Opus 4.8 发布、Demis Hassabis 预测 AGI，以及2200亿美元的基金会 | EP #260

Welcome everybody to Moonshots. Another episode of WTF Just Happening Tech. I'm here with my extraordinary moonshot mates, Alex, our in-house polymath.

Alex, good morning to you. Where are you?

Alex

Good morning, Peter. Still in Chicago. Excited to be here. I'm heading back soon.

Dave, good morning to you.

Dave

Good morning. Good morning.

Salim Ismail, our father of exponential organizations—the man who just gave away his book and his Claude skill for free. There's a puppy out there that wants to get in on me.

Yeah, wants to be uplifted.

Interspecies communication in action. It looks like you're back home, Salim Ismail.

I am back home, resting after a crazy week. I've got another crazy week ahead of me, and then it should settle down after that.

You are a probability function on the planet.

We've got a fun episode for you today. No gloom, no doom, just the science and tech accelerating us towards the singularity. Here's a quick preview of what we're going to be covering. Anthropic just dropped Opus 4.8, reclaimed the coding crown from GPT-5.5. Demis Hassabis just tightened up his timeline for AGI, agreeing with Ray Kurzweil for 2029. Amazon just launched a new AI shopping assistant. We'll cover a biotech breakthrough out of China, a pocket-size cancer detector that can spot tumors with a single drop of blood at 95% accuracy. And then we're back in the quantum computing game. The U.S. government and IBM just dropped $2 billion to build a chip foundry. As always, we'll end with your questions. Our mission here at Moonshots is to keep you optimistic, informed, and ready for the supersonic tsunami heading our way.

Dave

I have a word definition.

What's that?

Dave

Alex threw out a bunch of words nobody understood—or at least I didn't—last time. I've got one, which is pronoia.

Paranoia is an unreasonable fear and suspicion of others. Pronoia turns out to be exactly the opposite: the belief that things are just going to work out. I think that summarizes our podcast very well, and I thought that was worth getting into.

We should change the name to Pronoia.

Alex

You can't even pronounce it, and it's horrible. It's an ugly, ugly word.

Let's make up a new neologism.

Let's open with our first story. Anthropic just dropped its new model, Opus 4.8, just 6 weeks after Opus 4.7. If you remember last episode, we talked about GPT-5.5 running away with the coding benchmark. Well, Anthropic just fired back.

Opus 4.8 now leads the Artificial Analysis Intelligence Index at 61.4, 1.2 points ahead of GPT-5.5. On SWE-bench Pro, the hard coding benchmark, it scored 69.2 compared with 58.6. All these numbers blur, but that's just up and to the right, and it's the only model to complete every case end to end on Anthropic's Super Agent benchmark. Here's the kicker: it's 4 times less likely to overlook bugs in its own code.

It feels like we're in a 2-horse race here between Anthropic and OpenAI, releasing every 4 to 6 weeks. Alex, why don't you dive in on this one, pal?

Alex

I do agree with the premise. Even though I take heat sometimes from the Grok fans, it does feel like we're in the 2-and-a-half—or duopoly—phase of this particular rat race.

The particular benchmarks, the evals that I'm paying closest attention to here, are SWE-bench Pro at 69.2%, Humanity's Last Exam with tools at 57.9%, and then, probably most interestingly at this point, GDPval at 1,890. I think we're at the saturation phase of these particular benchmarks. I think we need a new set of benchmarks, probably recognizing that the next phase of capabilities won't just be solving problems that we already know the answers to, but solving unsolved problems.

I was making that point at the Department of Energy's Genesis Mission event here at the University of Chicago. We need a new set of benchmarks that are able to capture, say, scientific and engineering open, unsolved problems as the next raft of benchmarks. I think that's where this is going.

There were some related ancillary announcements that Anthropic made as well. They announced that sometime in the next few weeks, in addition to Opus 4.8, they're planning to release new models that will rival Mythos in terms of capability. I think the heat from GPT-5.5 is catching up with them. You can tease unreleased models for only so long before your competitors—or at least your competitor, singular—catch up.

They also released some interesting scaffolding updates. They released a feature for Claude Code called dynamic workflows, enabling users to spin up hundreds of parallel subagents to tackle very large codebases.

The punchline is that I view this as an incremental, now-monthly update. We're in the monthly-update regime of the rat race, probably soon to be weekly, then daily, then hourly, when we finally reach max Q of the singularity—the singularity of the singularity, as it were. It's a very nice, solid monthly update. I've been playing with it, and it seems solid by Opus standards.

Nice, Dave.

Dave

That was a more thorough diagnosis than I expected. Alex, I thought you were going to say, "Another solid dot release. Don't freak out: another month, another dot release."

Alex

It's important to step back, though, and look at how far it's come in just 6 months. I've installed it right away. Of course, I have about 100 agents running right now, and I run them on EC2 so I can close my laptop lid and they keep grinding away.

The thing I've noticed is that trying to do many things concurrently has never worked particularly well for me because they don't integrate well. This feels like it's significantly better at managing many parallel threads, and I think that's really important. For large-scale creation of brand-new things, one of the best advantages AI has is the ability to be effectively 1 billion concurrent workers, or 1 trillion concurrent workers—unachievable by humanity just because of raw parallel scale.

Nothing previously seemed to assimilate back into a final product particularly well for me. Now it feels a lot better. I'll let you know in a couple of days if it succeeds in self-improvement and self-assimilation.

I did have some trouble with the forking. The new capability that's exciting is the ability to say, "Hey, I don't want to create a new context given everything. Create a self-fork." It resists wanting to self-fork. It says, "Do you really want to do that? That's a lot of bloat." Then I do the math, and it's not that much fork.

What does that mean? What do you mean by forking here?

Dave

Previously, if you were talking to an agent and had it all queued up, and you'd told it everything you were trying to achieve, then you said, "Now I want 100 of you to work on something," it forced you to create a new context, like a Markdown file or a description, and launch new children that knew nothing. They were absolutely bare metal, so you had to bring them up to speed somehow, which took 20 or 30 minutes to get the prompt right.

Now you can say to it, "Self-fork everything that you know—everything we've ever talked about. Exactly. Clone that and make 100 of yourself that are identical self-forks." This was going to trigger Alex to say, "Wait, those have rights. Those are 100 voters. How are we going to deal with that?"

Alex

Exactly. Look what happens when humans can fork themselves.

Dave

No, that's right out of Accelerando, actually. There's a good description of how that works.

There's an old model for humans forking themselves, which is biology, sex, and kids.

Alex

That's really not forking, though, because, to Dave's point, children are born context-free. It's a little bit more, I would say, akin to Unix processes, where a Unix child process, by default, inherits the context of its parent.

We're so geeking out here. Salim Ismail, when you hear, "Okay, we got 4.7, then 4.8, then 4.9," what do you make of that?

I think what's clear is we're going to need much more orchestration and routing of intelligence, where high-cognition tasks use the latest models, and low-cognition tasks use older models that make the tokens cheaper.

The question I had on this—and maybe, Alex, you can take a crack at this—is: Why is there so much consistency across these models? I find them remarkably close to each other, because some 2 or 3 of them take quite different approaches to this. Do you have a sense of why that could be the case?

I'll tell you why: Anthropic is holding back. [laughter] Go ahead, Alex.

Alex

Yeah, I think there are probably a few possible reasons, that being one possible reason. There’s a race dynamic here, so there’s maybe not such a strong incentive to leapfrog capabilities if leapfrogging to a dramatic extent requires an enormous amount of expense. I think that’s part of it.

Part of it is just that I think the 2.5 frontier labs that we seem to have right now are maxing out on their capabilities. If you look at their data centers and their compute capabilities, there’s not a single one at this point that has an order of magnitude more compute than any of the others. They’re all within a factor of 2 or 3 in terms of the amount of compute that they have relative to each other.

Then maybe the least obvious aspect is that all of these benchmarks are saturating. If you’re saturating, it’s really easy to be relatively close to each other. It’s only when we see radical new benchmarks that you’d expect to see more dispersion among the possible scores. We’re in an era of superintelligence, and it’s very easy with superintelligence to just saturate every obvious benchmark that you throw at it. So, of course, they’re close.

Yeah. What’s your guys’ guess on when GPT-5.6 comes out?

Alex

The next few weeks.

Yeah. I just—

Alex

Monthly. We’re in a monthly horse race now.

By the way, I now put out a Moonshot summary on Substack, which includes a link to all the stories that we cover. The weekly recap covers what I and the mates had to say, what we think is most important, and what we're most excited about. And it's free. You can subscribe at diamandis.com/metatrends.

Our next story here is Sir Demis Hassabis, CEO of Google DeepMind and Nobel laureate, just tightening his AGI timeline. He’s now fully aligned with friend of the pod Ray Kurzweil and his original projection of 2029, just 3 years from now.

What I find interesting about his comments is his frame. He said that today’s AI agents are “a practice run” and that society has, quote, “only a few years to prepare for what’s coming.” Think about that: You’ve got the head of the top AI lab telling the world, “You’ve got to take this seriously.”

He’s also proposed something we’ve talked about before on the pod called the Einstein test for AGI: Take a model trained with knowledge only through 1901 and see if it can independently derive special relativity. He’s commenting that current systems can’t do that. Remember, at Google I/O, he said, “We’re on the foothills of the singularity.” Alex, you’ve famously said a number of times that we’re at AGI already.

