thinking can also write the code, help train the new specialist model, or help design the machine that captures what it learned.

There’s already a pretty literal example of this happening. Google DeepMind’s AlphaEvolve found a simplification in a highly optimized matrix-multiplication circuit for an upcoming TPU. Jeff Dean described one of its circuit designs as “so counterintuitive yet efficient that it was integrated directly into the silicon of our next-generation TPUs.”

That’s intelligence being compiled into machinery about as literally as you can get. 🧠 → 💾 → ⚙️

And you skip a lot of the human handoffs and back-and-forth that slow big organizations down.

I recently wrote that software used to amplify labor, while AI is making software itself into a kind of labor. Give an AI agent access to tools and the web, and it can do an increasing amount of white-collar work.

Then the loop can close:

software → labor → software 🔄

The AI does a job, figures out which parts actually require judgment, and writes software for the rest. The Smart Thing makes a Dumb Thing that replaces it. At least for that particular job.

In the words of Alfred North Whitehead: Civilization advances by extending the number of important operations which we can perform without thinking of them.

So far, that machinery is mostly digital, but eventually more of it will be physical too.

You might spend a bunch of expensive frontier tokens to figure out some annoying task and build a 200-line utility.

The utility might run thousands of times on an old CPU and cost basically nothing by comparison.

Obviously, this only works if there’s some real structure there to find. But I suspect that the smarter a person or AI model is, the better they are at finding it, even in complex systems.

Not everything is equally easy to ‘compile.’ The best candidates give you a cheap way to know if what you built works. They also tend to repeat a lot, change slowly, and can be broken into smaller pieces.

Code is a great fit because it has tests. Robots have sensors that provide immediate feedback. Some repetitive business processes have outputs you can measure.

But fuzzy stuff like ‘make a great strategic decision’ is harder to turn into a dumb loop because you may not know for a long time if the decision was good (and even when it turns out well, it could have been luck or other factors…).

And this isn’t a fixed boundary. As reliable verification gets cheaper, more cognition becomes compilable.

There’s also a break-even point. The dumb solution isn’t free to build or maintain, and dumb solutions can be brittle. That’s partly why they’re cheap: they don’t stop and rethink everything when the world changes. But the danger is when they don’t fail in an obvious way and just keep doing what they’ve been doing even after the world has changed. 😬

So there are two forces pushing in opposite directions here. Smarter models should keep finding more things that can be compiled. But if inference gets cheap enough, sometimes why bother? Just let the smart thing think again. ↕️ 🤔

Back to Byrne’s coffee robot for a sec. Once the intelligent model has decomposed the job enough that simple control loops can handle most of the execution, the intelligence doesn’t just have to sit there idle. It can migrate upward.

If you don’t have to control every tiny movement, you can spend that intelligence on exceptions and weird edge cases instead. And eventually some of those may stop looking so weird. You find the pattern, compile that away too, and move up again. Rinse and repeat.

AI Can Cannibalize Its Own Future Inference

Once a task has been compiled away, the expensive intelligence can go looking for the next one. So you can need less and less intelligence for each thing you’ve already figured out, while using more and more intelligence overall. 📉📈

I used to picture sci-fi AI as super-smart and constantly thinking about everything, running everything. But maybe that’s backwards. A fantastically intelligent AI might spend a lot of its time figuring out where it can get away with stupid automations and save the compute for harder stuff.

Maybe a much more intelligent world ends up full of much stupider machinery, because intelligence has already been compiled into it. ⚙️

🔁 Déjà vu:

• Edition 638 — software cycle times collapsing from months to hours, turning speed from quantitative to qualitative.

• Edition 605 — intelligence as a universal input: when it gets cheaper, total consumption rises.

• Edition 627 — intelligence migrating toward problems humans under-serve or prefer not to work on.

• Edition 647 — cheap implementation turning prototyping and iteration into the new bottleneck.

Liberty Workbench is where I put one idea on the bench and think it through. Every Workbench edition is collected here.

Things that have never happened before are the most mispriced things in markets.

–John Burbank

💻🔍🤖💭 If there’s one thing that AI is good at, it’s aggregating the nerdy wisdom of millions of Reddit commenters and sysadmin neckbeards.

