Demis Is Out as DeepMind CEO, Revolut Founder Yacht Drama, AI’s Great Reverse Bank Run | Diet TBPN

What just happened at Google DeepMind? Crazy, crazy morning, uh, shakeup, a reorg, stepping down, stepping up, stepping left, stepping right. There's a lot of different terms. We'll break it down. We'll explain the organization before, now, what it looks like going forward, what this means, where this goes. So this morning, Demis Hassabis, the legendary AI researcher, the founder of DeepMind, a true scientist, a true scientist, uh, the subject of Sebastian Malaby's book, The The Infinity Machine. He announced that he's stepping down as CEO of Google DeepMind. He's also stepping up into the role of Google DeepMind chairman, and he will also be Alphabet's chief scientist. Is it a promotion? Is it a demotion? Is he quitting? Is he getting fired? Who knows? But the most important thing is that he's still there at Alphabet. He has a different role, and that's what's important. What is the nature of this new role? Google CEO Sundar Pichai also announced that Google DeepMind Chief Scientist Geoff Dean will be leaving the company after a 27-year run to start, along with Google Senior Fellow Sanjay Ghemwat, a public benefit corporation to accelerate discoveries in machine learning, science, and engineering. And if you're not familiar with Geoff Dean, he is the Chuck Norris of computer engineering and software engineering. There's a whole list of jokes around Jeff Dean's jokes.

Why don't you read some of them, John?

What are my favorite Jeff Dean jokes? There's a whole bunch. Any opportunity to repost the Jeff Dean Informatica link. Compilers don't warn Jeff Dean, Jeff Dean warns compilers. Jeff Dean's keyboard, it only has two keys, a 1 and a 0, because he programs in binary. It's that type of thing. "Unsatisfied with constant time, Jeff Dean created the world's first O algorithm." So he can bend time. There's all these programmer jokes. I don't know. They're a lot of fun. He knows the last digit of pi, that type of thing. But he's been a legend. These were on Quora maybe 15 years ago, maybe more. People were so excited about him, and for good reason. He invented tons of things.

Jeff Dean doesn't call APIs. APIs call him.

Exactly, that type of thing. The lore around Jeff Dean has been substantial throughout his entire career. He has been very key to so many different Google projects, integral in MapReduce and actually being able to scale Google's systems.

Does this make any sense? Chuck Norris counted to infinity twice. Jeff Dean did it in parallel.

Ah, okay. Yeah, that's kind of funny. It's a programmer joke, but I don't know. But yeah, it's just funny that he was able to have this level of aura at Google as a non-founder, just a great scientist. And it was so long ago, before there was so much attention on talent below the founding team and below the investing team or whatever.

Truly a living legend. Yeah, completely. Google jersey will be going in the rafters.

Yes, for sure.

But he's not done.

He's not done. And he's starting a new company, which is very exciting because there's a lot of opportunity in machine learning, science, and engineering, obviously. So Google DeepMind's CTO and Google's chief AI architect, Korey Kavuchoglu— I can't pronounce that, I'm sorry— is being promoted to SVP and will lead DeepMind. So Korey is now running DeepMind. But interestingly, there's no new CEO of DeepMind, which always seemed to be an odd decision because DeepMind was not its own reporting unit. Like when the, when quarterly earnings happening happen, you get YouTube numbers, you get Google Cloud numbers. And so having a CEO of those divisions made a lot of sense of those organizations. But DeepMind was always a, was always a unit that sort of cut across everything because the work that DeepMind does on Gemini and the models made a lot of sense to then into YouTube. And now there's a little Gemini button when you're on YouTube and you can just go to a 1-hour-long podcast and say, when do they talk about this? And it'll just answer you. And so that was like a feature that was vended into that. Obviously goes into core Google Search. Gemini models help with ads, they help with the actual Gemini products like they get vended everywhere. There's a Gemini button in Chrome. There's also a Gemini button in Google Docs. There's also a Gemini button in Gmail. And so it was, it was never like its own, even though it was its own unit. Like the value that it created at Google was very, was very cross-organizational as opposed to, you know, the YouTube team has a CEO and they have revenues and they have ads and they have a product. And yes, it's integrated into the other things, but it very much is a standalone organization, reports its own financials and has a CEO. So no CEO for DeepMind, at least in the short term. We'll see where it goes. On X, Demis says the move will, quote, allow him to focus on long-term strategy. And accelerating scientific breakthroughs. He's done a lot of that already, uh, including leaning into his work at Isomorpheic to help cure disease. He fleshed this out more on Google's blog. He said, I've decided that now is the right time for me to hand over my day-to-day operational responsibilities at GDM, uh, so that I have the time and space to focus on the big picture and help influence what is to come, uh, uh, to the best of my ability. In my new role, I will continue to work closely with Sundar on strategic and global AGI matters and to advise Corey, Josh, and the GDM leads, uh, from our awesome new London Platform 37 offices. As part of this transition, I will also be leaning into my role at Isomorphic where we are making, uh, extremely rapid and promising progress to accelerate our mission there even faster. As you have heard me say many times, I've always believed the number one application of AI should be to improve human health. That is a great message. Very different from the SpaceX S-1. The number one application of AI is enterprise software.