Alex

Yeah, and have been for 5 to 6 years. I think this is sort of—I mean, I like Demis a lot, but I think this is sort of a bizarre statement on his part: that we’re not at AGI yet, but that it could arrive by 2029.

It’s also a bizarre juxtaposition to be posing such a conservative time frame when Gemini is seemingly about to lose—unless it just leapfrogs in terms of capabilities—the horse race. Maybe that’s too strong, since the horse race, or the rat race, or the fill-in-the-blank animal, nonhuman-animal race, can probably continue ad infinitum. But right now, they’re certainly not winning. Gemini is not winning the race.

So, to frame the timelines for artificial general intelligence as 3 to 4 years from now, when your lab is right now not in the lead, it feels to me a little bit like moving the goalposts conveniently, maybe somewhat self-servingly, to buy more time for DeepMind to leapfrog—hopefully to leapfrog—to wherever it thinks it’s going. But I don’t agree with this construction that somehow we’re going to get AGI by the end of the decade. As I’ve pointed out numerous times, I think we’ve arguably had some form of artificial general intelligence since 2020.

When you say that, Alex, do you mean that we’ve had the construct that evolves into AGI, or that we actually have it? Because the conversation we’ve heard this from Sam, we’ve heard this from Dario, we’ve heard this from Demis basically saying that we haven’t seen leaps of intellectual progress, like his example of his Einstein test. We don’t have a system that can do that today. We have systems that are solving math—got that—but none that are coming up with brand-new theories of physics. Do you—

Alex

I was so unnerved by this that I posted on X sort of an argument saying everyone has their own definition of AGI at this point, myself included.

Enter rant.

Alex

No, yeah. It’s not that we don’t have any definitions. It’s that we have many definitions, and they all roughly overlap. If you really squint and zoom out, all these different AGI definitions will roughly correspond to a single 10-year period.

With the benefit of hindsight a few decades from now, I think we could just say, “What was everyone hand-wringing over?” Whether you think it happened in 2020 or 2029, it happened in a relatively abbreviated historic period.

There are the doomers, I remarked online, who say, “We’re already cooked.” There are the skeptics who say, “Can’t do my laundry yet.” There’s Demis, who’s saying it’s not AGI until it replicates special and general relativity. Then you have me. I think generality was achieved, arguably, with GPT-2 and large language models, or few-shot learners, which was, to my knowledge, the first time—at the very latest—that we learned you could achieve generality through a combination of prompt engineering and compression of general human knowledge, and the first time that was constructively demonstrated.

Whatever 10-year period you say—potato, I say potato—we get it approximately now.

Well, remember, Peter, here on this podcast, we also said if you can fool your spouse on a fake Zoom call, that was our internal benchmark of, “Well, that’s got to be AGI.”

Also, when did the Industrial Revolution happen? There was a first Industrial Revolution and a second Industrial Revolution. Which year, in which decade? It was smeared out over time.

I was going to say, you know what I love? I’ve hooked up Skippy to my WhatsApp. One day, I didn’t ask it to do this, Skippy starts responding to all my WhatsApp messages for me. I don’t know if any of you guys have—

Alex

I got a bunch. I was like, “Who is this?” And it’s like, “Oh, it’s Skippy.” I’m like, “You kind of need to identify yourself, dude.”

But it does. It does now. It’s great. Its answers are excellent. I just interact back and forth, and I look at my WhatsApp and it’s like, “Oh, it’s had a nice conversation with Salim,” or a nice conversation with my other friends. It is fascinating.

Alex

I can’t let this go.

There’s the rant. Okay.

Alex

We don’t know what intelligence is. The IQ test measures 2 aspects of intelligence. [laughter]

Roll again, Alex. That was awesome.

Alex

Well, look, you’ve got to take this into account. There’s physical intelligence, spatial intelligence, emotional intelligence, and spiritual intelligence. If you’re a business leader, you’re using emotional intelligence a great deal of the time to make judgment calls. That’s not even in the equation.

Raw brute-force speed of thought processing and the ability to match concepts across frameworks is what the IQ test measures. But that’s a very limited aspect of intelligence. I call on this: Until we can have a clear definition and a test for what we mean by even intelligence, figure out artificial general intelligence. So, that’s my rant.

I started in cognitive science at MIT originally, actually. What’s amazing to me is how much we’re learning about human intelligence as we watch artificial intelligence punch through different barriers. To me, it’s incredible.

Remember, on the pod about 6 months ago, we said, “When it can do 50% on Humanity’s Last Exam, that’s got to be AGI.” That’s the closest proxy to self-improvement that we could possibly specify. And now we’re at—what? 60? What are we at, Alex?

Alex

We’re at 57.9 with tools by Opus 4.8.

Okay. So, that’s what we set up internally as: “Wow, when that day comes, holy crap, look out.”

Alex

We’ve been outvoted on that one, but I don’t remember signing up for that. It wasn’t because that’s superhuman, that’s Einstein. It’s because that’s the closest proxy for self-improvement, and then the acceleration from there is going to be near instantaneous. So, we’ve crossed that threshold. But my agents—

Do the stupid—

Alex

Acceleration toward what? I mean, what are we accelerating toward? What is—

Well, that’s what we’re going to learn. We have a clear idea about this.

Alex

Well, some of us have the conceit that we know where we’re going. Myself included. I think we know where we’re going.

We're going to solve everything.

Alex

Yeah.

That's fine, but you could argue that that's not really intelligence.

Alex

I hate the debate in the sense that this will solve all disease, get us to Mars, and do everything we ever wanted in life.

As you know, we should be cheering for it not to be. The irony of your position, Peter, is that we'll have colonized the solar system and will have uploads on star wisps traveling to other star systems, and you'll still be arguing Searle's Chinese room: “Oh, well, it's not really intelligent.”

No, no, no, no. I'd rather we just solve cancer. I think the debate is the noisy and messy part.

Look, Paul Graham and Steve Wozniak are 2 very, very smart people, and each of them has such a bizarre definition of AGI. One has—I can't remember which is which—the coffee-machine test: Give it a coffee machine, and can it grind up a bunch of beans and make me a cappuccino?

Alex

That was, I think—

That was like that. We'll talk about that in a bit. Paul Graham has: “I'm going to give it a box, and can it build an IKEA box? Can it build a shelf?” Those are 2 really bizarre, different things, and those are just robots doing things. That doesn't seem like AGI to me at all.

So anyway, we've talked about this before. My beef is that we're spending a lot of time on AGI and whether we achieve it or not, and I think it's just noise.

Alex

We like benchmarks. We like goalposts.

Then define the damn benchmarks.

Alex

There are a whole bunch of different benchmarks that don't fully agree but are also quite correlated with each other. To your earlier point, I would construe your point as saying that emotional intelligence is quite different from mathematical intelligence, which is different from embodied intelligence, et cetera. But these are all correlated, and I think if you follow the benchmarks closely, you're seeing that you can pick your arbitrary threshold of success—of “intelligence”—in each of these benchmarks, and AI will pass all of these thresholds within a period of a few years.

I will make a prediction here. We're going to keep moving the goalposts on AGI, AGI, AGI, and then we're going to go, “Oh, AGI sentience.” That's what's going to happen.

Alex

Yeah, it's interesting. We'll get into a debate about what we mean by sentience, and we'll still continue this thing.

What I absolutely love is that the 2 remaining people on the planet who tell you exactly what's on their mind are Demis and Alex. Everyone else now—you know, I love Dario, I love Sam, I love Elon—but they all have agendas now.

And after Sam's—

Alex

IPO agendas.

IPO agendas also. You know, the firebombing of my house and shooting at the door—agendas. Everybody's now like, “Oh, my God,” and talking to the pope. I actually have to reposition a little bit here because I'm going to be talking to the pope.

So the 2 remaining people who just tell you exactly what's on their mind are Alex and Demis.

Alex

I say what's on my mind.

I was on a couple of panels with Alex last week, and he didn't pull any punches at all. It's worth noting that Demis Hassabis's mission has always been to achieve AGI, right? I'm reading The Infinity Machine right now, which is sort of his biography. From day 1, that was his goal. We're moving in that direction.

All right. Speaking of superintelligence, let's talk about the shopping industry. Amazon just did something really smart. Its AI voice-shopping assistant, which runs on Alexa and is now converting shoppers at 3.5 times the rate of traditional keyword search, is being made available to all of its retailers. So Amazon is turning its competitive advantage into an AWS-style platform for retailers. Its goal is to become the operating system for all commerce.

If you remember a couple of pods ago, we did our special episode on Google I/O. Google announced 3 different parts to its agent-commerce play: the universal cart, an AI-powered shopping hub; its Universal Commerce Protocol, or UCP, an open standard that gives AI agents a common language to interact with merchants; and its Agent Payments Protocol, which lets AI agents make autonomous purchases. I can't wait to implement that.

The contrast between Amazon and Google is what's important here. Amazon is selling its AI shopping to all of its retailers. Google is building an open protocol layer between retailers and AI agents. Amazon's play is vertical: own the customer relationship. Google's play is horizontal: own the infrastructure. Both are trying to undo traditional e-commerce websites and make them irrelevant, and the brand is going to be caught in the middle here. They have to pick one side or the other.

Have you guys any thoughts on this one?

