Apple 的 AI 危机与 ChatGPT 的全球统治：Dave Blundin 与 Salim Ismail | EP #162

Apple's top executive calls the AI crisis ugly and embarrassing. I've tweeted this a dozen times. Siri sucks so badly.

Salim, stand up for Apple, because I'm not going to.

Dave Blundin, you run an extraordinary incubator and see hundreds of AI entrepreneurs. How many companies do you back per year?

Normally, it would be 10. Now it's about 30.

OpenAI is raising money at this insanely high valuation.

The AI revolution totally justifies the price.

Technology delivers any domain from scarcity to abundance.

There are 2 ways to make a fortune as a foundation-model company, but one of them, which I think Sam and Elon both realize, is displacing Google and being a direct-to-consumer brand.

Now, that's a moonshot, ladies and gentlemen. Everybody, welcome to Moonshots, and a special episode of WTF Just Happened in Tech? I'm here, of course, with my dear friend, Salim Ismail, the head of Exponential Organizations and an extraordinary friend on the forefront of all things exponential. I'm also here with a new friend—actually, an old friend, new to this podcast—and that's Dave Blundin. Dave and I go back to our college days at MIT. We were roommates at Theta Delta Chi. It was a heck of a set of rooms there. It was myself, Dave, and Michael Saylor.

While I went off to medical school, Dave graduates and starts a company called DataSage, exits it for a billion dollars, and goes on to build 20-plus companies. In particular, he built something called Link XPV, a venture fund that's been investing in AI companies coming out of MIT and Harvard with an incredible success story. I don't know, Dave, how many unicorns you've had, but I think I've lost count. Anyway, good to have you on the pod, Dave.

Oh, awesome to be here, Peter. Thanks for having me.

Salim, you just landed. Where did you come from?

I was 3 days in Helsinki, 3 days in London, and managed to get back without too much damage.

Do you know where you are right now, by the way?

I'm back in New York City, so I'm safe at home.

You started to say 3 days in Helsinki. I was going to say 3 days in “Hell-sinky,” right? Got it. Well, it sounds like your travel back was pretty insane. Anyway, we've got New York, Boston, and LA, and there's a lot that's been going on in the AI world. I wanted to really dive into it because I feel like the pace of insanity is accelerating, and it's kind of hard to keep track, but our job is to make this digestible and fun.

1. OpenAI Valuation Gamble

I'm going to start with something that continues to blow me away, and that is the fact that OpenAI is raising money at this insanely high valuation, even when there's this debate about whether it's going to remain a nonprofit or a for-profit, and when there are lawsuits flying left and right. But, man, oh, man, I love their product, and so do my kids and my wife. Dave, what do you make of this?

Well, I don't know, Peter. Are you in or out at $300 billion? You tell me. I think the challenge on this one is that the AI revolution totally justifies the price, but nobody knows if foundation models are going to be winners. DeepSeek disrupted the market. You saw what happened to Nvidia's stock that day. It's a free-for-all right now.

It's very much like the internet was back in the early days, where things haven't settled down yet. It's not obvious whether you can bet on any given part of the AI economy, but you know that the whole thing is going to have $20 trillion of impact in just a couple of years.

Hmm.

So this is either really cheap or really expensive, depending on whether it's a winner or not.

That's what makes you a successful venture capitalist: figuring that part out. Salim, same question to you. Are you in or out at $300 billion?

Personally, I'd be out, but there are 2 things here. One is that there's massive competition, and there seems to be a talent drain at OpenAI, et cetera, et cetera. On the other side, maybe Masa and SoftBank are seeing things that we don't see, and it's risky to bet against Masa historically. He makes counterintuitive bets, as he has done for a very long time, and is more right than wrong.

Yeah.

What about WeWork?

So you could take it both ways. Personally, I'm out at $300 billion.

Okay, yeah.

Interesting.

What about WeWork? Exactly. You have to imagine that Microsoft has inside information, but Microsoft's the biggest beneficiary of OpenAI because they get access to everything across the board. I'm curious about—you know, it's a— I think the estimate we had at the Abundance Summit, Dave, was $1 billion a day being invested in AI at this moment in time.

Mm-hmm. Yeah. It's amazing how everybody's leapfrogging each other, too. DeepSeek disrupted everything, but now Gemini is on top of the benchmarks all of a sudden.

Yeah.

It just keeps leapfrogging back and forth; it depends on who released most recently. But the underlying trend there is that the rate of improvement is beyond anything anyone expected, so the total size of the pie is bigger and faster than anyone would have estimated a year ago.

The pessimist scenario here is that Ilya Sutskever isn't at OpenAI, and Mira Murati is clearly not at OpenAI. If you've got all of your former co-founders starting new foundation-model companies, doesn't that scare you a little bit?

Yeah.

Then there's a counterargument in the counterargument. If you have $40 billion, of which you're earmarking half for compute, there is a capital-wins-the-race kind of argument you could make here, too. If you're ahead in the capital raising, you're just ahead.

Yeah.

So there's your argument.

You know what's interesting? Mira, I think her founding valuation was like $3 billion or $4 billion, and Ilya's funding valuation was like $9 billion. That's pretty damn good as an entrepreneur. It's like, “Wow.”

Ilya's actually at $30 billion.

Oh, $30 billion, great.

Which is just—

Fantastic.

Just a shocking, shockingly big number. But investors are risk-adjusted-return machines, and when the upside, even if it's a medium- to low-probability chance of winning the race, is in the trillions, then it's still an intelligent bet. So there is that factor, too.

Everybody, I hope you're enjoying this episode. You know, earlier this year, I was joined on stage at the 2025 Abundance Summit by a rockstar group of entrepreneurs, CEOs, investors focused on the vision and future for AGI, humanoid robotics, longevity, blockchain, basically the next trillion-dollar opportunities. If you weren't at the Abundance Summit, it's not too late. You can watch the entire Abundance Summit online by going to exponentialmastery.com. That's exponentialmastery.com. I'm curious, Salim, can OpenAI go public with all of the overhang it has with the nonprofit debate and with the lawsuits that are in place?

If you can raise this much money in the private markets, why would you even think about going public? You could stay private for a long time and wait for that. Typically, people go public because they want to raise money, et cetera. If you have no problem raising money, then why go public?

You see that across the board. There's almost no public AI companies with over 10 million of revenue, so you either have the really, really big ones or nothing.

Well, Peter, you usually have access to everything. Did you see the pitch deck on this one? Did it talk about community pathways?

I didn't.

Okay.

Nor would I have had the capacity to get into it. What we haven't talked about, which just happened, was the merger of X and xAI, which I did see. I'm an investor in xAI, and that was a pretty spectacular move.

Mm-hmm.

Right? Any thoughts on that one?

I mean, a really easy way to block off his downside and just merge it in, right?

Yeah.

The combination of AI plus the massive data set is incredible.

Yeah. I have a very firm opinion on this one. There are 2 ways to make a fortune as a foundation-model company, but one of them, which I think Sam and Elon both realize, is displacing Google and being a direct-to-consumer brand.

Mm.

In that model, Google's worth a couple of trillion dollars. It's $300 billion a year of almost pure cash flow. You don't have to be the best AI model forever if all of the consumers go to you as the first place they go on the internet, and clearly people are moving at an incredible rate toward ChatGPT or Grok as the first place they go on the internet.

So I think Anthropic isn't playing that game, and Perplexity is. I didn't see the deck on this one either, but one of the first things that probably was in the deck is: “Look, here's Google, here's where people are going, and here are the horses in the race.”

Now, Elon’s got all the Twitter data. He’s made that clear over and over again. Up-to-date, real-time information is a big part of what people search and ask about: sports scores and all the stuff that’s in the news today. I think they’re uniquely positioned to win that race regardless of the underpinnings of foundation models. So that would make those very good bets regardless of core AI.

If you’re—

You have—

…an xAI shareholder, you should be very happy.

Yeah. No, I am. But you have to believe that OpenAI is going to become a multi-trillion-dollar company to invest $40 billion at that valuation. I mean, you don’t invest unless you’ve got a 10X upside for a company like this. You know, Dave—

Yeah, well, I’ll give you one—

Yeah.

Go ahead.

Go ahead.