Harness the power of the entire galaxy.

I mean, there's that, but also just in the actual application, it's like enterprise workflows, which is—

Yeah. So, some type of shakeup has been expected for a long time. Yeah. The DeepMind organization has, you know, made some real contributions to the space, but has been lagging. It's been a point of embarrassment for, I would say, for Google, for, I would say, a year plus. They've always felt behind even though they created so much of the fundamental research that has, you know, gone into this entire chapter of the, you know, and certainly not behind on video.

The video stuff's really good.

Yeah, sure.

But then you get kind of tied up because The Chinese models are like, actually, we will do Mickey Mouse.

Yeah, but here's the thing. At no point has Google been breaking out their true AI revenue in a way that gets shareholders excited. Yeah. And yeah, that's good. What would— yeah. And there's just been so many, so many question marks. There was a moment, I mean, over a year ago, felt like Google had some real momentum. Yeah, but that momentum was very much lost. And so there's this question of You know, when Noam left, which was about a month ago, that felt— that was like more shocking in some ways than even this. This feels like expected. And then to leave what feels like a hole in the organization is pretty, pretty wild. Stock's down. It was down 5% around the open. It's recovered a little bit, just down 3.5%. But I mean, the big thing is like they just started, like Google as a company, I don't think was able to be at least sufficiently AI-pilled, right? They started selling compute. A lot of researchers I'm sure were frustrated with that. And that could have led to some of the historical—

Tyler over there thinks NVIDIA shouldn't sell the chips if they're really AGI-pilled. The much more extreme version of Google should keep the TPUs for themselves is NVIDIA should keep the GPUs for themselves. Yeah, I mean, that was something that was discussed as, okay, if you believe you're in the foothills of the singularity, you will want to scale your compute. If you're seeing traction in the products that you're making, you want to be GPU-rich. You don't want to be selling TPUs to other companies. So, there's got to be some tension, but Google is a big organization. There's multiple organizations within Google and they all have different different KPIs. So one team might say, yeah, my mandate is to sell as many of these as possible externally. Other people might say, my job is to keep as many of these internally. And there's a lot of debates there. So we'll see.

Obviously, Google has made a massive bet on Anthropic and is becoming their largest financial backer, especially through some new reporting on some of the lease guarantees that they're doing. Yeah, but anyway, still so surprising Google Obviously one of the greatest companies of all time, has every possible advantage in this race and yet has just consistently struggled. Susan Zhang, who is currently on leave from DeepMind, posted a little somewhat of a fable. Backdrop: Kings all asleep/on vacation while minions slave away. August 2024, first big fail. Fast forward January 2025, only competent VP bails after one terrorist lead demanding an even bigger slice of pie breaks the camel's back. Summer 2025, a year ago, second big fail and king of terrorists climbs the ranks above the slumbering execs.

This is a crazy summer.

2020. Summer 2026, slumbering execs and terrorists who only show up for successful announcements move on. Leaving just the king of terrorists behind. Lessons: If you're high enough up there, you can pretty much be a terrorist and hold orcs hostage while continuing to ask for more and more and more until there's absolutely no more aura or gold left to plunder, and then just move on to greener pastures to do it all over again. And we'll leave it up to you.

This is a crazy, crazy post.

