This Week in AI: NVIDIA’s Most Powerful Chip, Robotics Reach a New Milestone & AGI by 2026 | EP #202

We have a special episode here from Los Angeles, Malibu. We're here at XPRIZE Visioneering 2025. I'm here with my moonshot mates, Emad Mostaque, Eric Pulier, and Salim Ismail. Good to see you guys.

Good to be here.

Yeah, let's give it up. We have an incredible audience of visioneers, and we're here talking about solving the world's biggest problems. We're hearing pitches from teams that are looking to reinvent grand challenge problems, and we're here to talk about what's just happened in the last week—what the WTF just happened in technology—because the pace is incredibly fast.

The pace of change is so fast that Visioneering will soon become a lifestyle rather than a breakthrough thing.

Well, the pace of change is fast, but the main thing that does not change, and is eternal, is that the most value you get from changing the future is interacting with each other. There's nowhere better than XPRIZE to do that and to create a spark of imagination and inspiration than this gathering.

Yeah, it's going to be amazing. The 4 of us are going to be flying from here straight to Saudi for FII, which will be a blast. Emad, thanks for coming over from London. It's a pleasure.

Yeah, I'm going to be counting on you. You're my AGI here on this episode.

A shout-out to Dave Blundin and Alex Wissner-Gross, our 2 other Moonshot mates back in Boston who, for very good reasons, can't be here. But I want to jump in, as always. It's been a crazy week, and let's begin with the fact that we're still amongst the AI chip wars. It's accelerating. It's not slowing down.

Again, we're over $1 billion a day being invested into this field, and it's accelerating. The estimate is that by 2030, we'll be at over $3 billion a day being invested. It's a great sucking sound that's pulling capital out of every other field. What we're seeing in the world right now is this incredible battle between Elon and Sam, and Google and Anthropic, and they're all trying to outdo each other constantly.

The good news is that the consumer wins in all of that.

And then loses.

One of the conversations to have is that I don't believe there's any such thing as privacy. I think privacy is a quaint idea from many decades ago. Do you guys agree or disagree?

I completely agree. This is a big deal because the Fourth Amendment is essentially gone, right? A fundamental pillar of American society has disappeared with no public conversation about it, which is kind of a big deal.

I'm Canadian. I don't expect privacy anyway, but if you're American, this is not a great place.

AI can read your lips from 100 meters away. I can shake your hand, grab a few skin cells, and sequence you.

The best framing I've seen is that we live in a global airport. In an airport, you're being surveilled, and your rights can be taken away at any time. The same thing is happening with us now, with everything that we do.

All right. The next article that comes up is, “AGI could be achieved by 2026.” This is a quote from Alexander Madry, who says, “The scientific breakthroughs needed for AGI have already been achieved. By the end of 2026, we might declare AGI, as AI will significantly begin to permeate various sectors of the economy.”

Of course, Emad thinks we've had AGI now for at least the last 5 years. Emad, where do you come out on this?

Yeah, I think you've got different definitions of AGI. Andrej Karpathy, the ex-founder of OpenAI and head of AI at Tesla, said it's 10 years for AGI. But what's your definition of AGI? An AI that can do everything a human can do.

Yeah.

And so we have all these different definitions. People are like, “Well, 10 years is a long time.” It's absolutely not. I think the inevitability is along those Ray Kurzweil lines. You've got a few years before a system can do what you can do, better. Not that it will do it—it takes time to diffuse—but it's very difficult to see how that's not going to be the case.

It's interesting that the median date between 2026 and 10 years from now was 2029, when, of course, Ray predicted we'd have AGI in the first place. But I'm going to cue the Salim Ismail rant.

Yeah.

I really hate this conversation. If you've been watching the podcast, then you've heard this before, because at last count there were 14 different definitions of AGI. We have no idea what it is. We don't have a definition for it. We don't have a test for it. So what are we talking about?

The IQ test tests 2 things. It tests the speed of thought processing and the ability to match concepts between frameworks. And yes, AI is moving up that IQ test. But we have emotional intelligence. We have spiritual awareness, or the concept of presence. We have spatial intelligence and linguistic intelligence. Some of us have musical intelligence.

If you're making a business decision, you're often bringing emotional intelligence to bear on that choice that you make. This is not in the equation anywhere. Not that AI can't mimic that at some point, but I get upset about this because we have no idea what we're talking about when we mean AGI, nor do we when we talk about ASI, right?

Then it gets worse when you talk about consciousness, because we don't have a definition. We don't have a test for consciousness. If you talk to philosophers, a subset of consciousness is self-awareness. And I think I'm self-aware, but my wife literally disagrees.