Well, one counterexample. If SoftBank makes many, many, many, many ecosystem investments in addition to its core platform investments, the other way you could make money—even if this is illiquid for many years to come but is the centerpiece of an entire ecosystem—is that you can basically land any venture deal you want.

Because to me, what’s really obvious right now, just as an investor, is that the use case—the vertical applications—is the best investment theme I’ve ever seen by such a wide margin.

For sure.

And every one of those boats is rising with the tide. So it is possible that the rationale here is not about trying to get a 10X on this deal. Maybe just breaking even or 2X on this deal is a great outcome. But you get 20, 30, 40 other investments that are in an ecosystem, and you can easily make up for it. Well, you’re not losing money on this either.

Yeah, I would guess—

But that’s an argument—

…SoftBank and Microsoft want to see all the application layers sitting on top of OpenAI. Yeah.

Yep. Yep.

2. AI Startups Scale Faster

You know, Dave, you’re seeing—I mean, listen, for those who don’t know, you run an extraordinary incubator basically on the campus of MIT, like a block from MIT, and you see hundreds of AI entrepreneurs throughout the year. How many companies do you back per year at Link XPV?

Well, normally it’d be 10. Now it’s about 30. Everything’s 3 times faster and bigger than I’ve seen before.

Crazy. So how are those entrepreneurs thinking about OpenAI versus Anthropic versus Grok? What’s the feeling? Do they have a favorite?

No, they don’t. They’re all multi-API. They all use all of them in parallel, especially the code-generation companies. But all of them use everything.

The debate they have is: Do we want an open-source Llama 3 or DeepSeek, and do we want to run it on our own hardware, or do we want to be API-dependent? The problem with being API-dependent is if it’s slow or—OpenAI, you saw what happened with usage this past week—it can get overloaded.

Yeah.

And that can be bad for your user experience, so they like to have an open-source model running in their own universe, just as a control point. But every one of them is using everything, because you see how they leapfrog each other. You can’t afford to be—

Hmm.

…tied to one and then find out, for one use case—

This—

…one of them leapfrogged the others.

This is the part I struggle with in this particular investment: the foundational-model space has already been commoditized. There are several out there that are nearly as good as each other, and therefore it’s really hard to make a single bet in favor of one. An ecosystem approach or access to some weird applications would be the only way of rationalizing this.

You just got back from Europe, Salim, so what’s the buzz there? How does—what was it like in Helsinki and London and wherever the hell you were? I mean, is it—do they view the U.S. as dominant? Are they—is Mistral their darling? How do you think about it from a European standpoint?

I think what we’re seeing, what I’m hearing there—and I was in a whole bunch of conversations over there—are two things. One is how incredible what’s coming out of the U.S. is, both technologically and politically, which we don’t need to get into.

Yeah.

The second part was that there are so many amazing cracks at the piñata coming from China and India—open-source-type DeepSeek equivalents—that this whole thing is moving faster than anybody possibly could have believed.

India? What’s coming from India?

Well, what they’re doing in India is they’re actually applying it in hacky ways. We did a talk a couple weeks ago, Peter, and I said that I saw a startup that had a full AI doctor coming out toward the end of this year, and I got 3 pings, 2 from India, saying, “We’ve already got a fully fledged AI doctor live and running. Here’s the link.”

Yeah.

I was like, “Whoa.” So I think on the application side, they’re actually using some of these models very aggressively, probably with local use cases, but it’s starting to get—why? Because the democratization and the cost of access are so low, right? You can try it for lots of things.

India desperately needs it, right? With 1.41 billion people.

Mm-hmm.

The only way they—

Yeah.

…survive, and they know this, is by democratizing education and healthcare using AI.

Let’s go to our next story, and this paints the picture of the universe that you’re living in, Dave. Full disclosure: I’m a partner with Dave in Exponential, in Link XPV. But here’s the headline, if you’re listening to this versus watching it on YouTube: Mercor, an AI recruiting startup founded by a 21-year-old, raises $100 million.

So, Dave, let’s tell the story of Mercor because it’s crazy. This is—and by the way, this is one of two $2 billion success stories in the portfolio, right?

Yeah, and I’ve researched that. That has happened, according to ChatGPT—maybe not totally reliable—but it’s happened about 5 or 6 times in the history of the world, and now you’ve got 2 that I know of in our portfolio—

Well, define—define what happened 5 or 6 times. Say it again.

Reaching a multi-billion-dollar valuation in under 2 years from founding day.

Nice.

Which, when you think about that, that’s crazy, right? People work a lifetime to get nowhere near that target. So it makes sense that it’s extremely rare. But I suspect in 2 years we’ll have multiples of that in terms of case studies to point to.

So this is one of the first. It’s a real bellwether. But every team I meet now is like, “Yeah, I know Mercor. I know it really, really well. It’s an inspiration. I love it.” So the story has definitely gotten around the startup community.

But the thing that makes this story exceptional relative to the other mega-successes is that Brendan Foody, the CEO, just turned 21 recently. And that’s a sign of the times. When you look at startups over history, the gating factor to getting to a huge valuation quickly usually means you need to learn something, you need to do something that takes time. Usually, that’s managing hundreds of people. You can’t just be 20 or 21 years old, wake up, and be a great manager of a team of 1,000.

But now, with AI as your workforce or your workforce multiplier, you can get there with 15 or 20 people, which is well within the comfort zone of a great 21-year-old leader.

Yeah.

So that takes a lot of the time to market out of the equation. People have been predicting this for a while. They just didn’t really believe their own predictions until now that they see it. But there’s going to be a company with 2 or 3 people that reaches a multi-billion-dollar valuation sometime very soon.

Yeah, we’ve been saying that for a while, Salim, right? The one- or two-person—

Yep.

The one- or two-person unicorn.

Since last year.

Yeah.

Yeah. And it’s happening, and it’s going to happen at an increasing pace.

Let me share 2 data points that I find fascinating that come from you, Dave, so you can expand on them. One is that I think it’s 70%, 75% of incoming MIT freshmen want to start a company before they graduate, which is pretty extraordinary.

And the second is that, if you look at a histogram of the age at which a founder started a VC-backed, billion-dollar unicorn, it used to be the early-to-mid-30s when they started. Now there’s a huge peak in the 20- to 23-year-old category. So, Dave—what’s happening there?

That’s right. That’s right. Well, I think the reality is that 20- to 24-year-olds have always had that in them, but they were hobbled. Everything is about unhobbling these days. So, what was holding them back? They’re obviously at their peak IQ.

What holds you back? Credibility has always held you back.

Incubators and accelerators now account for the vast majority of successful startups. When we started doing this, maybe 7% of companies went through any kind of accelerator program. Now it's the vast majority, and they way outperform the ones that are not in an incubator or an accelerator. Why is that? The incubators and accelerators are all about providing everything that a 20-, 22-, or 24-year-old founder hasn't learned to do yet.

For our companies here in this office, we do the accounting, we do the big data, we do the networking, and we do the legal and incorporation. Two weeks ago, I bought an apartment building across the street from MIT that can hold 6 teams—4 bedrooms, or 6 bedrooms. 60 units, 4 bedrooms each. That was actually based on this Mercor story. I was telling our team here, “Look, these guys went up in value from the day we invested. They went up $20 million a week in valuation.” If I could go to Brendan and say, “Hey, Brendan, I can save you a week looking for a place to live for your team,” that pays for the building by itself.

Wow.

So I just wrote the $5.5 million check and said, “Get the building. Let's start populating it.” It's just all about those un-hobblings. Now you have capital, you have support, and with your AI workforce, you have the workforce that you need to get to these targets.

I was teaching at MIT this morning, actually, and I was telling the students—it's Foundations of AI Ventures, so it's all about creating these companies—“Look, guys, right now you just gotta sprint. Don't overthink things. Look at Mercor. Look at what they built. It's really, really clear that there are many, many other Mercors to be built in parallel markets. Don't overthink it. Just run like hell.”

By the way, you should take a second and tell people: What does Mercor do?

Oh, that's a good point.

Yeah.

I didn't even mention that. Mercor discovered that you can interview people all over the world with an AI face and voice far more effectively than an HR department can interview them, especially for technical jobs, but it's growing rapidly from there. You can imagine that the biggest customers are Meta, and then the other AI-forward companies, like a Tesla-type company, are right behind that.