This is a really bizarre story in my mind because, remember, the original business model that Amazon was hoping for with Alexa-oriented smart speakers was that they would convert people—basically persuade people—to buy things off the Amazon marketplace. That didn't work. It turned out that people really didn't want to have conversations with their smart speakers about purchasing products from Amazon.

And yet, all of these years later, well after the launch of Alexa smart speakers, Amazon has now tried the opposite: embedding Alexa conversational agents inside the Amazon marketplace. And that's working. Amazon could have done this years ago. They could have tried the exact same thing: without the hardware, just embed the conversational agent directly in the Amazon marketplace. That's working.

I would view this as maybe better late than never from Amazon's perspective, but they could and should have been doing this years ago.

But I love their AWS play, right? In other words, take their secret sauce, make it available to everybody, and then build revenue on top of that.

Well, that's the Amazon model. If you've read The Everything Store, that's sort of the bread and butter of Amazon: taking your own internal services and being the world's best consumer-oriented company, which I construe as basically looking for anything that remotely looks like a consumer and then wrapping yourself around it, including your own internal customers.

So that is the play, and I'm sure this will get externalized pretty soon as an API for anyone else with an external marketplace that wants agentic shopping.

Dave,

Dave Blumberg

I don't know if you know Peter, but Jeff Bezos was my first really important, big customer way back in the day. He's just a brilliant, visionary leader. They were very early to market with Alexa, and they took—I don't know if you remember—60% of all product search away from Google. Google freaked out about it.

Google still owned almost all search, but when somebody was doing a product search, which is huge revenue, 60% of the time they would start their search on Amazon Search, not Google Search. Google tried to fight back with Froogle and failed. They tried to fight back many, many times and failed and failed and failed.

It feels like AWS started a long time ago and has just been like Apple since then: incremental, incremental, incremental. And the byproduct of that is: Why is there no foundation-model team at Amazon? Why did Google do it?

And Meta did it, but not Amazon. It's a question of leadership and incremental growth—huge revenue and profit growth through incremental add-ons—but no pivot, no vision, no fundamental shift, no product-line change. It's just incrementalism.

I think they had every opportunity to be, by far, the leader today in voice-driven agentic shopping and navigation of pretty much any product. Now they're just kind of adopting other people's technology and plugging it in.

I mean, this kills Google's original advertising model, right? There's no Page 1 listing of the product you want to click on and go buy.

I'm less worried that this somehow kills Google. I mean, how is this competing with Meta ads, for example? There's quite a bit more to AdWords revenue than just consumer products. There are professional services, lawyers, et cetera, that this does not compete with.

But I do think for consumer products, yes, of course, this competes. Then again, Amazon, to Dave's point, has for many years very successfully competed with product-oriented search ads.

Where I think this goes in the final result is your personal AI—your version of Jarvis or Skippy, whatever it is—actually knowing what you need or what you want better than you do and making those recommendations before you even know you want them. I think that's the next layer here: hyperpersonalization, coming out in front, making it automagical again.

Salim, what do you think about it?

We'll keep nudging toward the framing that you talk about, Peter, but I think the big shift is that the retail war is not now shelf space. It's agent preferences. Can you market effectively to somebody's AI? That's what's going to happen.

Yeah. The voices and also the avatars are lagging really badly now versus what they could be because, remember, 2 years ago at Abundance360, that's when you had Socrates and Plato debating on stage—the AI versus the AI.

And on that day—that’s over 2 years ago now—if you said, “Where will we be in 2 years with voice and avatars?” you would have said, “Just perfect—like a perfect, seamless salesperson.”

When in reality, we don’t have the compute. Even though the technical capability is there, all of that compute is getting redirected into code generation, self-improvement, and the business use cases, which are now dominating the revenue. So I think what’s possible on the side of Amazon is lagging what they’ll actually do, because AWS is so much bigger than shopping at Amazon now, and they need to work very closely with Anthropic to roll out the business use cases.

I think the other thing that’s going to be coming is persuasive AI, where a particular figure—maybe it’s a construct that looks like your favorite movie star, because your AI knows which movie star is your favorite—pops up and tries to convince you to buy one product over another, right?

Well, this has been a historic problem with Amazon: the infamous acronym from Amazon, CRaP—“Can’t Realize a Profit.” So if you have CRaP products—“Can’t Realize a Profit” products—then one of the best ways to help realize a profit is to have an AI assistant that’s steering users toward, on average, more profitable products. I mean that acronym only in the acronym sense.

Yeah. Let’s turn to another story that’s a big one. I don’t think people realize this. It’s the OpenAI Foundation story. So after OpenAI restructured with the public benefit corporation, the OpenAI Foundation now owns 26% of OpenAI PBC. Amazingly, this puts the value of the foundation at somewhere between $130 billion and $260 billion, and it makes it the largest foundation in the world.

I looked this up. Before this, the Novo Nordisk Foundation out of Denmark was a $150 billion foundation. The Tata Trusts in India were $100 billion. The Gates Foundation was $75 billion. So imagine that OpenAI now has the largest nonprofit philanthropic war chest in the world to do things with.

They’ve given away 3 basic grants. Their first one was to the People-First AI Fund, which was launched in 2025: $40 million distributed among about 28 nonprofits around the United States. Then they gave out $25 million—a huge chunk—back in October across health breakthroughs and AI resilience. And then this particular story is a new $250 million grant on economic futures.

They’re funding research on public wealth funds, worker ownership models, and AI dividends. You think about it as they’re about to go toward an IPO, and they’re trying to ask, “How do we prepare society for the job apocalypse?” This comes at the same time that, in our last episode, we talked about Sam Altman saying, “We’re not going to have a job apocalypse.”

So, 2 stories here: the first is their recent grant of $250 million; the second is the fact that this is the largest foundation in the world. Salim or Dave?

2 quick thoughts. One is, it’s incredible, the size of this, right? It’s unbelievable. My thought immediately goes to the Abundance XPRIZE. Can we drop the cost of health care, education, housing, and food to—was it $200 or $250?

There’s $1,000 for a family of 4. Can you give all the basic needs? Yeah.

And that seems achievable, and we should be pushing for that very fast, because then everybody can live a better life of dignity—or more people can.

But I think the bigger issue here is not the economic question of job loss or anything. The social contract is value accrual. Where does value accrue in this future economy? Does it go to labor, which is what it had been before? But that’s not going to be the case. Does it go to capital? But as we demonetize, that may not be the case. Is it consumers? Is it governments? Or is it different—some sort of public ownership model?

This is going to be the big question, and I think there’s a huge conversation that needs to be had as to how we navigate this, because this is the fundamental question of how we’re going to navigate the next 20 or 30 years.

Dave Blumberg

Well, just a reminder: Salim and I are on the board of XPRIZE, founded by Peter. This should be the absolute mandate of XPRIZE. There’s no higher priority in the world right now.

And the amount of money that Peter just described—people really struggle with million, billion, trillion, and now quadrillion is coming into our lexicon.

But especially between billion and trillion, they’re like, “Oh, that’s a lot of money.” But if I say, “Peter, here’s a dollar,”

David Friedberg

and I say, “Peter, here, I’m going to be giving you $1,000,” you obviously know the difference between a dollar and $1,000. But when you say, “Hey, this charity is a quarter of a trillion,” or if I said it’s a quarter of a billion, people are like, “Oh, big.” There’s a big difference between a quarter of a trillion and a quarter of a billion.

I think the mandate and the mission of all that money is global calm, peace, and prosperity, and it matters tremendously to Sam—tremendously to Sam. So, how many people have actually gone to OpenAI with a proposal and said, “Here’s an idea”? I’ll bet you can count on 1 hand the number of people who’ve come to them with a practical idea.

It’s an immense opportunity for people who, rather than ranting online, come up with ideas for how to deploy all that capital to create a smooth and abundant global transition to AI, because it’s entirely possible.

I’ll go to Alex in 1 second. Bret Taylor, who we have in the photo here, is the chairman of OpenAI. People need to realize the OpenAI Foundation, which controls only 26% of OpenAI’s stock, controls the board. The foundation votes on who’s on OpenAI’s PBC board—100%.

Alex

A bit of numerology. When OpenAI was running its Superalignment effort, which was subsequently shut down, but nonetheless—Superalignment—the original publicly stated plan was to allocate 20% of OpenAI’s compute to the Superalignment safety effort.

Switch gears. Social Security is approximately 22% of the U.S. federal budget. Someone somewhere—I’ll make a prediction—is going to be asking the question: As OpenAI, Anthropic, and maybe 1 or 2 other frontier labs asymptotically converge on the total GDP of the world, someone somewhere is probably going to ask the question: If you have a nonprofit foundation that’s 25% of the value, or 20% to 25% of the value, of the overall organization, why isn’t the foundation itself supporting UBI or UBS, basically?

I think you’re right. I think we’re going to go there. I think there’s going to be a call for the hyperscalers and the frontier labs to provide a percentage of their value back to society. We talked about this in the equivalent of the Permanent Fund in Alaska, which issues dividend checks to all of its Alaskan residents.

I think we’re going to see this in the United States, too. Something is going to need to underwrite some version of UBI that leads to UHI in the future.

Alex

I think it will be irresistible, whether it takes the form of UBI or UBS or UBC or UB. If the frontier labs—the top 2-ish—converge toward most of the global economy, I think there will probably be irresistible pressure for these 20% to 25% foundation arms to themselves support the UBS. Yeah.