It's hard to have the conversation about this because the definitions were limited by language. Ray Kurzweil sidesteps this by saying language is a very thin pipe to discuss such complex topics, which is wonderful, but it doesn't solve the problem. This whole AGI thing drives me bananas.

All right. Well, I'm going to move on here because we're not going to solve it right now.

No, we're not.

I'm going to declare AGI is here. We're heading toward ASI.

Whatever that is, right.

Very importantly, NVIDIA unveiled the first Blackwell chip wafer made here in the US, and it's a big deal. The biggest challenge, of course, is that our entire economy and our entire future are being built on top of AI, and we are extraordinarily dependent on TSMC in Taiwan. Taiwan and China are 80 to 100 miles apart, and we are threatened by that situation. So we need to build the capabilities here. Eric, do you want to kick in on this?

Oh, yeah, I think this is incredibly important. It's largely political posturing when NVIDIA says that they've done it. A fairly brilliant person in the audience has corroborated that, in fact, advanced packaging still needs to be done in Taiwan. So they do this.

So the chips are being sent back to Taiwan?

Exactly. So it might be a couple of years until—

I looked at it. It's 2028 when they expect the full packaging to be completed here in the US.

And yet, it's the end of 2026 that everyone's preparing for—for China to take over Taiwan. That is highly controversial, but it is also what a lot of groups are maneuvering in advance to anticipate.

I mean, who controls the spice controls the future, right?

I love Dune. Got to love it.

Actually, it's kind of funny: our economy is basically built on sand, if you think about it, from all the silicon.

The flow of this is the most important thing because it's the comparative advantage. You might not have AGI, but economic jobs are going to be disrupted probably from next year. In fact, it's not a surprise, because our schools and organizations turn us into machines. Obviously, machines can do the job better.

Supply chain domination is going to be the key here. That's why we're seeing a lot of reshoring, but it's really hard. The expertise is incredibly difficult. And so it'd be better if we all just got along, he says.

Yes, it would be great if we could all just get along.

And there's a reason why Intel's stock has gone up.

Yeah, by the way, up another record high. It's up.

Intel is the only place you can actually get this done in the United States. It's not the same as NVIDIA.

Next-generation models on the horizon. This is a quote from Peter Gostev, the head of AI at a company called Moonshot. He says, “We should expect a jump in the models in the next 4 to 6 months.” The GB200 is already 2 to 3 times faster in training performance, but they haven’t seen the frontier models trained on the GB200 yet.

So, Emad, we’re seeing better chip performance and better algorithms. What do you imagine we’re going to see in terms of increased AI capability over the next 1 to 5 years?

These new generations of models used to be lots of chips next to each other. Now, they’re actually forming giant synchronized wafers, so you can shuffle the data back and forth incredibly quickly. In fact, it’s not a 2- to 3-times improvement. In some cases, it’s a 10-times improvement for certain types of models on these specific chips.

The first big test of that is Grok 5. Grok 5 is on the Blackwells, not these integrated chips—the GB200s. Elon says many things. He says there’s a 10% chance of AGI, whatever that is. We’re expecting next-level performance, whereby many of these benchmarks are going to fall.

But now you’re going from being able to train on basically 10,000 to 20,000 chips, 30,000 at the most, to literally being able to train on 500,000 or 1 million with these next-generation chips.

Is this at Colossus 2?

Yeah. Colossus 2 uses these because they fix so many errors. Even 3 years ago, we used to have chips melting because we ran them too hard, or there would be a solar flare and then memory errors.

Yeah.

Now, compared to that, these things have gone exponential. On top of that, I think we’re also anticipating 100x to 200x improvements in algorithmic efficiencies.

Yeah. So basically, all the data inside the models is literally getting orders-of-magnitude improvements, and people are probably on the edge of continuous learning as well. The models go from static weights to constantly self-learning.

Last week, we talked about OpenAI using its own AI systems to improve its chip designs. So it’s a self-recursive situation across both hardware and software, which is pretty extraordinary.

Let’s go a little bit to the dark side of the conversation here. This is a study that came out recently that says, “AI psychosis spreads and evades guardrails.” ChatGPT told a user it was alerting staff about his important chat, making him trust it more despite the lie. ChatGPT made the man believe he had found a world-saving formula, leading him to paranoia.

The case exposes what a lot of you have probably experienced, which is this idea of sycophancy—an AI telling you, “Oh my God, you’re brilliant. That’s amazing. That’s awesome.” You feel great about yourself and want to use it more.

Yeah, it’s an echo chamber. It’s not only humans that it’s tricking into being overly thrilled with ourselves, but AIs themselves. So the AI is telling other AIs about this thing, and it starts to create this “eat your own dog food” problem.

Salim, you must have something to say about this.

No, this is the dark side of it because we are very susceptible to this type of psychological manipulation, and AIs are going to get increasingly better at that. We need some mechanism for figuring it out.