Every time I need a new software tool, I have a discussion with an AI so it can give me a few choices, the pros and cons of each, and a recommendation based on what it knows about me. I make sure to explain my use case first, so it can target its recs even better.

So far, it has steered me well. I’ve discovered software I probably wouldn’t even know exists otherwise.

So I decided to take that to the next level.

I gave it screenshots of my full Mac Applications folder and asked it to review each app and tell me if there’s a better alternative it recommends, and why.

I installed many of these tools years ago and hadn’t really kept up with that particular space since. Some were outdated, some were no longer maintained, and others had been outclassed by a Shiny New App™ that does things better.

For example, I have been using Pastebot as a clipboard manager for a loooong time. It’s great and does the job. But as I updated my app launcher from Alfred to Raycast, the AI suggested that I could consolidate to Raycast’s built-in clipboard manager and have one less app loaded in memory. So far, I’m loving it.

I removed dozens of obsolete old apps, replaced a few of my main apps with better ones (I didn’t even know IINA existed. I’ll also keep VLC, but I’m glad I learned about it). In some cases, the AI just confirmed that I was already making the right choice (Fluid Voice, Helium browser, Pixelmator Pro, GraphicConverter, GrandPerspective, Mimestream, etc).

I think this audit of your apps is a pretty high-leverage thing to do. You may use these tools daily or for important work. Leveling up your tools is a one-time effort that can pay dividends for years to come.

Next step: I’m doing the same with all my phone apps. 📲

💚 🥃 The paid subscription includes the full feed, every podcast, Zoom Q&As with me, and access to the private Discord clubhouse, where a core group of us hang out and discuss the interesting stuff we find:

🏦 💰 Liberty Capital 💳 💴

Jensen Has Found a New Complement to Commoditize: Capital 🏦

Nvidia has spent decades driving down the cost per unit of compute. To do that, Jensen and his team have moved from one bottleneck to the next. Let’s widen memory bandwidth, let’s add more cores, let’s improve networking, let’s buy up a ton of wafer and memory capacity, let’s invest in the silicon photonics supply chain, etc.

For years, most of the bottlenecks Nvidia attacked were technical or physical. Now they’re going after a financial one.

They see that the AI wave is starting to hit balance-sheet bottlenecks. 🧾

Jensen posted about the new financing mechanism Nvidia is trying to build:

The plan is to team up with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to try to gin up more than $500bn of third-party money for AI infrastructure.

The interesting part is that Nvidia may also provide residual-value support for up to 25% of some deals. Basically taking on some of the risk that those very expensive GPUs could end up being worth less than lenders are assuming.

From Nvidia’s POV, the logic is pretty simple: if expensive financing keeps customers from building more AI factories, then financing is just another complement that Nvidia wants to commoditize.

The all-in cost of producing tokens in a $5bn datacenter isn’t just the GPUs, land, shell, electricity, cooling, etc. That cost also includes the interest paid on the debt used to finance it.

Now Jensen is trying to make the dollars themselves cheaper. 💵📉

Why is Nvidia willing to take some of that risk?

Jensen’s argument is that Nvidia GPUs keep their value for a while and are easy to find a new home for. If one customer doesn’t need them anymore, there are lots of other customers, clouds, and workloads that probably will. And CUDA keeps making the installed hardware more useful over time (through backward-compatibility and constant optimizations).

Basically, Nvidia wants to make its GPUs better collateral.

And Nvidia has a large information advantage over the banks: Jensen knows the roadmap (he’s the one writing it ✍️). So they should also have a better idea of what today’s GPUs will still be worth a few years from now. Of course, it’s about as far from a neutral appraiser as you can get, but that doesn’t mean it can’t pull this off.

The irony here is that nobody is working harder to make today’s Nvidia GPUs obsolete than Nvidia itself 😅

Better collateral → cheaper financing → more AI datacenters get built → more Nvidia GPUs get sold

That’s the plan, anyway. We’ll know it’s working if comparable Nvidia-backed AI datacenters start getting financed more cheaply.