You can imagine if you went to a bank, like Bank of America, and said, “Hey, I want to use AI to interview all your future hires,” they're gonna vomit, right? Their HR department's gonna go, “What? No, I don't know why we would do that.”

That's why they're gonna go out of business.

“We're doing this. We're scared.” Exactly. But if you're Meta, you just embrace it and run with it, and they love it.

Yeah, Salim?

I think there's one more thing that's going on with these young founders. When I was building computing systems originally, the internet wave came along, and you had to be kind of young enough to be native in Internet speak, HTML, and all the protocols to be able to get your head around it. Whereas people like me had to unlearn a whole bunch of stuff and relearn new ideas.

We saw the same thing with blockchain, where there's some bylaw that you have to be under 25 to program a blockchain, right? It's literally— you skew younger because just getting your head around it requires a huge amount of effort if you've been around for a while. Whereas if you're young enough, you go, “Oh, there's this thing. Oh, this is what it's capable of. I just start using it.”

I think these AI systems are very much the same. You have the benefit of youth. You don't have to unlearn 10 years of a previous paradigm. You can lock onto this new thing and move quickly with it, so I think there's something about that in there also.

Yeah, it's being a digital native.

We try and get everyone to read The Future Is Faster Than You Think, because it really makes that point. Also, Malcolm Gladwell's Outliers makes the point that Bill Gates and Steve Jobs were born the exact same year. That can't be a coincidence, right? And it's not, because they were both 21 when the PC started to come out, and at age 21 you're old enough to start a company but you're young enough that you're not trapped by all the baggage—all the “I'm halfway done with my law degree—”

Kids.

—or whatever.

Yeah, right.

Kids.

Amazing.

Yeah, whatever it is.

Here's our next story from Fortune. It says, “1 million users sign up for ChatGPT in 60 minutes after a viral image-generation feature is unveiled.” For about 2 days, every single person was pumping these out on X. I mean, don't these people have a life? Is this what's gonna get them to sign up?

This is their life, Peter, so you gotta get to that, Peter.

I guess this is me.

This is why you have to be digitally native today.

This is me and Palmer Luckey in the image here. I can tell Palmer because he's wearing a Hawaiian shirt and he's got a mullet, but I don't know that that's me. At least it's not my self-image of me, but what the heck?

Sure.

Yeah.

We promised everybody that something would blow your mind in AI every single month from here forward until we hit the singularity, and then, of course, thereafter every single minute. This is a good example of it.

One of the first companies we incubated here at CourseAdvisor, I sold it to the Washington Post, to Don Graham. As part of that deal, I ended up consulting to him right as he was trying to save the Washington Post from the Internet. It was a really tragic but fun year, because there's no way you were gonna save the Washington Post from the Internet. It was a relentless onslaught, and mainstream media was pretty much doomed because the Internet was taking the audience.

We're kind of living that all over again. Every time something like this happens, the eyeballs came from something else. But people love to interact and engage and create a lot more than they like to be passive and watch a movie or watch a TV show, and that's where it's coming from. It's coming out of there and it's going here.

This particular thing will come and go, but something else will override it that's even more interesting. Those are gonna keep coming out at light speed now, and it's gonna take the audience away. It's a massive industry that's just gonna move to interactive, AI-driven things. The number-one use case for AI is companionship—

Yeah, for sure.

—and that's only going up from here. We can talk about that.

I saw a note on X from Elon saying he expects short video movies are gonna become dominant very shortly. And then Emad posted, “Full motion-picture length next year.” I think it's gonna be interesting, right? When you go to some new version of Netflix where it's got this massive amount of films that are being crowdsourced and crowd-voted up, and the level of creativity— we've seen the decimation of Hollywood a few times. I think we're gonna see yet another wave of disruption coming.

Did you play with Sesame at all—the super-super-empathetic AI voices—

I played—

They've just been out a few weeks.

I played with them on ChatGPT. I played with the unhinged version of Grok, which is kind of insane. If you've not played with it, I recommend not doing it around your kids. So, no, I haven't, but I think empathy is solved, Salim—

Yeah.

—we've had that conversation, right? These AIs are far more empathic than any human I know.

You could mimic empathy and get 99.9% of the way there, and we're easily there.

Yeah.

When you listen to the super-super-engaging voices, they're not married to the best LLM content yet, and that'll come out within the next—I'm gonna guess within the next 1–2 months. It's a little bit technically challenging to put the 2 things together, because to get super sympathetic, you have to run very, very quickly. To get the best possible answer, you have to run a 1-trillion- to 2-trillion-parameter model. That'll all come together. It's just inference-time compute-bound right now, but that'll come together very, very soon.

But when you hear those voices, you're like, “Oh my God—”

Yeah.

This is just so captivating. I can't believe it. The Sesame website got so overloaded with people playing with it that it was up to millions of concurrent users, and then you couldn't get in. I think it's settled down now. You can play with it if you wanna. Check it out at sesame.com.

Here's the next article from TechCrunch. It said, “Earth AI's algorithms found critical minerals in places everyone else ignored.” This is the story of abundance. This is the key point I was trying to make in the book—God, 13 years ago—that everything's there; we just haven't found it or gotten it in a usable form yet. AI discovers copper, gold, and silver in regions overlooked in Australia, combining historical geological data and custom drilling. We've talked about this, Salim, a number of times, haven't we?

We have. The data point I used to use before was that, in the last 20 years, about 40% of all oil and gas deposits have been discovered using digital searching and new sensors, right? Therefore, that's just going to go to other areas. This was a matter of time. It's fantastic to see it. I think it's such a great example of democratization and demonetization.

Yeah.

It hits right to the core of your abundance thesis: technology delivers a domain—any domain—from scarcity to abundance.

Yeah.

Do you know anything about the sensors they used to discover this? Because this story was new to me.

I don't, but I'll tell you, there's a great story I wrote about—it was probably in Bold.

This is—

Goldcorp—

McEwen?

Yeah, yeah. No, it's Rob McEwen.

Rob McEwen.

He owned a large gold mine, and he gathered all of his engineers and geologists together one day and said, “Show me where to find the next 6 million ounces of gold.” They basically said, “Huh? I have no idea.” And he goes, “You're fired,” right?

Then what he did was something very unusual. He took all of his geological data, which, for a mining company, is usually kept in a safe, and he—he didn't... The web wasn't up then. He put it on DVDs and announced a Goldcorp mining competition. I forget how much it was. I think it was $500,000.

It was $500,000—

Yeah.

—for whoever could find the next 6 million.

And so anybody who wanted to could write to him, and he sent them a CD-ROM of all their mining data—all the secret stuff, right? What was amazing was that he had hundreds of teams enter, and the winning team—there were 2 teams that won. One was in, I think, Australia, and I think the other one was in Russia, and they never visited the—

Yeah.

They never visited the mine, and they found the next 6 million ounces of gold by looking at the data. The answer is in the data.

His market cap went from $50 million to $3 billion—

Yeah.

—at a cost of $500,000, just by breaking convention. It's such an incredible story. And the way they did it was they saw patterns in the data that said, “Oh, we mine nickel, and when we see that pattern, it typically indicates a deposit over here.”

Yeah.

And so they used the kind of pattern recognition you might have for nickel or some of these other metals, and they applied it to gold. The gold people always had their own way and never looked at the nickel data, right? And so, just by cross-mapping, they thought, “This is where we think he'll find a bunch,” and they found it.

Yeah, it's great.

I remember the original question was, “How big should I have the gold processing facility?” His people were saying, “We don't know how much gold there is in there.” And he's like, “What am I paying you for?”

Yeah, Rob is now a—

It really bugs me.

Rob's a trustee or important member of the XPRIZE, one of our benefactors, so it's been fun to get to know him and his wife, Cheryl. Yeah, Dave, you were saying?

That reminds me a lot of the original thesis of the Avatar XPRIZE, where the whole goal was, “Hey, you don't have to go places to do things.” Why would you go places? You've got high-definition cameras anywhere. In this particular case, the data is not going to be visual anyway. It's going to be some kind of IoT-based sensor.

Yeah.

So what's the advantage of going to Australia to look? I mean, it makes no sense. Just stay where you are and get the data to come to you. It was a good early use case that proves the point.

Mm. I should also note for our beloved listeners that both Dave and Salim are members of my board of directors at XPRIZE, which is fun. I get to see you once a quarter at those meetings.

It is fun.