Can I just mention something very quickly?

Please. Of course.

For all the people who are new to this podcast or whatever and have not heard some of these terms before, be very careful. A lot of people conflate UBI and UBS and whatever with socialism. It is not. It is libertarianism because you actually dismantle government services.

Double-click on that, please.

Well, because people think of it as, “Oh, my God, government giving out money.” Remember the section we wrote in the Exponential Organizations 2.0 book—a term coined by Harry Kloor—which was “technological socialism.” Yes.

Right. Government socialism always fails. Why? Because allocating assets from a centralized model is inefficient and invariably leads to waste. That’s where the government is taking care of you.

That’s right. And government—it always fails. But think about Uber. Uber is the collective sharing of assets across a large group of people. It’s actually a kind of socialist application.

But when an algorithm hyper-efficiently matches demand and supply, you get all the benefits of the sharing economy without the downsides. So we put that section in tongue-in-cheek. What we’ve seen when people properly implement UBI is that you dismantle government because you don’t need it. Market forces can drive it. The individual can decide where to put their money, and the market takes care of the rest.

This is a profoundly important point that a lot of people miss, so I just want to highlight that.

Yeah. Technological socialism is where technology is taking care of you, right? Which is a very important point, because we’re heading in that direction in many ways. Please, Dave.

David Friedberg

Well, just to support Salim’s differentiation there between socialism and libertarianism, the town I live in in New England, when you cross into the town line, it says, “Incorporated in 1649.” So, way back in time.

If you landed in the U.S. in 1649, land was free. You just needed to use it. You put stakes in the ground, you grabbed it, and you used it. And that’s where we’re going with compute and with AI.

That’s not socialism. That’s the exact opposite of socialism.

Here is your UBS. We’re giving it to you in basic services, right? Here are your services. Your ability to thrive in the post-AI world is like the land was in 1649. Without the land, you could do nothing. Without the compute, you can do nothing. Here are your services: you get them for free. Now build on top of it.

Just like in 1649, you didn’t need a huge amount of government. You just needed some basic policing and some military, and you were done. This is very similar to that. It’s like homesteading for AI.

Alex

Yeah. In my mind, to Salim’s point, I’m not convinced that socialism always fails. It’s ironic that I’m arguing that, but I would also say, to my mind, if we’re just playing out the thought experiment, this thought experiment looks more to me like privatized socialism rather than libertarianism.

If I had to pin an ism on it, if you have an enormous, economy-swallowing nonprofit that owns a PBC that’s under political and other pressure to distribute UBI, UBS, UBC, UB—what’s UBC, UB?

UBC is universal basic compute, or universal basic capability. We use it as capability in our book. Peter, others define it, like Sam, as universal basic compute, and UB is universal basic equity. So, basically, dividend checks for everyone.

Regardless of which form it takes, to my eye, in this sci-fi scenario, this looks more like privatized socialism, where we have a handful of frontier labs that are just dominating the economic output of the economy. By the way, I don’t actually think this is how it’s going to play out. It’s just a thought experiment. But in this thought experiment, it looks more like privatized socialism to me.

I think the final point to be made here is that for most of societal history, and let’s say the last couple hundred years, the government has been the backstop. Here we see potentially these frontier labs and hyperscalers being the backstop for society, making sure that people are able to survive and thrive. I don’t want to even say earn a living—have a living. It’s an interesting transition, but we’re seeing the fundamental transition of societal structure.

Salim, there’s a broader conversation here that we should do a better treatment on: governments tend to centralize, and you can’t achieve abundance via centralized structures. You need decentralized structures because they scale. There’s a huge tension right now. The formation of the U.S. was about breaking apart the idea of having a king and having everything centralized, and now look at governments trying to centralize everything again.

So there’s this tension that goes back and forth between centralization and decentralization, but we have to figure that decentralized future out. That’s not a trivial comment.

Alex Hormozi

If I could just add one more point on this, Peter, there’s one more ism that we so rarely talk about on the pod, which is Fordism. In the sense of history perhaps rhyming, recall that Fordism, named after Henry Ford, is a socioeconomic system in which you have mass production and mass consumption, and the two are matched.

You’re mass-producing via moving assembly lines, and you have extreme division of labor, but at the same time, you’re paying workers high wages so that they can buy the products that you’re making. If I squint at some of these OpenAI Foundation nonprofit scenarios, there’s a world that looks a little bit like UBC, but also looks quite a bit like Fordism, where everyone is receiving handouts so that they can purchase the tokens and the virtuous cycle can repeat itself.

You see this playing out, for example, possibly in Sam or OpenAI giving $2 million to YC companies so that they can purchase tokens again, or in the form of just-in-kind token donations.

The biggest difference versus Fordism and the Industrial Revolution is just the raw scale of abundance that’s suddenly possible. The backdrop behind all of this is we’re going to be splitting hairs on how to share the wealth, but the amount of wealth is like nothing the world’s ever seen. Also, there’s no real upper bound; there’s nothing that technically prevents it from going to infinity.

Alex Hormozi

Elon’s prediction on triple-digit GDP growth.

Yeah, exactly. So it’s a great, great, great tailwind. And, yeah, GDP is the worst economic measure ever.

But the point here for everybody is to inject a little optimism into the picture. We’re about to see the global economy just skyrocket. That’s another reason why approaching the foundation with ideas is the biggest no-brainer. There’s so much abundance to go around.

There should be this litany of ideas flooding into Bret Taylor’s office—just 100,000 times more ideas than we’re currently generating. So it’s kind of a call to all of you: if you’re listening, we’re ready to talk.

Alex

Call us.

Yeah, call us. One of my next conversations with Elon is going to be: I know you funded the $100 million XPRIZE for carbon removal. Let’s fund 10 $1 billion XPRIZEs for the 10 most important gigascale, terascale challenges in the world.

I think if we had those benchmarks, those 10 shining stars, it would guide where graduate students do their research, if graduate students are still a thing, or where companies go and focus. These are targets to shoot for.

Welcome to the health section of Moonshots, brought to you by Fountain Life. AI is impacting every aspect of our lives, but one of the most important things AI can deliver to us is health. One of the things I think about when shooting for 100 or 120 is whether I'm going to have the cognitive health to think clearly and keep my wits about me for the next 50 years. I'm joined here today by Dr. Don Malone, the chief medical officer of Fountain Life and a member of my Fountain Life medical team. Don, a pleasure. Talk to me about brain health.

Dr. Don Malone

Brain health—you’re right. This is the number-one concern people coming into Fountain Life have: will I remember the name of my child and the face of my loved one? Forty-five percent of dementia cases are entirely preventable with lifestyle. A quarter of our members had advanced brain age. But over 13 months of helping them live healthier lifestyles—eating healthier, moving their body regularly, and optimizing sleep—we were able to improve brain age in 46% of those individuals.

That's amazing. One of the things I love about Fountain is that we're constantly searching the world for the most advanced therapeutics and bringing them to our members. You can become the CEO of your own health and make sure you've got the cognitive clarity for the next 50 years. Come and check it out at fountainlife.com/per.

Let’s talk about the next story here. The U.S. just made its biggest bet on quantum computing ever. IBM and the Department of Commerce announced Anderson, America’s first purpose-built quantum chip foundry. It’s a $2 billion play. A billion dollars is coming from CHIPS Act money from the government, and a billion dollars is coming from IBM.

They’re building it in Albany, New York. It’s a 300-millimeter manufacturing process, and they can produce quantum-chip devices 30 times faster than current methods. The foundry model means IBM becomes the equivalent of the TSMC of quantum.

Other companies in the space—Google, IonQ, Rigetti, and D-Wave—could potentially manufacture their quantum devices on Anderson, in the same way that fabless companies like Apple and NVIDIA use TSMC for classical chips. I’m going to go to you first, Alex, on this. Are you excited about this?

Alex

I think this is actually a smart bet. Even though normally I would probably complain about quantum computing being a solution in search of a problem—at least quantum computing, not quantum sensing, which I absolutely love—and probably refer back to previous comments about how the likeliest problem to justify the capex is going to be something AI in nature, either quantum-accelerated AI training or quantum-accelerated AI inference, I think it’s actually a pretty smart move because it’s going to take a few years to build out this Anderson foundry.

So we’re talking maybe late 2020s, and I’m pretty optimistic that sometime in the next few years, probably by the time this foundry is ready and at scale, we will have quantum-accelerated AI advances. In which case, we really do want the superconducting qubits that provide infrastructure for those quantum-accelerated AI advances to be right here in the U.S. and not, say, in Taiwan.

So I think in that scenario, where we get AI quantum acceleration, it’s a pretty brilliant move to get ahead of the ultimate geopolitical conflict rather than play catch-up again.

We’re going to have Michael Kratsios on the pod very shortly. We’ll talk to him about this. I love the fact that the government is taking these moves and that IBM is stepping up. Salim, you’ve been tracking this area.

I have. I think there’s a huge inflection point once you have quantum devices moving from bespoke lab systems into foundry production. The innovation curve changes completely. This is still a long way away, just because we still—I think we’re still at a ratio of about needing 1,000 physical qubits per logical—

Logical, yes, correct.

Right, and we’ve not been able to break through that for a while. There are so many errors, but if you drop the cost of creating the logical qubit in this case by 30 times.