Try to figure out who's who in this story. Doesn't strike me that Susan will be headed back to the company after this post. This is a crazy, crazy post. She says, I've been on leave since January. Got pretty burned out by certain people who just say yes to everything to avoid ever making a hard call and leave it all to the hungry minions to backstab each other until success finds a cursed hole to crawl out.

To be fair, I was looking at Tyler's alt account, and he talks about us like this all the time. He's always talking. He refers to you as the slumbering exec and me as the terrorist. The terrorist. I mean, I believe in free speech. I defend your right to say these things, Tyler. I am offended, but, you know, you're going to— Nobby, nobby. No, it's a wild, wild thing.

One of the crazier subtweets from a big tech.

Yeah, but I don't know. I was never sure that she worked there. It felt like it was maybe like an alt account. And it was just like an anon, but it does seem like, I mean, she was saying right here, I've been on leave from this organization. So I don't know. And somebody calls her the Homer of our times. Another interesting wrinkle here is if you look at what Demis was talking about, writing about just a few weeks ago, it wasn't as much focused on the isometric scientific advancements. I know he's super deep in that, and that's what he's said he's been focusing on, and I love that message. He says, "It's time for AI to prove its unequivocal value in the world, and what better way to demonstrate that than to help finally cure diseases like cancer." 100% agree, 100% agree that if you get a new cure for cancer, you save someone's life and they can go and say, "Yes, I took the new cancer cure and I was cured. AI is doing its job." A lot of people will be very happy about that. Perfect. But that's not what Demis was writing about a couple of weeks ago. The last substantial piece that Demis put out was called The Framework for Frontier AI and the Dawning of a New Age, in which he argues for a frontier AI standards body that evaluates frontier models before they're released for risks associated with superintelligence. It's at the very least not crazy to assume that Demis's new role at Alphabet is maybe related to what he's advocating for in this essay. There's been a lot of back and forth in Washington about the interpretation of that essay, and a lot of AI researchers have stepped up and said, like, yeah, we basically agree with this. This framework seems pretty good. It includes open source models in there.

Yeah, to me, Demis could be an amazing candidate as somebody to lead a new regulatory body.

Lead America as the president. You think he's running?

Not quite, but I think we got a researcher in there at this point.

Be great.

Yeah. No. So big question now. So Jeff Dean is out. Noam Shazir is out. Demis is out. Who is going to fill that void from a leadership standpoint? How are they going to be able to recruit the best researchers in the world? That's a big question mark. I wouldn't be surprised to see Google just shift into more of an NVIDIA-style role as like a capital partner to a number of other labs as they, you know, continue to fall.

It's, it's maybe more of like an Amazon and Microsoft role where you have a great business, you're also great at building infrastructure, you also have a semiconductor team and a chip team, and you can play up and down the stack, but maybe you're not as, as focused on this like full stack straight shot thing and you're partnering with other labs on, on different pieces of the puzzle. It'll be interesting to see where everything goes.

Tenno Bruss says Demis is ousted as DeepMind CEO and Jeff Dean and Sanjay are leaving Google to start a new lab. They're all being very careful to frame these as positive shifts, but there's no way in heck Demis would have accepted this willingly, and there's no way Sundar happily accepted Jeff doing this as a totally independent new public benefit corporation rather than a bet under Alphabet. Tough to see an interpretation other than Jeff losing confidence in working on AGI under Google. Very bad.

We need a Sebastian Malaby sequel. We need a sequel.

AlphaBet overall.

I would love for him to write a sequel to just— and even zoom out and look at the AGI wars generally, like he did in The Power Law about the early venture capital firms and how they all formed and battled each other. That would be great. Anyway.

Rohan Anil over at Core Automation says, pour one out for the big G. It's so over. We did have a funny moment with Google AI Overviews a couple days ago where it said that one of our friends was married to another of our friend's wife. And it was quite confident in that.

It was.

And I have no idea how it even thought of that.

Yeah, I think it was because they did a podcast together or something like that.

Like they were linked in a blog post together, something like doing a podcast together.

She wrote about him or something or wrote a profile and it just like kind of put the names together. Very odd. Yeah, I don't know. Yeah, you have to wonder what it takes to get AIOverviews into like a reasoning mode that can actually do the fact-checking so the hallucinations die down because Well, yeah, they have to figure out a way to make that process so inexpensive because they're serving it. Yeah, billions.