All right, next on our news—

Well, getting to—

Okay.

Hang on.

Yeah.

Getting to inspire teams to work on the hardest problems in the world is about as fun a job as you could ever get. So this is an awesome thing.

It is. It really is.

Yeah.

3. Apple Confronts Its AI Crisis

All right, so you've got to love this next title: “Apple's top executive calls AI crisis ugly and embarrassing.” Oh my God, it's about time, guys. I mean, listen, how—I've tweeted this a dozen times: Siri sucks so bad. It really pisses me off.

Oh, God.

Honestly, it's like—

Salim, stand up for Apple, because I'm not going to.

No, no. It's the standard big-company innovator's dilemma. If you're a big company, it just takes forever to roll out features, right? You have to go through control frameworks and systems testing, integration testing, and privacy testing.

Yeah.

My favorite story around this is at Yahoo, when I was running their incubator. It typically took Yahoo about 8 months to roll out a new feature because you'd code the feature, and then you had all this testing, brand testing, and all this other stuff. A hundred teams would look at it and send it back to you.

Then Zuckerberg comes along and says to his developers, “If you think your feature's ready, go take it live on the site. Here you go. And you better be right, because it'll cost you your job if you take down the site.” The developers were so excited, they were rolling out a feature a week and totally blew past Yahoo, MySpace, and everybody else.

This is, I think, the difficulty at Apple: the control framework of keeping careful and cautious in a big company totally takes out the ability. If you're on the consumer internet, the 2 characteristics you better have are speed and risk.

Yeah.

Well, I feel like I'm living a repeat movie here. I hate to say it, but I loved my original Mac. I'm sure you did too, Peter. After Steve Jobs got booted, it got so bad that you literally had to move over to a PC—it was torture—or a Sun at the time. You hated every minute of it, and you just couldn't live on the Mac anymore. It was such a piece of crap post-Steve Jobs.

Then the Newton came out, and the whole company was basically bankrupt. Steve Jobs comes back, and the next wave of Macs, which were Linux based, were just unbelievable, like out of nowhere. It's the cult of personality. This is the thing that Elon Musk has mastered now. You just have to be a cool leader online that people want to work for to get the critical talent you need to stay on the cutting edge.

And you need a through line—

You know—

You need a through line of a vision. One of the things I learned years ago is that the best movies in the world have a single director and a single writer, and they have a clear vision and execute on it. The movies that are crap are the ones that have gone through 15 revisions, and there's a third director—

Committees.

—and it's a mess, right?

Yeah.

If anything, Jobs was the orchestra conductor who had a very clear vision on this. The problem is he built this—

In Apple's defense—

Mm.

This is about as hard a thing as a big company could ever face: when do you integrate some cutting-edge new thing when you don't know how long it's going to be around, even if it might only be a week before it's out of date by the time you've implemented it? The speed of the development of AI, as Dave referenced earlier, makes it unbelievably hard to know when to place your chips.

Yeah.

Well, I still think it's pathetic, but putting that aside, this is what makes America work and what makes it great: the constant turnover of talent. At the end of the day, startups have no chance whatsoever against huge companies like Apple, Google, Facebook, and Microsoft, except for the fact that this happens. That engine of turnover is actually very, very healthy.

I hate to see the dysfunction in my devices, but it does keep the startups moving, and it's just the nature of talent.

Can we talk about your worst iPhone Siri moments? Mine that really pisses me off is, I'll text Kristen, and her name is right there. That's who I'm texting. You can spell the name. It's K-R-I-S-T-E-N. And it will spell her name three different ways in the course of the text. It's like, no, no, it's right there. And it spells my name P-E-T-E-R. It's like, honestly. I mean, consistently spells it with an I in there. It's like, stop it. I don't know.

Yeah.

Try it on yours.

Can't comment, as I use an Android.

Ah, okay.

Yeah.

Yeah. Well, you ask my wife, I'm saying “effing Apple” like 10 times a day. She hears it. It's just a litany of things. But the thing is, I'm not price sensitive. I'm ultra-time-sensitive.

Yes.

I need things to work.

Oh.

And that used to be Apple's brand, right? You know? And now it's, “We make money on every song, we make money on every movie you download.” Like, okay, but what's the core mission, guys? Just make the out-of-the-box experience perfect.

Make it beautiful, and it works.

You have the money to do it, you know?

Yeah.

Yeah.

4. AI Rebuilds Education

Yeah, for sure. All right, here's another fun story from Fox News: Texas private schools use a new AI tutor that rockets students to the top 2% in the country.

I actually know this story. Mackenzie Price, who runs this school, is a friend. She's been a member of my Abundance360 membership, and she's built this school in the Texas ecosystem, serving a lot of the families out of Tesla and SpaceX.

The adjoining story here, which I think is fascinating, is we just saw that Beijing made a proclamation that, in September of 2025—this coming fall—primary and secondary schools will baseline AI as part of their educational program, and then Estonia did the same. I'm pissed that we still think of AI as shackles for our kids or cheating for our kids, versus a jetpack. Thoughts here? Dave, you have 3 amazing kids. You have been mentoring them in AI. Talk about this, would you?

Four.

Four amazing kids, yeah. Mentoring them in AI. Talk about this, would you?

Yep. Yeah, I don't know if you remember, but Mira Wilczek, one of our partners here, wrote an entire book with GPT-2, which is incredible when you think about it in hindsight. She took it to Davos in 2019 and showed all the world leaders onstage, “I wrote an entire book using AI. Get ready for what's coming.”

She might as well have finished the presentation by saying, “Buy NVIDIA stock and try and get invested in OpenAI if you can.” All the world leaders were like, “I don't care. I don't know what you just said.” It's just like, oh my God. Talk about being asleep at the wheel. But the kids are all over it.

Yeah.

Now, because of what we do, I had all 4 of my kids go through MIT's 6.S191, Introduction to Deep Learning, which anyone can take online. So I had them all go through it early. My wife is the best wife ever, unbelievably supportive of all of this, happy to lend the kids to any of these AI missions.

But now when you talk to the kids, they're like, “Yeah, it's crystal clear that education is completely going to go to AI. What I can ask the AI today, even with today's technology, in any of my classes—anything that the teacher says is already in there—and it's much more efficient for me to talk to AI to get the same lesson.”

So they're going through the motions, trying to get through it as quickly as they can, knowing full well they're going to switch over to full-time AI-based learning from here forward.

This, plus the previous slide about Apple, et cetera, and the 21-year-old discussion, reminds me of my favorite commentary from Douglas Adams, who wrote The Hitchhiker's Guide to the Galaxy. He said, “Anything in the world when you're born, we call that normal. Anything that's invented when you're young, that's called a career. And anything invented after you're 35 years old is just bad. Just bad for the world.”

It's just a great glimpse of human nature to see stuff and just not get it unless you're below that age where it's native to you.

Yeah, I'll be really interested to see how this plays out, because the data's going to be crystal clear that this is a better way to learn, and the kids are going to be clamoring for it.

Meanwhile, you have all these teachers' unions, and I've been talking to the governor here in Massachusetts. We have a phenomenal governor, Maura Healey. She's brilliant. But it's going to be a challenge for her, though, because it's one of the biggest unions, and they're not going to want to lose their jobs.

That is the immune system.

Yeah. So, Salim, your son and my 2 sons are all 13. They're in middle school. How do you think about this?

Well, what I'm seeing is the school allows them to use ChatGPT, but theoretically not for homework. What I see him doing is looking up a math problem on AI, coming back, looking it up over here, coming back, and frankly, he watches how the AI solves it, and he learns from it, and then he does it.

How much different is that from real life? I mean, that is the way in which we learn. We watch a master at it, and we mimic it. You get the concepts into your head, and then you ground it by repetition. I think that's just a standard pattern.

It's just that, if you go back in history, learning and education was one-to-one. You had your tutor; you were an apprentice to a master. You had your guru if you were in India. And then we turned into one-to-many, where you had one teacher and many students, right?

And then with the internet, we turned it again, many-to-many, where many students can have many teachers. But now we've blown open the supply side, and the teaching can be—any one AI could teach and blow the socks off almost any teacher in the world.

That is going to change the game, and there are going to be some countries and some education systems that take advantage of that. We're going to go from push-based education, where you try to get kids into a classroom and you're trying to cram algebra into them, to pull-based learning systems, where you take on a project, you pick an MTP, and you pull down the learning you need for the task you're trying to do. I think as we make that transition, it's just really painful for the legacy systems.