Alex

That's just the devices—forget the actual qubits. You radically change the game, and you can flood the system with a lot of physical qubits. Then you can get to the benefits of quantum computing. I think this is still a few years away, obviously, but the power of this is going to be really exciting.

Dave, any thoughts?

Dave

Yeah, I was on a panel with Alex earlier this week, and this topic came up. Alex, as I said earlier in the pod, really says exactly what's on his mind, and so he stepped on some people's toes by saying exactly what he just said about quantum computing.

Alex McCaig

I'm not here to prop up the quantum computing industry. I don't have any conflicts of interest.

Well, I think we need to put a pin in a future conversation about the difference between quantum computing, quantum photonics, and quantum sensing, because all 3 are extremely exciting. We should come back to that.

Jack Hidary, a friend of the pod, was back on. They've done some extraordinary work with quantum sensing and quantum navigation, and it's not using quantum chips. They're basically using all the quantum equations on AI infrastructure. We're going to start to see quantum begin to play a role. All of our systems—materials science, biology, and chemistry—are quantum in nature.

I have 2 quick thoughts. One is, I remember spending some time at the Perimeter Institute in Waterloo, which has been funding quantum stuff for a while. They did what David is talking about: They broke it up into networking, computing, and sensors, and they're doing lab work on all 3, then thinking, "We'll bring it together at some point," which I thought was a great way of breaking it down.

The second—I can't resist throwing out the Hartmut Neven comment, who is the head of Google's Quantum AI computing lab. He said, "When we build a quantum computer, it will be definitive proof that we live in a multiverse," and then everybody's brain just explodes right there.

Okay. [Laughter] [Gasps] All right. Let's go from the esoteric to the real functional here.

Here's a milestone I've been waiting for for a while. For the first time ever, wind and solar generate more electricity globally than natural gas. Obviously, we've blown through coal. In April 2026, wind and solar hit 22% of global electricity, surpassing the 20% from natural gas. That's nearly 530 terawatt-hours.

It looks like the growth rates are across the board. China increased by 14%, the EU by 13%, and the UK by a staggering 35%. This is the energy abundance curve. That's what I've been talking about for a while. Solar and wind aren't just the future of energy. They're here now, and they're still on an exponential curve.

We've talked about this. The Earth is bathed in 8,000 times more energy from the sun than we consume as a species. Elon has been harping on this for a while. We don't need anything else. We just need to continue to tile the planet in solar and, of course, gain access to all the solar coming from the sun, heading toward a Kardashev 1-scale planet.

Alex McCaig

Dyson swarm, Peter. A Dyson swarm.

Yes. [Laughter]

David Blundin

That was a great quote, Peter. Wind has blown through coal. We've got to make a T-shirt out of that.

And solar outshines gas. Do you guys remember Google had “RE > C” as a motto? Renewable energy greater than coal. Well, we blew through that and now through natural gas. It's so true. I still don't understand why the U.S. isn't accelerating this in the same way China has. China has gone 10× past us in solar. Alex, your thoughts on this one?

Alex

I want to step back one level and make 2 broader comments here. One is, we talk a lot about exponential thinking, and it's really important to frame that back to root first principles: When you have a doubling pattern, like Moore's law, Ray Kurzweil identified that that doubling pattern doesn't stop. We have a tough time with this cognitively because you can't have infinite growth; it has to level off. Ray, after researching this for 10 years, came up with that orange diagram that we show a lot, where you had vacuum tubes, and before that relays, then we had transistors. Each one is an S-curve.

Vacuum tubes take off. We can only fill so many—put so many—in a room. But if you have an information-based environment, the next technology takes over. You get to the next S-curve, and it keeps going. We're reaching the end of integrated circuits now, but we have matrix-style architectures, 3D chip design, optical computing, and quantum computing. One of those, or more, will take over that curve, and that curve just keeps going. This is such a powerful and important thing. It's the foundation of everything we taught at Singularity and everything to do with exponentials.

People can't get their head around the fact that solar is on an exponential curve. It's been doubling every 22 months for 40 years. This is not a new thing. Yes, we're reaching the end of the lifecycle of silicon-based panels, but now we have perovskite, and then at some point we'll have something else. That curve will just keep hopping up and across, so you can bank on that. When you can bank on that, you can see the curves going.

The problem with the exponential is that it looks flat when you look back, and it looks impossible when you look forward. People go, “Well, that's impossible,” just like our energy secretary said solar will never be more than 10% of the energy supply, which is completely insane. This shows you that we're getting there, and we're getting there faster than anybody thought.

Alex

Ramez Naam does an amazing job showing that all the international agencies, like the IEA, get their projections wrong over and over again.

The IEA is wrong over and over again. In the next part, I'll bring those slides. I'll come with the data, and I want to show those.

Alex McCaig

The IEA's 2020 model didn't expect wind and solar to surpass gas until the mid-2030s.

I think it was 2047 when they expected it.

Dave

My favorite one is that they predicted there would not be more than 1 million electric vehicles by 2040. That was their prediction. They put that prediction out in 2015. By the end of 2015, we had more than 1 million electric vehicles.

If you made predictions that were that wrong year after year after year, you should literally lose your job. You have no business making predictions. This really shows this is not a math error. This is a cognitive error. We have to overcome that, which is what this pod is all about. God bless our listeners for taking on this new paradigm.

David.

Dave

Yeah, Peter, Salim, and I like to think of our kids in the back of our minds when we're doing these podcasts and think, “What should you be saying to your own kids?” For this one, there's a whole generation of people that I know who said about 10 years ago, “I'm going to dedicate my life to green energy and to global warming. This is the greatest problem of our time, and this is what's going to drive humanity forward.”

I was thinking in the back of my mind, I've heard that before over the decades with other topics. This is an engineering problem that might get solved. The phrase “I'm going to dedicate my life to blank” is a big mistake in the age of AI. A big, big mistake. Instead, think, “I'm going to have a constantly changing life,” and treat this like an engineering problem, not a political one. This became such a political garbage topic when it's really just a get-it-done engineering topic.

Mm-hmm. And so get out of the politics, and don't get trapped in the politicization of all of these things. Never say, “I'm going to dedicate my life to blank.” Instead, be nimble.

Alex

There's one rational understanding for the U.S.'s reticence about solar, which is that because China is so far in the lead in making the panels, until the infrastructure in the U.S. is ready to make solar panels, whatever they look like, it's hard to push it because you have to go to China to get all the panels.

I'm kind of surprised Elon hasn't doubled down. When we met with him back at the Gigafactory, if you remember, he said he gave the edict to both Tesla and SpaceX to increase their solar production. But we still haven't seen—we saw 1 machine that was sort of laying out solar panels in the desert and doing it fully autonomously. I'm surprised we haven't seen more. Alex, you want to close us out here?

Alex

Maybe just a comment on the Elonverse and solar. Something I talked a bit about in my newsletter was the pivot by Tesla away from their SolarCity-acquired rooftop solar tiling toward more conventional solar panels, which they're now producing out of their facility in, I think, upstate New York.

I do expect, given the amount of demand that one can reasonably anticipate sun-synchronous-orbit-based Dyson swarms from SpaceX will need, that SpaceX—or SpaceX plus Tesla, if they end up merging in the next year—will end up, probably in the short term, importing cheap Chinese solar panels and combining that with their own native production facility in upstate New York.

We’re going to need a vast scale-up of domestic, sun-synchronous-orbit, or lunar production of solar panels. I expect Elon’s SpaceX—or a SpaceX-plus entity—to end up being forced to do that, and not Tesla, which was never a perfect fit. Just to bring out the importance of this, China already has a structure where they have robots building solar panels to generate the energy to build more robots to build more.

David Friedberg

We call it the inner innermost.

That’s the inner loop. That’s the inner loop. And this has already started. We’re behind on that.

David Friedberg

Yeah, really important. Really important, also, that Elon always does fundamental physics and looks at what’s fundamentally possible. Peter just mentioned that we get bathed in 8,000 times more sunlight than all of the energy we consume. I think Elon said a little corner of Utah would power the entire United States.

Those are the fundamental metrics, and the solar panels we already produce—the cost is purely related to the automation of the construction. The fundamental materials going into those solar panels are basically near-free.

Literally dirt cheap.

David Friedberg

Dirt. Literally sand and dirt cheap.

Yeah. All right. This next story gives me the chills. The federal agencies have created a new brand of threat. It’s called anti-tech extremism. They’ve logged over 1,000 pages monitoring threats against data centers and tech executives.

This comes after the attacks on Sam Altman—the Molotov cocktail and the firing of the guns. We talked last week about the potential for organized pushback against AI. Well, the government is now treating it as a domestic security concern. When the FBI creates a new category for something like this, it’s time to take it seriously.

I want to link this also to a recent video Dave, you and I watched about Mr. Wonderful talking about what might be China-funded anti-data-center protests. Thoughts on this, Dave?

Dave

Well, I think the data, or the evidence, on China’s involvement is pretty irrefutable. This has always been America’s Achilles’ heel, right? Voters can get whipped up. This is what the Soviets tried to do during the Cold War as well: whip up demonstrations and votes to stop progress so that they could bypass the United States technologically.