Yeah, billions. Nathan Lambert says major restructuring at Gemini will be studied forever as the incumbent with all the advantages not being able to get going.

They got going. This is, this is not correct. They completely got going.

They got going, but they certainly didn't build real momentum. And then he says, P.S., OpenAI accomplished their original goal.

Lucas Byers says, P.S., of the century.

But yeah, the backstory is that a lot of the initial energy around OpenAI was a concern that Google would control AGI and a fear from the various early participants around that scenario.

Yeah. Interesting.

And almost even more positive news, the moment you've all been waiting for, CarPlay is coming to pontoon boats.

Let's go.

Finally. Finally, finally, finally.

Have you ever been on a pontoon boat?

Yes.

Explain. What, what context were you on a pontoon boat? What does a pontoon boat even look like? This is a terrible—

it's for cruising. It's for cruising. Team, pull up a pontoon boat.

It's not, it's not like a monohull, but it's not a catamaran.

It's something between like a cigarette boat and a houseboat.

Okay, that's a really wide gap.

Well, it's right in the middle.

Mastercraft Boat Holdings today said it's adding CarPlay to its upcoming Crest and Belize pontoons. The company is working with Savvy Navy to bring CarPlay and on-water navigation to boaters.

And Mastercraft Boat Holdings Inc., of course, a public company experiencing a god candle, really up 0.72%, up $0.18.

This is cool. This is very cool. See?

Yeah, this is a double-decker.

Double-decker. See, out in California, we don't know anything about these pontoon boats because these don't work at the beach, right?

Correct. But I was sort of correct, somewhere in between a houseboat and a—

Aren't you from the land of a thousand lakes, right? Michigan? Isn't that the land of a thousand lakes? No, me. Minnesota. Oh, you're from— OK. Are you a pontoon boat guy?

Pontoon boats are huge in Minnesota.

Minnesota.

Okay.

Yeah. How many—

they're for lakes.

They're for lakes, basically. How many hours have you spent on pontoon boats in your life?

10,000.

10,000 hours? I'm an expert. The man, man, put in life's work on the pontoon boat, apparently.

You're in the wrong job, buddy.

I think, I think you gotta become a pontoon boat captain.

It is crazy to think in another life Ben could have been the top pontoon boat salesman in the entire world. Moving over to—

okay, we got—

what is the story of the day in technology?

Startup billionaire sued over superyacht broker fee. Everyone's talking about this. This is potentially bigger than what's going on at Google today. Uh, in London, the billionaire founder of one of the world's most valuable startups recently bought a superyacht valued at an estimated $400 million. Yeah.

Continue.

Replete with a glass-bottomed swimming pool and a helipad. Do they make— wait, do they make super pontoon boats? Can you get a $400 million pontoon boat? What does that do for you?

If you— Almost certainly.

Can you imagine just a 300-foot pontoon boat?

The world will create infinite ways to make you part with your hard-earned money.

Yeah.

And I'm sure we could arrange a— a almost half a billion dollar pontoon boat.

Okay, team, AI images. I want to see what $400 million gets me if I'm going all in on a pontoon boat. I want to really dominate Lake Michigan. Is Lake Michigan too big for a pontoon boat?

Yeah, I think the pontoon boat is not for like big waves.

Big waves.

You can't, it can't be like—

But there's not big waves on Lake Michigan. You can't surf or anything? I don't think so. Surf. Yeah. Anyway, so Uh, it was a $400 million superyacht. It had a glass-bottom swimming pool and a helipad. There was just one problem. According to this reporting in the Wall Street Journal, he didn't pay the broker's fee. That is according to a top luxury yacht broker who is now suing Nick Staronsky, a co-founder and chief executive of digital banking giant Revolut, alleging the entrepreneur went behind their back to avoid paying commission on a purchase they had worked on. Cecil Wright and Partners, a luxury yacht broker, filed the legal action in London's High Court over Staronsky's acquisition of Nixie, a 335-foot vessel with 4 terraces. The broker alleged in a court filing that it was negotiating the purchase for Staronsky when his team secretly contacted the yacht seller directly and agreed to buy the boat, avoiding roughly $20 million in standard commission, the amount the dealer now claims it is owned— is owed. In January, Staronsky's team allegedly told Cecil Wright to hold off, hold off on contacting the vessel seller, Canadian billionaire and former hockey player Patrick Dovigi, and said they would talk again the following week, according to the filing. The filing claims that by next week, Staronsky's team had already arranged to buy the boat directly from Dovigi. Bypassing the broker entirely. It was Cecil Wright that brought the yacht to Staronsky's attention and introduced the Revolut co-founder and the owner.