Yeah, I think the teacher unions are going to be a major issue. I mean, I went into my kids' new school, because we took them out of the old school really to get them into a school that was much more progressive in learning and philosophy and, I think, leadership skills.

The realization I've had is, if you allow your kids to use AI in middle school to solve middle school problems, it's meaningless. But you could give your middle school kids college problems and have them use AI and learn to use these tools to do extraordinary things that they never thought they could do. It drives confidence, it drives up-leveling of their ambitions. And I think that's really what we need to get to. Dave, sorry, what were you going to say?

Well, I mean, there was a period of time where everyone at MIT needed to learn to use the slide rule, and it obviously became absolutely stupid at some point. We're in that moment. It's just happening way too quickly for people to adapt.

But I can't tell you how many middle school and high school students who never would have written code are now vibe coding with Lovable, Replit, or Cursor and loving it. Vibe coding is just flat-out fun, but you're creating the exact same stuff that a Python developer or a C developer would have done last year, and that is just great for the kids.

But when you look at the underlying curriculum that they have to go through in middle school and high school, that curriculum hasn't been updated since Isaac Newton. Seriously, that is the exact calculus. And the chemistry—yeah, we added a little bit of quantum to it—but it's basically the same chemistry, same biology.

Hasn't changed, yeah.

But human knowledge has grown, what, 1,000,000 times?

Yeah.

Is this the right thing to learn just because there's an SAT or an AP test for it? Is that why we're going to learn this? That's just nutty.

And you'll never use it again.

Yeah.

I just looked at a friend of mine when my cousin's sons were...

Taking his SATs, I flipped open the book to look at his ACTs and SATs, and I'm looking at the questions they're asking—

Yeah.

—and I'm going, “This is crazy.” When I knew this stuff, I would never use it again for the rest of my life.

Well, no. What you do need to know has grown so much in that same timeframe. You don't have time to cram useless stuff into your brain in today's day and age, and the education process is doing this massive disservice by occupying all the time.

Yes.

Yeah. You have that realization that you're not going to use this stuff when you leave. Unfortunately, that hit me in the middle of university because I went through this co-op program. I would be going out to work every 4 months, and I did a couple of work terms at Control Data, helping design some large-scale chip boards, et cetera.

I came back to school and was listening to a lecture where something is orbiting a star at X number of light-years' radius, and what's the orbital path? You're like, “I am never going to use this. Look, please let me get out.” I barely got my degree.

5. Vibe Coding Changes Software

Lord help us. It was about 13 years ago, I had my two kids, my two boys, and I remember at that moment in time, I made a decision to double down on my health. Without question, I wanted to see their kids, their grandkids, and really, during this extraordinary time where the space frontier and AI and crypto is all exploding, it was the most exciting time ever to be alive. I made a decision to double down on my health, and I've done that in three key areas.

The first is going every year for a Fountain upload. Fountain is one of the most advanced diagnostics and therapeutics companies. I go there, upload myself, digitize myself, about 200 gigabytes of data that the AI system is able to look at to catch disease at inception. Look for any cardiovascular, any cancer, any neurodegenerative disease, any metabolic disease. These things are all going on all the time, and you can prevent them if you can find them at inception. So super important. So Fountain is one of my keys. I make that available to the CEOs of all my companies, my family members, because health is the new wealth.

But beyond that, we are a collection of 40 trillion human cells and about another 100 trillion bacterial cells, fungi, viruses. And we don't understand how that impacts us. And so I use a company and a product called Viome. Viome has a technology called metatranscriptomics. It was actually developed in New Mexico, the same place where the nuclear bomb was developed, as a biodefense weapon, and their technology is able to help you understand what's going on in your body to understand which bacteria are producing which proteins. And as a consequence of that, what foods are your superfoods that are best for you to eat, or what foods should you avoid? What's going on in your oral microbiome? So I use their testing to understand my foods, understand my medicines, understand my supplements, and Viome really helps me understand from a biological and data standpoint what's best for me.

And then finally, feeling good, being intelligent, moving well is critical, but looking good. When you look yourself in the mirror saying, "I feel great about life," is so important, right? And so a product I use every day, twice a day is called OneSkin, developed by four incredible PhD women that found this 10-amino-acid peptide that's able to zap senile cells in your skin and really help you stay youthful in your look and appearance. So for me, these are three technologies I love and I use all the time. I'll have my team link to those in the show notes down below. Please check them out. Anyway, hope you enjoyed that. Now back to the episode. So let's turn the subject to coding. I've seen a number of people out there. Emad, on the Abundance stage, said this a number of years ago: “In 5 years, there are going to be no more coders.” And so learning to code traditionally, people are saying, is now dead. Don't teach your kids to code. Dave, what do you think about that?

Yeah. It's amazing how, to a degree, they're ahead of the curve, actually. I think this is a common thing in your network, Peter: If everyone, going back to Ray Kurzweil, said exactly what they said 5 or 10 years later, they would be Jensen Huang. But because they're such great forward thinkers, they're always ahead of the curve, and that was exactly right. It's all happening faster than anyone would have anticipated.

We're in a transition point right now where, if you want to build industrial-strength software, you still need to code, but it's going to be game over, certainly within 2025, and then everything will move to vibe coding. The good bet right now is to build whatever comes easily through vibe coding. Don't even look at the code. If it doesn't do what you want it to do, just wait.

As soon as you get in there and try to debug it, one of our companies, Blitzy, writes 3 million lines of code in a single night. So 3 million lines of code—that'd be $300 million worth of R&D that would have taken you 3 years if you had $300 million lying around, and you got it in a single night. And if you say, “Well, it doesn't do exactly what I want it to do. Let me start debugging it”—

You're dead.

How do you debug 3 million lines of code? You're dead. So what you do instead is you just keep talking to it and fix the modules you can through voice, and then just wait a couple of months, and the rest will get fixed too. And that's—

That's—

—where we are.

6. Models Race Toward AGI

That is insane. All right, our next story is dominance of Gemini. Gemini 2.5 Pro is the most powerful model yet, and it's free. So I love this.

First of all, I love Google. I've had a long love affair with Google just because I knew Larry, Sergey, and Eric early on, and they were big supporters of the XPRIZE, so I'm super biased about it in that regard. I still go back to the fact that Google really led the charge in AI to a huge degree, with their papers on attention and the early work that they did. They chose not to release their AI code early because they felt like society wasn't ready for it. But when OpenAI released ChatGPT and the world was like, “Oh, OpenAI,” it forced them to take action, and we've had that conversation before, Salim. So I like the fact that they're still up there and leading the charge.

Thoughts. Salim, how do you feel about all this?

I think this is what Dave talked about earlier in the leapfrogging. All these models are just going to step past each other in different ways. I was kind of blown away by how good 2.5 was. It was really, really impressive across all these different domains as well. So I'm really, really happy to see it.

Larry and Sergey, I understand, came back to drive the initiatives here.

Yeah, Sergey did, yes.

Yeah, it gives you the sense of why a founder-led initiative is such an important thing. You bring that energy of the original vision, the drive, and the passion, and then everybody gets motivated. I think there are a bunch of different commentaries here, but I thought the model was fantastic.

Yeah. Yeah, it's funny, I've got Gemini 2.5, Anthropic, ChatGPT, and Groq open all the time in tabs on my Chrome, and I'm just—it's interesting what I use for what. Anyway, Dave, what are you thinking here?

Well, first of all, I have to say, one of the best things that ever happened in the Nobel Prize world is that Demis Hassabis got a Nobel Prize, and then Geoffrey Hinton, who I've loved since 1986, because he invented backpropagation. But they had to really stretch the definitions of the Nobel Prizes to give 2 AI guys prizes in chemistry and physics, but it was the right 2 recipients.

Go ahead.

Huge props to the Nobel Committee for doing this. That took a big leap, and I really, really respect them for stepping out of their comfort zone for this.

Yeah. Yeah.

Well, they're going to have a real—

Yeah, I mean, think about how far—

They're going to have a challenge going forward, right? Because—

Yeah.

—almost every single major discovery is going to come on the back of AI algorithms.