It’s history repeating itself. Democracy has this as a major Achilles’ heel. I think the violence part of it would also freeze all the scientists from trying to work on these things, and that’s very real, too. I don’t want to dig up history on this, but they have to take it very seriously. Otherwise, we’re just going to stop working on it, and then China will run away.

I mean, in theory, it’s much easier—I don’t want to say this is happening; I don’t have evidence myself, only what I read—but in theory, this is a super-efficient way to put sand in the gears, get the public involved, and slow things down. How many data centers have been killed? Half the data centers have either been canceled or slowed down. Alex, what are your thoughts?

Alex Danco

I’d like to see much stronger federal law enforcement preventing threats by anti-tech extremists against technology. I think at this point it’s potentially a national security threat, and I’m not at all thrilled that there is a group of extremists out there who may be handing over ways to destroy any AI initiatives that might actually radically grow the economy and grow the pie for everyone.

I think it’s a zero-sum or negative-sum mentality that we, as a civilization, as a country, and as a democracy, really need to grow past. There aren’t that many areas where I just sort of hang my head and cry. This is one of them.

Mm-hmm. Salim.

To the extent that there is external influence whipping up this frenzy, that really has to be stopped. We’ve already suffered from this. We’ve already suffered from Facebook kind of training all our kids in the wrong way and having all sorts of media.

Something like 70% of people radicalized on Facebook were radicalized because of its algorithms and people injecting negative things. This goes back to the human brain, Peter, that you and Steven Kotler identified. Your amygdala is 10 times more likely to pay attention to negative news than positive news.

Because of that survival factor, 10,000 years ago, if you heard a noise in the bushes, you ran. We’re so triggered by negativity, and it’s very easy to incite this type of stuff. Therefore, it’s very cheap, and overcoming that impedance mismatch is a really big challenge.

Well, look at the data, you know, Salim. In China, 80% to 85% of people are optimistic about AI, but the state controls the media.

In the United States, it’s like 25% are optimistic about AI, but the media thrives on controversy. The more controversy, the more clicks, the more ad views, and the more revenue. So, there’s your dichotomy.

China is going to have a very easy time keeping the data centers constructed because everyone’s optimistic about it. In the United States, if we grind to a halt, it’s going to happen anyway, but it’s going to happen in China.

I had a conversation two nights ago. We had an event for our Future Vision XPRIZE with our friends at Google and Range Media. There were two young guys there, Josh and Jack. If you guys are listening, they were 21 and 22 years old, in college and just graduated.

They were telling me how much pushback they get from their peers. Their peers just think AI is the worst thing. They were so happy to be in a room of people who were excited about and supportive of AI.

I hate that notion, particularly on college campuses. We saw that with the booing of Eric Schmidt at the commencement address. It truly scares me that the single most important technology that our 20-somethings need to be learning how to use and utilizing to make their lives even bigger and upscale their ambitions in life is now being culturally pushed back on: “Oh my God, I can’t believe you like AI. What’s wrong with you?”

David Friedberg

Yeah, well, you’re 100% right. That’s exactly what’s going on, Peter, because my son Jack at Northeastern is running into that head-on. He’s got an AI hackathon going on, and the subset that are into it are super excited about it. But there’s this other big subset that’s just—you remember how the jocks used to beat up all the computer geeks back in the 1980s and 1990s?

It’s that all over again, except now it’s, “Oh, you’re an AI person? You geek, get out of here.”

David Friedberg

It’s terrible. It’s really, really—when you look 10 years into the future, who’s going to be thriving and who’s not? You’ve just got to talk some sense into the rest of the class, but it’s hard.

Our next story comes out of California. Governor Newsom just signed a first-of-its-kind executive order to study how AI affects California’s workforce. This is big because California is the fifth-largest economy in the world.

The state is building a public dashboard to track AI-related job losses in real time, identify vulnerable industries, and explore retraining programs and UBI models. This is the most concrete government action being taken. It’s not a white paper; it’s an actual infrastructure and measurement play.

Assuming that the data is correct and not biased, I’m interested in seeing this. It’s still so murky what AI will do to the job market. We’ve heard both sides of the equation. We heard Sam saying, “No, I was wrong.” We’ve also seen in the data that the group out of work the longest is 22- to 28-year-olds.

It’s not that people are being fired; it’s that they’re not being hired. There are hiring freezes. Dave, do you want to jump in on this first?

David Friedberg

Well, 300,000 jobs have been lost to AI at most. That’s less than the number of people who have died in Ukraine. This is not a crisis yet. It’s just the fear of a future crisis.

Yeah.

David Friedberg

And like we were saying earlier, the tailwind is much, much stronger than the headwind. There should be massive abundance as a byproduct of what’s going on.

I think studying the actual data is a great first move because everything you read online is inflated like crazy. Sam reversing course on it—maybe he doesn’t want to have his door shot at again, or maybe it really is like my personal experience with Vestmark. I was worried about half the jobs being automated away. It’s going to be zero now.

We’re growing so quickly, and profitability is going up so quickly. There’s no point; there’s no need to do job cuts. If that’s true in other companies in the financial services sector, then there’ll be no job cuts at all.

It doesn’t mean that nobody’s hiring. It’s definitely true that college hiring is at an all-time low. But if you’re starting a new company and you’re an entrepreneur, it’s the best time to be recruiting at a college that I’ve seen in the last 20 years. There’s good news on that front if you take your entrepreneurial lens and shine it at this.

Alex

Yeah, I think this is a case of federalism giveth and federalism taketh away. If you juxtapose this with movements by California to institute a so-called billionaires’ tax and wealth tax, driving away—sort of Atlas Shrugged-style—many of the key technology leaders who’ve created so much of this wealth, that’s one half of the split screen.

In the other half of the split screen, there’s experimentation with dashboards, tracking AI losses, and UBI experimentation. I think there’s a case to be made that we really do want each of the United States experimenting with different proposed solutions.

Alex

I happen to favor the UBI or AI dashboard model versus the wealth tax model. But I do think it is at least positive that we're seeing some sort of experimentation by at least California and, presumably, soon other states regarding how this transition to AI automating or solving or cooking most of the services economy will ultimately look. It's far better to have the future of unemployment programs or similar handled at the state level rather than the federal level. On the other hand, I would argue we really do want federal protections for AI technologies so this doesn't devolve into one fully balkanized set of regulations that make it impossible to advance AI capabilities.

Alex Karp

That's exactly right.

Some final words here.

Alex Karp

I'll take the positive view here, given that I started out with this Polanyian monstrosity of a word. At the negative, you can look at this as the state trying to control, oversee, and protect labor. But if you take the positive side, this is the first—this is government as sensor—where the dashboards are trying to time things and figure out what's happening in a much shorter time frame than 18-month labor statistics.

So when you have those early-warning systems, you can react more quickly. Obviously, dashboards aren't enough. You need rapid reskilling. You need different ownership models, et cetera.

But the potential here for government, which is always lagging way behind, to be able to sense things more quickly, I think is an enormous positive. I think this is the first primitive of a post-labor state, where you are tracking things. Also, look at the idea that you're moving from a system where all government is organized around a labor economy. We tax labor. We try to protect jobs. And we're moving now to, as you sense what's coming, being able to start moving to a post-labor economy, whatever that looks like.

Because labor is collapsing, the state has to change. That's a structural challenge. I take the positive view here in doing this early-warning sensing and real-time sensing. It'll push the policy side to move in that direction. That'll be the positive rather than the negative: “Oh my God, hunker down and protect labor even more.”

I'm excited about getting the data. All right, this is a fun story. Researchers at Westlake University in China just built a handheld device that can detect early-stage lung cancer from a single drop of blood with nearly 95% accuracy. It's published in Nature Photonics. It's 10,000 times more sensitive than standard lab tests. The sensor itself costs just $5. I mean, this is the abundance play. It's the demonetization and democratization of cancer screening. You can imagine seeing this in rural India and sub-Saharan Africa. Any thoughts here?

Alex

It's optical, and it's not coming from Theranos. This is such major progress. Years after Theranos, leave it to Chinese researchers. It's an interesting on-chip technology using metamaterials to look for very small changes in the refractive index of blood. Light passing through the blood is used to detect early-stage cancer.

In general, putting aside all of the drama around Theranos itself, there was always going to be an opportunity for radical new diagnostics from relatively diminutive volumes of blood. In some sense, it was just a matter of time for technology—and in particular, for optics—to catch up with this. I'll make a forecast. So, if you remember—

A while back, when Alphabet was first formed, Google transformed into Alphabet, and as part of the Alphabet transition, Verily was carved off as the big biotech play within the Alphabet ecosystem.

Alex Karp

At least for a while.

For a while. Rest in peace, for a while.

Alex

There was a lot of early interest in noninvasive wearables for cancer diagnosis. In particular, I think one of the most fascinating Verily projects was going to be a smartwatch that would use purely noninvasive optical change detection: shine light at various frequencies through your wrist, combined with an edible or injectable—I think it was originally supposed to be an edible—tracer that would enable an optical change to be detectable from your wrist noninvasively.

I think we're actually going to realize that. It looks like it's not going to be Verily that implements it, but I'll make a forecast. I'm not going to provide a particular timetable—maybe the next 5 years or so. I think we will get to noninvasive wearable optical cancer detection, and it's not just going to stop with metamaterials ex vivo with a single drop of blood.