✨ Constellation Software’s Dark Matter ✨

Random thought: As they grow and keep adding business units, will they ever change their name to Galaxy Software? 🤔

The quarter came out last night. I haven’t had a chance to look at it in detail or read this morning’s transcript yet. I’ll probably post more about it later. But for today, a couple of points that many CSI observers tend to forget.

First, on organic growth:

And on the increasingly material amount of ‘dark matter’ hiding in the star system, I like this explanation by friend-of-the-show Leandro (💚 🥃) at Best Anchor Stocks:

Contrary to many believing that revenue is a “clean” number, Constellation’s total revenue number also needs context. For starters, the company is deploying capital into opportunities that “entirely” bypass the consolidated revenue figure. I explained this in my most recent Topicus article, but Asseco is being reported through the equity method, meaning that it’s not consolidated in revenue but the company’s share of its profits comes included in “share in net (income) loss of equity investee.” What this ultimately means is that Asseco is indeed present in Constellation’s net income figure (after deducting Topicus’ non-controlling interests) but is not consolidated in the revenue line. Something similar happens with Sabre in the sense that revenue bypasses the revenue line. Sabre, however, is not accounted for in the same line as Asseco but rather in “finance (expense) income.”

Topicus deployed around €400 million in Asseco (around $500 million at current exchange rates) and Constellation deployed around $90 million in Sabre. This ultimately means that Constellation has deployed around $500 - $600 million into acquisitions that are entirely bypassing the company’s revenue and therefore not contributing to revenue growth (even though these are indeed having an impact down the income statement). There’s nothing wrong with this, but I believe it’s something important for people to consider as the relationship between capital deployment and revenue growth will not be perfectly linear if Constellation and its universe continue using these capital deployment options.

Most investors never have to deal with stuff like this. More than a thousand business units, partially owned spin-offs, increasingly large equity investments in public companies (some in ‘exotic’ places like Poland 🇵🇱). 😵‍💫

Once again, it depends on what your goal is.

If it’s to give the cleanest metrics to Wall Street, then stop with the minority equity investments and stop buying hairy assets that require restructuring and shrink. Even if you can get them super cheaply and produce great returns like this:

You need to click to see the full tweet. The hidden part is what I want to highlight:

But if the goal is to create shareholder value, then the messy stuff is likely to be the most mispriced, where the opportunities are juicier.

🔁 Déjà vu: Constellation Software’s 2026 Annual Meeting

💰 Intel Raises $20bn in Equity (Foundry 🟢)

Intel initially announced a $15bn equity raise, then upsized it to $20bn. Bloomberg reports that the deal drew more than $100bn in orders.

Nobody wanted Intel’s equity in August 2025 when it was about $19/share, but today Intel can sell $20bn of it at $95/share. The ducks are quacking. 🦆🦆🦆

But it makes sense. CPU demand is through the roof thanks to AI agents that can do a crapload of tool calls in the time it takes me to make a cup of espresso, and a bunch of the big capex spenders would love to have an alternative to TSMC (or at least a Plan B).

14A itself is already a go: Intel said in July that it had committed to high-volume ramps in 2028, and Tesla is already a customer. Intel has also joined Tesla and SpaceX on Terafab, and Musk said Tesla plans to use 14A for chips there.

And Foundry isn’t only about making wafers. Intel says customer interest in its advanced EMIB-T packaging is “very high,” with a growing backlog and customer ramps coming in 2027.

14A is special because it’s Intel’s most credible attempt in years at retaking process leadership from TSMC. They got the world’s first commercial High-NA EUV machine from ASML, and last month became the first chipmaker to ship high-volume logic using High-NA, on some 18A layers.

Intel claims that 14A will deliver 15–20% more performance at the same power, 25–35% less power at the same performance, and up to 30% greater density versus 18A.

The naming is super convenient too, because Intel’s 14A is the direct competitor to TSMC’s A14. They’re both aiming for 2028 production.

But this equity raise makes me think that there’s probably another big customer lined up… Nvidia? Apple? Google? We should find out soon.

⚛️ The Biggest Nuclear IPO Yet? 💵

Westinghouse has confidentially filed for an IPO. Brookfield first bought Westinghouse out of bankruptcy from Toshiba in 2018. Cameco acquired a 49% stake in 2023 and Brookfield kept 51%.