Well, that's why it's so brilliant that they cracked open the door here, because they're just early in recognizing that that's the way it is, but these innovations are still huge benefits for mankind. The fact that it came from an AI person and not from somebody who toiled away in a chemistry lab for many years is totally fine. Salim, you're dead right. It was incredibly gutsy of them to give—

Mm.

out those two prizes.

I was just in Helsinki. I have a lot of fun with the Nordics because they always complain that it’s hard to compete with German engineering. And I’m like, “Okay, give me one internet innovation that has come from Germany ever,” and there’s never one.

Yeah.

And yet you look at Scandinavia and you’ve got PHP, Skype—you name it; it came from there. And one of my favorite—

Wait, you just named the 2 that I know. I don’t know any others.

There’s a bunch of them.

Oh, there are. Okay.

Spotify. The list just goes on and on. But one of my favorite—

Wow. Yeah.

—stories is that I was doing a session in Copenhagen at Niels Bohr’s house, where they did the Copenhagen—

Wow. Yeah.

interpretation. Um, and I was talking to the head of the Carlsberg Foundation. He gave me this amazing story. In the 1830s, the master brewer at Carlsberg Brewery invented the yeast that didn't get you sick. For them, up to then, when you drank beer, you got sick.

And so the lawyers got all excited, saying, "Wow, we'll patent this. We'll make a ton of money by licensing this yeast to other people." And the brewer was super upset because he was a beer fan. So he made 140 copies of this yeast and sent it out to the 140 major breweries around the world without telling the lawyers. It was the first great example of open source. He open-sourced the beer. And 10 years later, the amount of beer drunk in the world had risen 10X.

This is a great example.

Have you checked out Humanity’s Last Exam? It’s the hardest benchmark—

I was about to ask you—

—ever.

Tell us about it, because that feels—

Yeah.

—that feels like the measuring stick we need to be paying attention to.

Oh my God. Gemini’s up close to about 20% on that test, which is nutty. You ought to splice in a couple of example questions from that test in there. You could never answer even a single one of them.

I—

These are the hardest questions in every field of human endeavor, including math, physics, and everything else. But just the domain knowledge to understand what the question’s asking you is off the charts for any normal human being. And this model—

Does it—

—does close to 20%.

Does it solve the inability for men and women to understand each other’s thoughts?

Yeah. No, that’s out of range. I want to ask you what you think of Llama, though. Llama 4 will be out before too long. To me, the Gemini 2.5 release is just leapfrogging, but it’s exactly what you’d expect. Every time there’s a release from any of the big guys, it leapfrogs everyone else for a little while.

I think what’s interesting to me is that the team at Google has to be very, very cautious with safety, security, and privacy, and they have that new paper out and everything. The team at OpenAI is fundamentally Microsoft’s bet, in addition to being OpenAI, and they’ll leapfrog this, I’m sure. But Mark Zuckerberg, to me, is… I have to tell you a story, actually.

Mark Zuckerberg is the sort of founder-leader still in control, driving AI with open source. We just invested in a Harvard team yesterday and signed the deal in Mark Zuckerberg’s dorm room and—

Nice.

—got some great pictures of it. They’re trying to walk right in their footsteps, but it was super fun. But what do you guys think? Do you think Mark will then keep up with these guys or not?

I think—

Well, I think the—

He has to.

—the great aspect here is the fact that open source will… We have a section in the ExO book called “Open Beats Closed,” and we’ve seen over time, throughout history, that open models beat closed models. We’re at that point now because there are just so many more eyes and attention that can go to it and evolve it very quickly. I’m super excited about what we’re gonna see in Llama 4.

And so Elon is the other sort of founder-originator guy in the race. Do you think he’ll… If you had to pick one, everyone has their specific advantages. So you’ve got 4 choices, and maybe there are others too. Maybe DeepSeek is a 5th choice. But which one would you bet on?

I tend to bet on Grok. I just think he’s got an unlimited checkbook, and he’s got a massive amount of data coming from X, Tesla, and humanoid robots. Yeah. Salim?

I think—bet for what purpose? Do you mean betting on building the best next model or lasting over time?

Hmm. Good question. How about just the Humanity’s Last Exam benchmark—who has the highest—

Oh—

—score 2 years from today?

I think it’s gonna be an unknown entrant.

Oh, no way. Really?

Yeah, I think something’s gonna come out of the blue—

Whoa. You heard it here first, folks.

—where people are building—

Okay.

Because—

Salim’s back.

—we’ve seen that throughout history. We thought nobody was ever gonna beat Yahoo, and then Google came along and blew that wide open.

That’s right.

And then Facebook came along and blew past that. OpenAI comes along out of nowhere and blows past everybody else. So I think that’s the pattern that we’ve seen throughout history. You’d be hard-pressed to bet against the unknown entrant.

Yeah.

Well, I’ll tell you one thing.

Be unpredictable.

Elon was very early to recognize that who gets the chips is a huge factor in that question. And so he put himself in a position now where he has a lot of control over—first of all, what gets manufactured at TSMC and Intel.

Yeah.

A lot of control over that, and then who gets the supply, he has a lot of control over that. So that is a pretty important—

And who gets access to—

He’s just such a forward thinker.

—gets access to transformers and electricity and the entire supply chain for getting intelligence out the door.

You know, I miss the idea of testing a human IQ on this. We were measuring human IQ, and then we had GPT-o3 or o1 with a 120. We haven’t seen any new IQ data coming out. I’m looking forward to that. All right. Here’s another one. OpenAI and Microsoft now back Anthropic’s Model Context Protocol. Who wants to take this on? Dave, do you wanna give us a quick 101 on MCP?

MCP is basically—so, with ChatGPT, you can do almost anything. You ask it to do something on your laptop, and it says, “Well, I can’t control your laptop, but here’s some code that would do it.” And you’re like, “Well, what do I do with that?” “Open the terminal session and paste it.” You’re like, “I’m not doing that.”

So then they come out with Operator. The problem with Operator is that it can delete everything on your laptop. And so Dario Amodei is actually the god of this. He is an incredible forward thinker in terms of how to get productivity while still having safety and security.

Hmm.

Somewhere in there, he said, “Well, here: MCP. This is gonna be an open standard. Make it free to the world.” But it’s a way that the AI can access and control things like your laptop, but also information that may come from other agents, or it may just be stuff out on the internet that helps the AI serve your purpose. It’s a simple protocol that enables all this, and it’s really pretty brilliant.

I think it might be very similar to HTML. HTML was not super complicated, but it just flat-out worked, and we’re still using it today. Here we all have the internet. What a game changer. So MCP is very much like that kind of breakthrough in terms of having the AI empowered to help you without wrecking everything.

But I love the fact that both OpenAI and Microsoft have backed that. It’s part of the sort of open ethos. All right, let’s go to our next item here.

Just a quick shout-out: if you go back to that story, Peter, what I loved was that this was based on OAuth, which was a protocol for sharing identity across websites. Something that people should look into is the Internet Identity Workshop, where all the folks working on identity systems at eBay and Yahoo get together and ask, “How do we handle cross-site identity, et cetera?”

OAuth was a protocol that emerged out of that. So this is literally the community practitioners getting together to create a protocol. And what I’d love to see with this MCP is the same thing happening again: the practitioners in the space getting together to say, “How do we do this in an appropriate and responsible way?”

Hmm.

So I’m really, really happy to see that.

Beautiful.

Yeah. I’m not 100% convinced that the other guys adopting it is them being altruistic and saying—

Yeah.

“Yeah, great invention.” I think it might be more of a, “Hey, we…”

“Let’s make sure we’re close to this.”

“Okay, let’s steal it.”

“Let’s steal it.”

Yeah, exactly.

Yes.

“Great, next.”

There’s definitely that. It’s the 40th anniversary of the IIW workshop—the Internet Identity Workshop—which is an unconference that’s been going for 20 years, where all these folks get together. It’s amazing to see them trying to figure out the block-and-tackle problems of interoperability at that level.

Nice.

Just a shout-out for them.

All right. Speaking about the importance of safety, Google released a paper on AGI safety by 2030, and what could possibly go wrong? I don’t know if you had a chance to see the paper. I took a look at it, and it talks about a wide range of areas.