Matt Young says, unbelievable, taking food off the table of that luxury yacht broker's family.

A spokesman for the Staronsky family office said, we are aware of the claim. It is without merit and it will be defended. He says not guilty. Nixie features a private wellness spa, indoor and outdoor cinemas, two movie theaters. Now we're talking. As well as a Japanese teppanyaki grill, according to Boat International.

Pull up this picture that Tyler just shared. I think this is what Nixie looks like.

Yeah, let's see it. When chartered, it is equipped with a variety of water toys, including a luxury sea bob water sled and inflatable blob launchers. Is this the Bondoon boat? The luxury pond?

Oh, sorry. That's— That's the $400 million pontoon boat that Tyler just generated.

That's pretty good. I like it. I like it. Uh, Staronski is one of Europe's highest-profile startup founders. Founded in 2015, Revolut is now the region's most valuable startup with a recent share value, uh, share sale valuing the digital bank at $115 billion. Wow, Revolut, what a tear. Originally built around low-fee foreign currency transactions, Revolut is widely used by international travelers and migrants. Sending remittances home. The company gained a full UK banking license in March after a lengthy battle. Wow. It took them over a decade to get a banking license. That is crazy. But they did it after a lengthy battle. Now they're looking to expand into the United States, having recently applied for a national banking charter. It's crazy that they can't go faster. Erebor got a charter so fast. And then there's been other— there's been other companies that have explored like buying a regional bank and then using that charter. Why?

It's almost certainly part of why Revolut is now such a massive company. Yeah, because they figured out they don't have infinite competition in the way that—

true—

many, many banks and neobanks have here in the US.

Earlier this year, Revolut reported record annual profits and a jump in its customer numbers. At the same time, Revolut said it remained focused on its goal of increasing its customer base to 100 million users by mid-2027 from more than 70 million currently. Well, good luck in the luxury yacht battle. Hopefully everyone gets what they're owed and everyone is compensated fairly. Hate to see people fighting over that. AI's great reverse bank run. In today's newsletter, Joe Weisenthal wrote about Leopold's Basilisk and what makes this particular mania feel distinct. People are betting on left-tail outcomes without sacrificing the right tail of financial opportunities. Doomers are getting fabulously wealthy. Let's see what he's saying. Here's the deal. Historically, if you want to cut off your left tail, you also have to cut off your right tail. Or to put it more specifically, if you are up big on a stock position, you want to hedge your gains by buying puts, you can do that. But if you want to maintain your puts over time, you spend a lot of money and erase any upside gains that you might otherwise be making. Or to put it even more generally, suppose you're afraid of the outbreak of war? Well, you can move to a rural farm and grow your own food, but then you miss out on the economic opportunities afforded by living in the city. People make these kinds of trade-offs all the time depending on what's worrying them. Pin that thought for just a second. One of the big questions with AI is whether it's just another technology or whether it somehow will be different this time. It's purchasable intelligence, kind of like purchasable electricity, or is it something more like a sci-fi novel where Humans get replaced or eliminated in some way as businesses, the big American companies selling AI, mostly operating as if AI is just another technology to be developed and solved. But we know that there's deep anxiety within the companies themselves that AI could play out in the way sci-fi novels have expected for decades. And this sometimes is reflected in their governance structures. Here's one stylized fact though that might argue for the it's different camp. If you are one of those people who's been reading AI philosophers and message boards for the last decade, familiar with paperclip maximizers since long before ChatGPT came out. You might have deep existential fears about something bad happening right around the corner. You might also have ended up very, very rich. With AI, there has been no trade-off. The doomers who were early on accumulating GPUs, or better yet, shares in GPU companies, have done extremely well. The put option on humanity has become the call option on the technology. You could bet on the left-tail outcome, not sacrifice anything financially. We've maybe had some shades of this before. For a while, Bitcoin was, or had characterized as an end-of-the-world hedge, a form of money for when the nation-state collapsed, and people made a lot of money on that bet. But a trader wasn't driven by the fear that Bitcoin was a world destroyer. Or take gold, betting on gold at certain times protected you against hyperinflation. But the fear that came with the bet did not entail gold being the catalyst for doom. Something else would be the catalyst, but you would be protected. AI is different. The long-feared catalyst for doom itself has presented an unbelievably profitable trade, at least so far. What's interesting about this is that he's sort of— he's saying like the doomers got rich. There are plenty of AI doomers who did not join AI Labs, did not participate financially, worked in nonprofits, just wrote books, worked on research, did not ever join anything that had economic upside. But I take his point that in general, if you were early to this, oh, well, like things are going to get weird, you probably were earlier on the train. And a lot of those people that were talking about AI and doom scenarios did wind up working at labs that did very well. So it's a good point. But there are some people that don't fall in this category. Long before we had the transformer architecture or GPT, people have been theorizing about the risks of AI doom. One popular thought experiment is Rocco's Basilisk, a 2010 thought experiment. Wow, when was the last time we got a new thought experiment? I didn't know they were dropping new ones in 2010. That seems kind of recent. What's the most recent thought experiment that's really taken hold and been thought-provoking?