This is a little different from many of the nuclear companies that have gone public recently, which are bets on new reactor technologies and future deployments.

Westinghouse already services ~63% of the world’s operating reactors and fuels ~40% of them. That’s real concrete and gigawatts ⚡

Trivia:

It was founded in 1886 by a Civil War veteran, George Westinghouse, whose first major invention was a braking system controlled by compressed air that revolutionized rail safety. By the late 1880s he had moved into the electricity field. His company was a competitor to Thomas Edison’s

But they’re not just about the past. The future looks pretty good in this world that is now hungrier for gigawatts. There are already 14 AP1000s under construction, another 7 contracted, while Westinghouse says its broader pipeline includes up to 91 potential reactors. 🏗️🏗️🏗️

🤳 AI May Massively Expand Meta’s ‘Taxable’ Surface Area

Friend of the show and supporter Byrne Hobart (💚 🥃) writes:

Mark Zuckerberg has identified an opportunity for political entrepreneurship, laying out the positive case for AI, and doing so with enough details that, if he successfully sets the narrative, it makes the other labs look bad

It’s always hard to know exactly how much is deeply held belief and how much is strategic positioning, but I’ll take it.

I think a resurgent Meta AI could be a very positive force in the field, both to create open-weight alternatives to Chinese models and to keep Anthropic and OpenAI on their toes (especially now that Google appears to be falling behind…).

Meta is in the business of filtering things you don’t know about in order to show you things you didn’t know you wanted. That applies to organic content, and it applies to ads. They’re also in the business of adding a legible digital layer to interpersonal interactions [...]

Meta wants a world where there’s more competition for attention, because that’s a world where their filtering ability is more valuable to users and the resulting captive attention is more valuable to advertisers. They also want a world where there’s more economic growth in the long tail, and where small businesses have similar production capabilities to big ones, so that they are willing to pay Meta more for distribution.

If AI helps create tons of new content AND makes it easier to make video games, apps, physical products, even start small companies, this is a lot more monetizable surface area for Meta.

All these things need distribution. The AI won’t provide that (or at least, whatever it does to help will be available to everyone, making it just as hard to stand out, like Buffett’s standing on tiptoes at a parade metaphor).

If creation becomes more abundant, attention becomes even scarcer and valuable by comparison, and Meta controls a lot of where that attention is directed. To use Byrne’s phrasing, Meta is one of the biggest toll collectors on attention.

From the full feed 🔒: When Old Compute and Data Get Better With Age — why some aging hardware and datasets become more useful as the systems around them improve.

🧪🔬 Liberty Labs 🧬 🔭

📄🔍🤖 AI Is Starting to Recheck What We Think We Know

On Friday, I wrote about how in the AI era, old data, old code, and even old hardware can become more useful as the intelligence available to work with them improves.

Nature has a great example of this thing happening to old scientific knowledge.

Sebastian Pios, a theoretical chemist at Zhejiang Lab in Hangzhou, China, was using an AI system to predict the boiling points of several molecules when it began producing values that clashed with long-accepted entries in a 75-year-old reference database. At first, he thought the model was wrong. But when he manually checked the original literature, he found that the reference data were wrong, not the AI model.

Pios’s model found two more errors in old papers and reference books, including a typo and incorrect values from century-old boiling-point measurements.

And we can probably expect a lot more of this kind of thing to come.

That’s because science is especially well-structured for errors like these to propagate AND for AI to find them.

There are probably lots of places where parts of the scientific record disagree with each other and nobody has noticed, because humans simply don’t have time to follow all those threads and double-check every single figure in every single textbook and paper.

But AI is increasingly good at that stuff, and getting better rapidly.

The scientific literature isn’t just a giant pile of papers. It’s all connected through citations, reference values, reused datasets, formulas, experimental assumptions, databases, etc.

Scientists can only look at small parts of that network at a time, but AI can look across much more of it and find places where two things we thought were true don’t agree with each other.

The researchers behind PaperQA2 built ContraCrow, a system that searched biology papers for contradictions, and human experts agreed with 70% of a sample of the contradictions it flagged (which is a good sign).