I was with Eric Schmidt down in Miami at the FII Miami Summit. I had a conversation that was super critical, and he said, “The biggest concern isn’t the upside, of course. It’s the downside.” The biggest concern isn’t the U.S. versus China. There are concerns there, and we can get to that if you want, but it’s everybody against the dystopian user. It’s the person who uses AI to cause havoc.

The number 1 example they use is a bioweapon that gets produced because that technology has gotten easier and easier. Eric said we need something like a Three Mile Island disaster—something that makes global news but doesn’t actually hurt anybody and scares everybody—so that we adopt reasonable guidelines.

That’s funny. I interviewed Sam Altman at MIT, at IIA, 2 years ago next week, actually, and we talked about this. He said proactively that the worst thing that could happen in AI is if we get into some kind of hypercompetitive, all-out race—maybe an arms race with a foreign nation or something like that—where nobody feels like we have the ability to slow down and think.

We have to move full speed because otherwise somebody else is going to bypass us. That would be terrible. That would be awful. But thankfully, that’s not happening right now, and we can take our time and think about it.

Sure. Sounds familiar.

I have to dig up that recording, actually, because here we are, 2 years later.

Yeah.

Yeah.

I mean, that is the case right now. Everybody is all-out. The latest estimate is $1 billion a day being spent on AI. It’s unconstrained. There’s no velocity knob. It’s all tilt to 11, with no on-off switch.

Hmm.

So—

I have a tangential question to this topic that I’d like to ask Dave.

Please.

I think the approach taken by the paper is perfectly thoughtful and so on. I struggle hugely with how you clearly define AGI. Dave, do you have, or have you seen, a decent definition of AGI?

Well, I’ve seen some really good definitions that are crystal clear, but then everyone disagrees about whether that’s AGI or not.

Well, that—

And it’s like, well, that’s just terminology. Who cares? Demis Hassabis has a really, really good one: Take all the information up until about 1910, and then try to have an artificial Einstein discover relativity given that data. If it can do that, then you know you’re over the threshold.

That’s a great benchmark. Let’s code it up. Then someone will say, “Well, that’s not AGI. It doesn’t have empathy.” Okay, that’s a perfectly legitimate complaint. But AGI was originally defined as the Turing test, and everybody agreed for 50 years, right?

Yeah.

And so—

We blew past that.

I think it’s crazy to change the definition right at the finish line.

Oh my God.

Like, as an AI—

The Turing test, as classically defined, was passed a long time ago, and no one noticed and no one cared.

Yeah, because we’re there. We’re exactly there.

Yeah.

Richard Socher has really great thinking about the different types of intelligence going forward—the 10 pathways that he’s carved out. He promised us that book he’s supposed to write. I don’t know where that is.

Hmm.

But it’s a really well-thought-out process. From here forward, it’s not about whether it’s equal to a human when it’s doing superhuman protein folding and superhuman cell simulations. Do we really want to make an AI that acts like a human anyway? Isn’t that the disaster scenario?

Can we just make it not self-aware but super-capable in all these areas that help humanity? It’s a very doable thing. Why don’t we just do that, and then everyone will be happy?

Yeah.

Wow.

Okay. I’m with you.

7. Physical AI Takes Shape

I’m going to move off the subject of AI for a moment and move to one of my favorites, which is flying cars, and share a short video. The title here, from the South China Morning Post, is “China clears first flying taxis for commercial takeoff.”

Here we go. Let’s take a look. Those of you who have been with me for a while have probably seen a variation of this vehicle. It looks like EHang from China, one of the very first to be produced.

I remember when Martine Rothblatt purchased about 100 of these vehicles for use in delivering lungs in the United States, and then she went and backed BETA in the U.S. Salim, you and I have chatted about flying cars a lot.

I mean, for fuck’s sake, why can’t we have these yesterday? It’s so much easier to fly a car than to drive a car autonomously. You’ve got so many fewer constraints.

I’ll make a controversial comment: Kobe Bryant would be alive today if we’d cleared autonomous flying cars 20 years ago or 10 years ago. The technology’s there. It’s regulatory, it’s infrastructure—you have to have dedicated gates at airports, special fueling mechanisms, licensing. It takes a long time, unfortunately, for human beings to get all that stuff into place.

This is where an authoritative government like China can get it done quickly, because they can just look through it and say, “Make it happen.”

Yeah.

I think it’s a game changer. I can’t wait for it to happen. I think we should’ve done it ages ago. Airport commutes are just a disaster.

Yep.

—

How long were you in line to get to JFK?

There was a 30-minute delay. It’s like LAX on a bad day, right? It’s like a 30-minute delay at 1:00 a.m. They were doing something, and God knows what. My car was stuck forever coming to pick me up.

It just doesn’t make any sense that we can’t break through that stupid barrier, at least for airport runs. Price is not an issue. People would pay a fortune just to be able to get there reasonably.

Well, there are a dozen well-funded eVTOLs—electric vertical takeoff and landing, right?

I know.

Joby and Archer are probably the 2 biggest ones in the United States. The plan right now, as far as I know, is to have them up and operational when the Olympics are coming in a few years here in Los Angeles.

The problem is the old adage: The FAA is not happy until you’re not happy, right?

Okay. Well said.

You know that better than anybody, Salim. We have drone delivery as well, which has finally broken through. We have Wing, which is part of Alphabet—

Yep.

—and Zipline, and they’re making millions of deliveries now, with drones flying basically autonomously.

This is so funny. I don’t know if you remember hearing the CEO of Wing talking about this: These little drones are flying, and now they’re operational in Dallas and will move to other cities. They have to get a Part 135 license.

Part 121 is United and JetBlue. It’s large airplanes carrying passengers. Part 135 is smaller aircraft carrying passengers. They have to abide by the Part 135 rules, which means your airplane has to have a seatbelt on it and needs to have all of the paperwork on it.

And, of course, you’ve got these little drones, and we need regulatory—

Well, Peter—

Yeah?

Well, it’s something we were talking about maybe 6 months ago. I don’t know if you acted on it, but island real estate is dirt cheap, either in the middle of a lake or off the coast. And that’s because it’s hard to get to.

You know, we were ahead of the curve talking about the land around nuclear power plants. And lo and behold, it’s gone way up in value. But one of the smartest things you could do right now is start a little fund to buy island real estate in anticipation of the arrival.

Yeah.

We have a fully developed business plan, developed by one of my community members, for exactly that.

Hmm.

We did a quick study. It turns out if you just circle Toronto Airport and Chicago Airport with a 200-mile radius, there are something like 10,000 islands and lakes.

Hmm. Mm-hmm, mm-hmm.

Right? It’s a ridiculous number.

There you go.

You could go around the world where there’s decent rule of law and just snap these up. There’s no way they’re going down in value. And now you get this: you just put a drone landing pad on each one and wait.

Hmm. That’s a good—

Well, I asked around the office about this topic, and everybody says, “You know what? China can win some races. That’s okay. This is one we don’t need to win. We can just import these things.” But, like you said, we need to get the regulatory thing figured out very quickly.

Yeah. Yeah.

But as long as we’re competitive in the core AI race, this is a good one to concede, because they’re so far ahead of us anyway, and just start importing them.

Yeah. I would pay the tariffs for that.

All right, so here’s another race that’s going on between the U.S. and China, and it’s the humanoid robot race. Of course, this year I had Brett Adcock, the CEO of Figure, on stage with me at the Abundance Summit.

Hmm.

Their recent release was that the Figure 02 robot can now walk like a human. Let’s take a look at this video here. This is Figure 01, which is walking like an old crippled man. And here’s Figure 02 coming along.

Yeah.

It still looks a little bit awkward to me.

Mm-hmm.

Doesn’t it? I mean, it’s getting better. But I’ll tell you what’s amazing: kudos to Brett and Figure. They have Figure 03 coming out shortly. They’ve got Figure 04. They’re iterating a new robot design every 12 to 18 months. It’s extraordinary.

I used to be a superfan of the Atlas robot back when it was being powered by hydraulics, and now it’s gone electric, and it’s amazing. So let’s take a look at the Atlas robot. This video really freaks me out. Check this out. Look at Atlas running like that, right? I mean, it’s much more human. Crawling.

Wow.

I mean, very human-like maneuvers, right? Let’s see what it does next.

That’s so cool.

Some tumbling. I mean, imagine this in the Olympics, in gymnastics.

It’s a little unfair, isn’t it?

But yeah, sure.

I mean, you know my beef with these—

Right.