Golden Retriever mode.

That's not a thought experiment, that's an axiom. That's a, that's a lifestyle. A thought experiment is almost like a paradox, something you can wrestle with.

We got to come up with some new I've been wrestling with golden retriever mode.

You think it's a thought experiment?

I think you can make it one.

I don't think it falls in the bucket of thought experiment. What do you think? No. What are you guys laughing at? Are you doing more pontoon boats?

Crazy text. I'm trying to find the most recently created.

Okay, Jackson, you are correct. Yes, that is the most recent thought experiment. Good stuff. Anyway, last story that I want to talk about. Jordi, if you got anything anything, let me know. But, uh, the founder of ByteDance apparently just banned his employees from distilling US Frontier models, which begs the question, what were they doing before? But, uh, according to this post, fears distilling US models would trigger Washington, that there might be pushback in Washington, DC over this. Obviously a lot of labs, they don't like getting distilled, We've seen all the complaints and it might put TikTok at risk again. Anthropic has accused Deepseek, Moonshot, Minimax, ZAI, Alibaba, all of distilling Claude, but they haven't.

I think, I think he's doing a little bit of comms and marketing.

He said we, we should be willing to sacrifice some short-term gains for longer-term goals. No distillation allowed even if we lose. Interesting story. Yeah, we'll see. We'll see. Maybe the distillation attacks will stop. Maybe they will continue unabated. Maybe superintelligence can weed it out if you're trying to distill.

Elon on the SpaceX call, it's probably also worth mentioning that our internal projections for reaching $1 trillion in revenue have moved up from 2031 to 2030. Mm-hmm. And Ryan Peterson says, eff it, we're making our 2031 goals our 2030 goals now too.

I love it. Well, there's another model that launched, Muse Code from Meta Superintelligence. Mark Zuckerberg posted again on X, the everything app. Wait, I got to follow him. Why am I not following him? Because he's just a new poster now. Releasing Muse Code in beta today. It's a terminal coding agent that takes on complete software engineering tasks powered by MuseSpark 1.2, a coding agent, a coding-focused model update. And so I'm sure people will be benchmarking this, seeing where it's stacks up, what it, how it, how it prices out, where it's good, where it's bad, all those things. Over in Microsoft land, they're going all in on GPT-5.6 SOL. A Microsoft exec just sent an internal memo that's designed to end token maxing and makes OpenAI's GPT-5.6 SOL the default model for internal use. So that was line one. On the Ed Zitron to-do list for Sam Altman. Make sure Microsoft likes the model. And seems like mission accomplished.

We'll see you tomorrow, folks.

Have a good day. Have a good Thursday. Is it Wednesday? Have a good Wednesday. We'll see you tomorrow.

We love you.

And we love you. Goodbye. Peace.