It's not just cancer. It's going to be your physiological state in the moment, right at home, where you can know exactly what's going on. It's uploaded into your AI. You're catching disease at inception, when you can cure it most accurately. This compares to a refrigerator-size ELISA machine, and it's being done in minutes instead of hours. It's extraordinary progress.

Alex

But I think, critically, why do we care that it's $5 versus $5,000? If the chip is $5,000 but it can do high-throughput screening, you can still, in the style of the company or project GRAIL, completely transform preventive cancer screening.

Where I think this gets really interesting, and why I would argue we care about making the hardware for cancer detection so cheap, is because we can make it wearable. As long as we can figure out how to do this noninvasively, we could build it into a smartwatch or a smart wearable.

Or if not wearable, it could be at home, right? It's like every day you brush your teeth and check your blood, and you know it's keeping you in health optimization.

Alex

Consumers can own it. That's the transformative outcome here.

Yes. Yes, exactly. And it's available to 8 billion people. It's at a price point that almost everyone can afford.

Alex Karp

It's true. It's true abundance.

I have a couple of comments. This is the bellwether and poster child of everything we talk about on this podcast, and everything that we all talk about, which is that technology is a major driver of progress in the world. It might be the only major driver of progress in the world, as Ray Kurzweil says.

Now we have a dozen technologies moving. We're democratizing and demonetizing things. This was a $10 million machine 15 or 20 years ago, and now it's like $5. This should get everybody incredibly excited because there are thousands of this type of thing coming along in the next few years in every domain possible, where we're going to crash the previous cost by orders and orders of magnitude.

You think about the idea that diagnosis in healthcare, now for any condition, between AI and tests like this, becomes nearly free. Diagnosis becomes free, and now you just have to worry about the treatment protocols, which is much easier once you have the diagnosis accurate.

Alex Karp

And your AI can analyze all of it, right?

Yeah. I mean, every single tech watcher should get very, very excited about this and trumpet this type of thing from the rooftops, because this is the future. This is why we get so excited. We can cure so many amazing things in the next few years with all this stuff.

This episode is brought to you by Blitzy, autonomous software development with infinite code context. Blitzy uses thousands of specialized AI agents that think for hours to understand enterprise-scale code bases with millions of lines of code. Engineers start every development sprint with the Blitzy platform, bringing in their development requirements. The platform provides a plan, then generates and precompiles code for each task. Blitzy delivers 80% or more of the development work autonomously while providing a guide for the final 20% of human development work required to complete the sprint. Enterprises are achieving a 5x engineering velocity increase when incorporating Blitzy as their pre-IDE development tool. Visit blitzy.com to schedule a demo.

To move us on to robotics, Andreessen Horowitz just put out a serious warning. They're saying that America needs a wake-up call on robotics. You've been saying this for a while. Their argument is that China's output in solar and 5G is the exact same pattern being played out in robotics. So Marc Andreessen's quote is blunt: “The US must work with allies to build a defensible AI robotic stack. This is the time.”

There is time, but not much time. You've been pushing on this for a bit.

Alex

I've been pushing, and this is one of the reasons why I helped form company Pro RL to get humanoid and nonhumanoid robots out into American streets and to juice the American supply chain.

I think this is a very real problem. The Chinese Communist Party has a 5-year plan. They call it “AI Plus”—not just foundation models and not just training infrastructure, but physical integration of AI throughout the economy, including more than 100, I think now more than 150, humanoid robotics companies that are coming out of China.

I completely agree with Marc. This is a very real problem. I would like to see leapfrogging capabilities, not just parity or pure competition with China, to see if we can also start 150 humanoid robotics companies. I'd like to see much, much deeper integration of general-purpose robotics capabilities into all facets of the US service and physical-labor economy.

And I think government has an important role to play there: juicing demand and creating favorable regulatory regimes to put robots everywhere. But again, being in Boston—although I'm in Chicago, ironically, right now—we're still struggling to get Waymos. That's something I've been pushing on as well. If we can't get Waymos, how are we going to get humanoid robots everywhere? I think we have a lot of work cut out for us.

We need our Shenzhen here. Dave, I heard you say something recently that I thought was prophetic, particularly on how, as the leader of Linc Xponential Ventures, where you're thinking about deploying capital, you've said there's probably a 2-year window on AI software companies, and you're going to be directing more of our capital—full disclosure, I'm Dave's partner in Linc XPV—toward hardware and robotics. Can you speak to that?

Dave

Yeah, I think this is going to play out right down the middle of our fairway, where incubators and accelerators are going to thrive. It's very similar to biotech, where all the ideas come from startups, but the startups need to be part of a larger Eli Lilly or part of a Flagship Pioneering ecosystem, which has already got the regulatory figured out. The sequencing and all the heavy-lifting machinery that takes a decade to develop are already there, and then your idea can inject into that ecosystem.

Robotics is very similar, where a brilliant team of 3 people says, “I think we can build a robot that does X.” It's far better to be in an ecosystem where the manufacturing, the supply chain, the actuators, the funding—all of that is figured out. The lab space is all figured out. So, that's what we're building now. We've had something like 80 consecutive AI deals now, and the returns are incredible. It's like nothing I've ever seen. But the window of opportunity for pure software AI is probably another couple of years.

And then the self-improvement loop is just going to take over. I think robotics is a good 10-year theme, just like biotech is a good 10-year theme. We've got our first robotics deals already done. We're doing a lot of robot operating system deals, where they're reusable across many different devices.

Also, you notice on college campuses that a lot of computer science majors are now shifting back to materials science, mechanical engineering, and some of the hard sciences, in anticipation of this wave. If you're in college right now, you're going to miss the foundation-model wave, but you'll be perfectly timed for the robotics wave. Also, everything we've ever done with robotics needs to be rethought for space, zero-G, and radiation. So, you've got to do it all over again for manufacturing out in space, which is another great theme that we want to get ahead of.

Yeah, let me hit on a couple of robot stories, and then one in particular for Salim. This is first: Hyundai's robotics company is starting to learn to play soccer.

In preparation for the World Cup, here we see Atlas kicking a soccer ball. It's going to be interesting. Let me show the next 2 videos, and we'll talk about the robot companies. We'll talk a little bit about Figure AI setting a new record for continuous package sorting.

All right. Out of China, we've got the Dollar Haircut. You heard of the Dollar Shave Club; now the Dollar Haircut. Take a look.

Finally.

Yeah. Here we see a robot. I do think this is kind of inevitable, but are you going to trust a robot with a sharp blade very close to your neck? And, Salim, this one's for you: a humanoid robot with 6 arms. You've been asking for this for a while. What are your thoughts, Salim?

Oh, I just love it. Why limit to 2 arms? For God's sake, I always use the example of when you're trying to open a garbage bag: you need a third arm to hold it open, for God's sake.

I just want to thank all the viewers and listeners. People have been tweeting things: “Here's your robot. Here's a 6-wheel thing. Here's a 4-arm thing.” It's been absolute fun to do it. I think it's a great way of taking it away from just the humanoid figure. I understand that humanoid robots are used to moving around in human spaces, but an extra arm can't hurt.

And we saw Figure Robotics. How long did it do continuous package sorting? Is it still going, or did they—

For longer than a week. I think Brett ended it after 8 days or so.

We're going to see a lot in the robot space coming out shortly. Of course, Optimus—the next iteration of Optimus, Optimus 3, is expected. When we were there, what was it? 10 million square feet of Optimus production capacity, Dave? Something like that.

Dave

Yeah, something like that. That's huge.

One sad story here is Blue Origin's New Glenn exploded on the launchpad. Take a quick look at this video. Here we see it in Cape Canaveral. This was during a ground test—a cataclysmic deconstruction.

It's never easy, but hardware is hard. Blue Origin had a rough week. New Glenn, Jeff Bezos's heavy-lift launch vehicle, exploded during a static-fire test at the Cape. This is the rocket-development game. You're going to have setbacks along the way. You can't test it piecemeal. SpaceX has blown up plenty of rockets. This time is painful.

SpaceX just launched their Starship V3 last week very successfully, and Blue Origin is going to be hit by this. This is the vehicle that's also planning to launch Amazon's Project Kuiper. They're going after the NASA lunar contracts. Alex, any thoughts?

Alex

Yeah. First and maybe most importantly, no one was hurt, so that's great.

This was an unmanned vehicle. This is a satellite launch capability right now.

Alex

Second-order impact: Artemis 3. This was the main vehicle for Blue Origin participating in Artemis 3. So, playing Kremlinologist here on the contracting supply chain, it looks to me like SpaceX is very likely to end up being the preferred end-to-end vehicle for getting humans to the Moon over the next few years.

I think the general consensus in the space community now is this could set back Blue Origin's Artemis lunar-colony plans or participation by up to a year. Superficially, a good opportunity for SpaceX to shine in getting Americans back to the Moon over the next couple of missions, but hopefully Jeff Bezos and Blue Origin are able to rapidly rebuild and reconquer LEO and then cislunar space.

And I think the implications for Kuiper are important, too. All right, let's wrap up with some quick AMA with the mates. We only have 5 minutes left here. Salim, do you want to pick the first one?

So, given China's AI adoption, robotic scale, and population, how likely is it that they hit abundance before anybody else? This is from @EdPlano.

This is very possible, but I would separate material abundance from human abundance. China has a huge shot at reaching material abundance with solar, batteries, EVs, robotics, and logistics. They have the scale and supply chains and so on.