Right?

Headstands. Oh, wow.

Oh, okay. Now we’re doing—now we’re gonna— That’s a different—

It’s a little—

Now we’re showing off.

—sport now, actually.

I mean, but check it out: cartwheels. I mean, honestly, it—

Wow.

—looks like... But just—look at the beginning here again. It’s running. Check this out, this running gait. I mean, amazing, isn’t it?

Yeah.

Yeah.

So—

You know my beef. Why the hell only 2 legs? You know—

Yeah, because it would be a horse.

—it could have 4. It could have 6.

Then it would be a centipede. Why not 100?

So much more flexibility.

Yeah.

I think we spend—

But—

—too much time trying to make them look human.

Okay, Salim. Salim, okay.

Yes, I’ll back off.

Okay. Forget this argument. You’ve made it in the last 3 conversations we’ve had. Putting it aside—

I know.

That was Atlas running, superhuman-like. Imagine that running after you down the street.

Yeah. Crazy.

I’ll go back to the thing that does blow my mind about this: once you teach one of these a skill, all 1 million other humanoid robots have that exact same capability.

Yeah.

And that shareability is something we’ve not achieved in human beings, and I think this is where humanoid robots have unbelievable applicability.

Well, I also think that the reason we’re doing humanoid robots is not because Elon says, “Well, look, the world is designed for human shapes,” blah, blah, blah. It’s not true.

What’s true is that this inspires a whole generation to work on robotics, and once you have the supply chain of components and also the software, you can make any form factor in no time after that. So why start with a humanoid and not with a spider or something?

Well, if you start with something creepy, then you just turn off a whole generation. You start with the video we just looked at, you’ve inspired a generation, and that’s more important at this moment. That’s what I think they recognize, and—

That’s the kind of wisdom that cuts through all of the crap we talk about sometimes, Dave. Thank you for that. That’s a great comment.

You’re not supposed to throw compliments to me.

All right, here’s—

I do appreciate it.

A couple of final stories here. Feeling the future: new wearable tech simulates realistic touch. I want to say, “New wearable tech stimulates realistic touch.” So this is—

Put a comma in there.

Fruit of the Loom is coming out with this new technology.

Okay. Well, that’ll sell. You got that right.

Yeah. This breakthrough in haptic tech stimulates pressure, twisting, and stretching under full-motion actuators. Listen, we saw this in Ready Player One. This is reality following science fiction. This is the emergence of our diving into the virtual world. Comments?

The porn industry is gonna have a field day with this one, obviously. There have historically always been breakthrough innovators in some of this.

This is where that whole example of putting away kitchen dishes and then simulating becomes really interesting for the avatar use case. If you can do haptic sensing in the robots on Mars and you’re on Earth or you’re off-planet, there are just some incredible things you could do there. Lots of applications.

Yeah, I totally agree conceptually, but when we did the Avatar XPRIZE, the latency kills you. When you start talking about something on the Moon or something on Mars, on Mars you’re talking about minutes.

You know what just happened, Dave? We talked about this at the Abundance Summit. Quantum communications have made instantaneous comms possible. I mean, that blew me away. All of a sudden, distance doesn’t matter. You could have instantaneous communication capability. So who knows?

Yeah. But if that works—and I’ve read about it—Einstein might agree, might not agree. Quantum entanglement is pretty compelling, so I guess I don’t know. I’m not that deep into physics to know if it’s real or not.

If we have—

But if it is—

If we have quantum entanglement, we’d do much more interesting things than haptic pressure testing and touch. So I think we’re gonna be way, way past all of that.

That’s a good point. But if the instantaneous communication from here to Mars can eliminate the 10-minute latency, if that actually works, then we’re either immediately gonna discover 10,000 other civilizations in the galaxy or we’re gonna know that they don’t exist. So that’s pretty interesting if that’s real.

We’ll solve the Fermi paradox.

Amazing.

Yeah.

8. Bitcoin Meets AI Markets

To wrap up with one Bitcoin story, Trump family enters Bitcoin mining with new venture, American Bitcoin. At the same time as we’re recording this, there’s been a sort of meltdown of all U.S. stocks, and Bitcoin dropped a good 10%. What happened to the idea that Bitcoin was an uncorrelated store of value?

Yeah.

It’s still not uncorrelated.

Yeah, it’s still super correlated.

Well, I mean, everything—the ETF, everyone moving to indexes, which are stupid. And I don't mean that they're literally stupid; they have no thought behind—

Gotcha, right.

—what they buy and sell.

Yeah.

Yeah, the ETF-ization of the world can't go much further without everything moving together. People are pulling money in and out of ETFs, and they're just selling indiscriminately. But you saw halfway through the day that Intel was way down, and then suddenly somebody realized, “Oh, wait, Intel's got great news today. It's immune from this,” and it was the one thing green on the ticker.

So it's funny how they created this huge arb opportunity in the middle of the day. Of course, all the quant funds are making 40% returns like clockwork. Now, how does that work? It's just picking up on that ETF inefficiency. Anyway, that's a bit of an aside, kind of geeky. I think what's going on here, though, is that data centers and chip access are an absolutely dominant part of economic success going forward.

Mm-hmm.

And what do you know? Bitcoin mining is a great way to make money on idle data center power supplies. It's hugely correlated with inference-time compute and training-compute build-outs. And so, of course, the Trumps are gonna be in the middle of those conversations and recognize the opportunity.

Nice. Agreed. All right, Dave, I added this story for you. CoreWeave shares rip nearly 42% higher, rising above the IPO price. So tell me about CoreWeave.

Okay. So CoreWeave is a heavily levered kind of bellwether of where AI IPOs are gonna go, but it's a tough case because it's got so much debt. What's amazing to me is that they wanted to price at $48. There was a little shaky demand, so they came out at $40, which is still great.

It traded down, and everyone got scared to death for 1 day, and then it skyrocketed. Now, even with today's meltdown, it's way up from its IPO price. That is, to me, very much the Yahoo! moment. Remember when the internet was starting to take off and Yahoo! went public, and that just blew open the markets? Then they stayed open for a good, solid 5 or 6 years after that.

To me, we could look back and say, “Oh, okay, this was the bellwether that said the markets are back open.” It says right there on the bullet: 2021 was the last time we had anything like this. That's a long time ago.

Yeah, it is.

I'm hoping, and I really feel like this is gonna crack open not only the IPO markets, but also the M&A markets—

Yeah.

—yeah, for all these AI companies. And that's what we desperately need. The alternative is that we lose the race in AI, but this has to happen for the machinery to keep running. It's great.

Salim, thoughts?

It's really great.

I'm super thrilled to see it also, because the markets need to open up, and I think that was one of the core promises of the Trump administration: to open up both M&A and the IPO markets. And it's great to see the sentiment behind it because, as Dave said, the next few will have as much excitement around them by definition, and it'll be kind of amazing. It might crack open the whole paradigm.

Yeah. So, Dave, listen, I loved having you on this episode.

Oh, likewise.

It's like, you know—

The most fun. I've been looking forward to it all day, but it's great.

Yeah, it's great. So everybody, listen, I know you love Salim more than me. I got it. You know, he's just better-looking, more intelligent, all of that.

Where do you get that? Does that come from the comments or something?

What? Let me know what you thought of having Dave on WTF. I love bringing the most intelligent people I know into this conversation to give an understanding of what's happened in the past week. And as the speed of this accelerates—and it's accelerating rapidly—I love having more intelligent people than myself to analyze the news, give opinions, and set me straight.

Well, also the insider's view, because, Dave, you've got so much direct knowledge of all these startups and who's doing what in the space. It's fantastic.

Yeah.

Well, I don't know if you remember, Peter, but all the conversations we've had with Michael Saylor and Sanju Bansal, and just walking around the streets of Cambridge—this is just the virtual version of it. It's been the most fun I've had in weeks, and I'm just absolutely overjoyed that you had me on, so—

I look forward to having you back.

Yeah, I'd love to.

Yeah, for sure. All right, guys. Get some sleep, Salim. I know you've been traveling the globe. And Dave, thanks for joining us, buddy. See you guys next time. If you enjoyed this episode, I'm gonna be releasing all of the talks, all the keynotes from the Abundance Summit exclusively on exponentialmastery.com. You can get on-demand access there. Go to exponentialmastery.com.