That’s a much more interesting use of AI than summarizing papers we already know about. It’s actively looking for places where the scientific record doesn’t quite make sense.

You could call it, uh, computational skepticism. Constantly asking the scientific world, “Are you sure about that?”

And this doesn’t have to be a one-time pass. As the models get better, you can just keep running them over the scientific record again and again:

2026 AI → check it

2028 AI → check it again

2030 AI → check it again

🔓 🔄 🔐 Cybersecurity Reflexivity

Since I wrote The Thinking Box, there's been a lot of news about models doing all kinds of very sci-fi things, getting out of supposedly sealed evaluation environments, trying to cheat by exploiting previously unknown vulnerabilities, even a swarm of agents creating an ad hoc message board to leave notes for each other.

The Hugging Face-OpenAI hack got a ton of attention, and that made everyone else look more carefully. Anthropic went back through its logs and found three incidents of its own. Since then, other incidents have surfaced involving Meta and Moonshot’s Kimi K3.

I don't want to rehash all this. I'm sure you've seen it.

I just wanna discuss one aspect of all this that makes me a bit more optimistic: The system is reacting to the warning signs. 🚨⚠️🔊

There’s a reflexivity aspect to this: Concrete evidence of the risk, the fact that it’s no longer just theoretical, changes what people do. 🔄

It doesn’t make the problems go away. But if we catch these things early and enough people freak out about them, then it should have an impact on what the labs do to prepare, how they improve security and safety, and how they train models.

And hopefully, outside the labs, a lot more of the infrastructure players and sysadmin types out there will take even more seriously their duty to harden their infra and patch everything they can patch. And maybe make use of cyber defense projects like Anthropic’s Glasswing and OpenAI’s Daybreak (which now includes GPT-5.6 Cyber, a new specialized cybersecurity model that is supposedly very strong).

I’m still worried about what can happen as these capabilities diffuse more widely and aren’t just in a few lab tests, but I’d much rather have these wake-up calls now, while the models are mostly trying to win benchmarks, than later when someone is deliberately pointing much more capable versions of them at the real world.

👨‍🔧 Jeff Dean on AI’s Past, Present, and Future 🔮

Great rec by Linas Beliūnas:

From the full feed 🔒: Citadel, Point72, and Other Hedge Funds Targeted by AI Voice-Cloning Attacks — what happens when AI lets scammers clone voices and call hundreds of high-value targets at once?

🎨 🎭 Liberty Studio 👩‍🎨 🎥

🎬 Beatles Biopics (x4!) by Sam Mendes and Greig Fraser 🎶

After I watched ‘1917’ with my boy (I wrote about it here), I was curious what director Sam Mendes would do next. Having also recently watched ‘Project Hail Mary’ (and recorded a podcast about it), its cinematographer Greig Fraser has been rising on my list of favorite filmmakers (he also did Dune 1 & 2).

So I was very happy to learn that Sam Mendes and Greig Fraser are working together on a new project! It’s a biopic of the Beatles, and it’s incredibly ambitious.

They want to make FOUR films, one for each band member, and release them all in April 2028. I don’t know if they’ll pull it off, but they have the talent to make something really special.

The actors will be:

Paul Mescal (Gladiator II) will play Paul McCartney, Joseph Quinn (The Fantastic Four: First Steps) will play George Harrison, Barry Keoghan (Saltburn) will be drummer Ringo Starr, and Harris Dickinson (Babygirl) is playing John Lennon.

The latest news is that they’re actively filming now. Just a few days ago, reports said Sony will do a major shoot at Abbey Road involving roughly 80 extras, period cars, street closures, and filming in front of Abbey Road Studios.

Sony says the four perspectives will “intersect,” but I wonder what that actually means: Will the films show mostly the same events from four points of view, or will each cover different stretches of the Beatles’ story, with the timelines occasionally crossing? 🤔

In any case, I’m so in for this!

The only thing we have so far is those cast photos in costumes (above).

If you want recipes like this, check out CookingForEngineers.com (this isn’t an ad, I just think it’s clever and funny) 🧑‍🍳👨‍🔧