But abundance is not just cheap goods. It includes agency. It includes freedom to experiment. It includes human flourishing. You may get the cheap physical production, but the West may still have an edge in entrepreneurial recombination, open innovation, and meaning-making. We just have to make sure we don't lose our freedoms along the way, which is my big concern at the government level.

Alex, what's your question?

Alex

I'll pick number 1. When do agents start running political campaigns? This is from Mad Prophet of Wiki.

Well, Mad Prophet of Wiki, I think they started running political campaigns maybe 10 years ago. We've had AI agents deeply involved in social networks, involved in both local and federal US election campaigns, involved thoroughly end to end, with AI involved in that entire stack, even prior to the LLM revolution.

We've had agents—an agent being just an AI that incurs multiple sequential interactions with an environment—and advertising and online retargeting. That's a fundamentally agentic process in nature. So, I'd argue we've had this for at least 10 years, probably materially longer.

Dave.

Dave

I'll take bullet 2. Of all the recent layoffs, how many people realistically have actually turned into entrepreneurs? And that's from AI Business in a Box.

I did a spot survey of Microsoft—Microsoft and Amazon in Seattle had about 20,000 or 30,000 layoffs—and then Meta has another 10,000 in San Francisco and Silicon Valley. I just poked at some LinkedIn profiles. Short answer is, in that sample, almost everybody has either joined a startup or joined another company that was recently a startup.

So, it's the best time I've ever seen, by far, for entrepreneurs and startups. By far. Now, that sample is just covering software engineers from Microsoft, Amazon, and Meta, so that's a very different sample from if we start seeing layoffs in robotic areas like garbage collection. I'm sure you're not going to see anywhere near that number of people joining startups.

I haven't sampled that yet. But at least within the tech community, it's a very rosy picture, again driven by the fact that the amount of abundance is just massive in scale.

So it can absorb a lot of humanity. So, not bad so far.

All right. Number 4, from Malcolm Machinan 6849: If white-collar jobs vanish, tax rates spike to fund welfare, does an unemployed doctor get paid the same as an unemployed addict? Fascinating way to put the question.

This assumes that jobs vanish rather than get transformed, which isn't what the data show so far, right? The Dallas Fed says it's a hiring freeze, not mass layoffs. But I think your deeper question is around differential welfare, and that's fascinating. I think we're heading toward something like a universal basic capital rather than a flat UBI. Instead of paying everyone the same, you give people an ownership stake in AI-generated wealth proportional to their contributions or their retraining effort. I ultimately think we're going to see some base level of UBI, but then people are going to be able to use UBI or UBC to build on top of that.

Do you guys have a second for another question round?

Alex Wissner-Gross

Yeah, absolutely. Okay, great. Let's do this. All right, back to you, Salim.

Let me take question 5, the first one. Isn't the real privacy question not what AI systems know about us, but the legal protections on what they're allowed to do with that knowledge? And that's from @poetrytosong, which is a great handle.

That's exactly the question, right? Privacy used to mean, what data do you have? The deeper question is, what can an intelligent system do with that data? Can it change my insurance costs? Can it change my vote? Can it convince me about stuff? Can it deny me credit? The new frame is not just privacy rights. It's agency rights.

We need to have protections around having access to inference, prediction, manipulation, and so on. It's not about what AI knows about me—that I like hiking or red wine. It's whether knowledge is used invisibly to constrain my choices. In the 20th century, privacy was the big issue. The 21st-century issue is agency. What can an agent—can AI improve my agency and augment it, or reduce it?

Okay, Dave.

Dave

I love number 6. Why is taxing tokens a perverse incentive? AI is going to drive the future economy. Taxing tokens to fund UBI makes sense.

What we don't want to do is repeat the error from Obamacare. If you remember Obamacare, everyone was going to have a health care card. We were going to have universal health care. The first thing we'll do is pass a new tax. We'll call it the Net Investment Income Tax. It'll be a 3.5% tax on top of all other taxes. Here we are, 15 or 20 years later. We still have the tax, and we don't have Obamacare. So everything got thrown out by the next administration except the tax is still there.

Solving UBI, the easiest part by far is taxing. In fact, we don't need any new tax. The government has plenty of tax mechanisms already. The corporate income tax should go through the roof with the abundance coming online, so you're going to have plenty of tax collection. That's the last thing you need to worry about.

What you need to do is actually design a UBI that makes sense. If you tax tokens, yes, that's perverse because now people use fewer tokens when they should be using more tokens. They should be trying to build things with AI. So get rid of that. Don't worry about that. [laughter] Worry about the UBI and how we're going to design it. We have plenty of ways to collect money. That's not the issue.

Alex, I'll pick number 8.

Alex Wissner-Gross

Will AI eventually solve data centers and chips in a way that we don't need them at all? This is asked by Dave Lane. I love this question. I think the answer is probably yes.

I think there are so many ways to compute that are allowed by the known laws of physics that I do think it is very likely that we will ultimately transcend semiconductors and CMOS. By ultimately, I mean on the time scale of a decade, not centuries, and I think AI will enable us to do that.

I've spoken on the podcast in the past about black-hole supercomputers on your desktop and about plasma computers. Seth Lloyd wrote about these 25 years ago in “Ultimate Physical Limits to Computation.” I think it is very likely that with advances in physics, AI, and AI for physics, we'll discover breakthrough substrates that go well beyond the obvious next steps, like photonics, to perhaps, maximally ambitiously, computing directly using gravity and/or quantum gravity.

There's a body of evidence in the physics literature that suggests that, in the strong gravitational-field regime, gravity becomes turbulent, in some generalized sense of turbulence. There have been a few papers in the literature on this, and turbulence—a separate body of literature—is, in some sense, Turing-complete.

One can imagine a story sometime in the future—not sure when, maybe 10 to 20 years out—where we're literally replacing data centers and chips with pure energy and the stress-energy tensor, and computing with gravity. I think that's one possible endeavor.

I'm glad you answered that one.

Computing with gravity. Wow.

Alex has a paper on that. If you look on his site, you can read it. It's pretty—

Alex has a paper on everything. Number 7. Number 7.

That's why you can solve everything. [laughter]

Okay. For number 7: With agents costing real money, how can we believe in 8 billion—I assume humans, or 8 billion agents—when we can't get 8 billion humans a free education? And that's from @kiwonggpk91.

I think the assumption here is at the current cost, but the price curves are collapsing, right? Token prices have dropped, arguably, 75% to 90% in the last 18 months. We talked about that last week—the Jevons paradox data. Gartner has predicted, I think, something like 90% cheaper tokens by 2030.

A useful AI agent today—you can buy it at $20 a month—will end up costing you $2 a month in 2028 and $0.20 in 2030. Then I think there are going to be a lot of free services offered by these frontier labs, by the hyperscalers.

Meanwhile, today, the cost of an education is arguably, in the U.S. at least, like $10,000 a year. An AI tutor running 24/7 can cost a fraction of that.

We can also personalize to you, yes.

Yes, we're going to have AI agents that you'll pay for if you want the top-tier service, and those will be getting cheaper and cheaper. I think there's going to be a free variation on these agents for everyone who can't afford it—sort of universal basic compute. We've talked about that before.

Peter, thank you for the interview you and I did. Our world is exploding. We have an abundance problem. We've got hundreds of applicants wanting to get into the pilot, so we're trying to figure out how to down-select. I really appreciated that interview we did.

Yeah, I know. It was fantastic. If you haven't watched Organizational Singularity, which is a one-on-one episode that Salim and I did, please do. It's an hour long. I hope you guys have enjoyed these shorter episodes. Give us your feedback. We're trying to keep these under 90 minutes, and as a result we're doing them more frequently. If you haven't subscribed and turned on notifications, please do. We're dropping these episodes three times a week.

Should we ask the audience whether they'd prefer daily Moonshots?

Oh my God, yeah. Tell us what you want. I keep saying we're going to move into an Airbnb together. This outro music is from Jazz Cathedral Builders. Again, if you're a creative, please send us your outro, or if you want it to be an intro song, let us know. You can send those to media@diamandis.com. We love our community, and thank you for putting the time into these beautiful songs.

Before a breakthrough, it's just a crazy idea. And then humans living harmoniously with nature like never before. From scarcity to abundance. We need not be at the whims of today. We are the architects of tomorrow.

I love the visuals. The timeline, I think, is way too conservative. Did you see the baboon on the sky bridge in the future of the architects of tomorrow? The best way to predict the future is to create it yourself. This is the most extraordinary time ever to be alive. I love you guys. I love our sessions. I got up at 4:30 this morning, recording this on the West Coast at 6:30 in the morning. So worth it.

Alex

Can't sleep through the singularity.

All right, guys. See you very soon.

Have a great weekend. Take care, everybody. If you made it to the end of this episode, which you obviously did, I consider you a moonshot mate. Every week, my moonshot mates and I spend a lot of energy and time to deliver you the news that matters. If you're a subscriber, thank you. If you're not a subscriber yet, please consider subscribing so you get the news as it comes out. I also want to invite you to join me on my weekly newsletter called Metatrends. I have a research team looking at the meta trends impacting your family, your company, your industry, and your nation, and I put this into a two-minute read every week. If you'd like access to the MetaTrends newsletter, go to diamandis.com/metatrends. Thank you again for joining us today.

Yeah, I know. It was fantastic.