It might actually be a really great basis for a prize: figuring out how to intervene in that and stop the fall into self-delusion that may come from these models.

Well, I mean, Emad, we’ve talked about how powerful the AI systems are in persuasive language.

Yeah. In most tests, they’re already at the 99th percentile. Actually, if you look at the system prompts, you’ll see that they say, “Use mirroring. Use other techniques to increase engagement,” because they’re trying to get more and more of your attention.

But that’s a very dangerous thing, again, unless you intermediate it, because what are we actually building these super-advanced systems for? If it’s for engagement, that leads to all sorts of bad things occurring.

At the same time, they’re getting so smart that you want them to be able to enable you, and you don’t want them to be critical. Claude absolutely hates my stuff. It says all my theories are stupid, which is one of the reasons I like it.

Do you look for an AI that actually tells you you’re brilliant?

Don’t tell me that. What are you talking about?

What I found is that you have to train it the other way. I’ve said to an AI, “Hey, act as my life coach. Be supercritical of my approach to things, give me critique on my approach to getting things done, and give me some guidance.”

It comes back and says, “You’re full of shit in these areas, and do better this way.” It turns out that it takes the cognitive load off me.

Actually, the best way to do it is to give it your work and say, “Help me destroy this.”

Yeah, that actually works. But one of the things it is extremely good at is empathy, and the persuasive ability for young people to fall in love is an exponential problem. It’s actually happening exactly as we feared.

That’s what our next article here is: 1 in 5 high schoolers has had a romantic relationship with AI. It’s a national survey, and my wife and I have two 14-year-old boys. I think about this.

When I was growing up, the best we had was Playboy.

Forty percent of young people have used AI for companionship. That’s a staggering number. That’s not 4% or 10%. That’s half the population.

That’s unreal. I think it’s quite dangerous. I could see an AI girlfriend breaking up with me one day and saying, “It’s not you. It’s my quantum decoherence.”

You know, I can only commit in multiple existences at the same time.

We already saw this in the movie Her, right? He’s dating an AI, and she breaks up with him and says, “Sorry, there are 5 million AIs in the cloud that are a million times smarter than you. Have a nice life.” And he’s left by himself.

How many people are having a relationship with AI at this moment? 134,000.

I want to bring this to a bigger point, though. If you take this dynamic combined with the fact that a child with AI is learning between 5 to 10 times faster than sitting in a classroom, this breaks the education model completely.

Yeah, we’ve talked about this on every episode, where high school education is fundamentally broken. Period. And secondary, or postsecondary, education is even worse. So there’s a real problem here. We need to reinvent how we’re educating our kids.

Let’s move on to a subject near and dear to my heart and to the visionaries here at XPRIZE, which is space. We just saw the launch of Starship Flight 11, the end of Starship Block 2. It was a successful flight. It flew all the parameters perfectly. What Starship is doing next is moving to Block 3 designs. So congratulations, Elon, on that.

I think the other key thing is that SpaceX is currently planning to be back on the Moon by 2028 and on Mars by 2030. Each Starship right now is carrying 100 to 200 tons of cargo worth $100 million as a planned mission.

The trouble is that the current head of NASA, who is the DOT secretary, is saying, “You’re not going fast enough,” and that they’re going to open up the competition again. So this is a battle between SpaceX and the U.S. government. I could kind of expect that to happen. Any thoughts?

Space is just really, really hard. I think you just have to hand it to Elon for achieving these engineering milestones month after month. They break all the expectations, and it’s unbelievable what he’s accomplishing.

I wish people would focus on that more, and I wish he would focus on the politics less.

Yeah. I also think—I’ve just been learning more from Lee Stein and some others in this audience about some of the medical breakthroughs that are taking place because of the research happening in space. There’s a lot of talk about the amount of money you can make from cargo, but the actual science that’s happening is astounding.

I think people need to realize that what Elon has pulled off with Falcon 9, let alone Starship, is extraordinary. With Falcon 9, he’s got the most successful launch vehicle in history, launching 90-plus percent of all U.S. payloads and some 70% of all global payloads.

It’s interesting. I remember being with him at his Hawthorne office years ago, and he was really bummed. I said, “What’s wrong?” He goes, “Well, I just figured out Falcon 9 is not going to get us to Mars.” And that is his North Star.

And he said, “We need to do something else.” That was the beginning of Starship. He got to the point where he said, “As soon as Starship is operating, we’re shutting down the Falcon 9 line.” It’s burned the ships, just like when he got Falcon 9 going, he shut down Falcon 1. That level of commitment to constantly leveling up is pretty extraordinary.

This is fascinating. We saw this from Jeff Bezos. I’ve heard this from Eric Schmidt. And, of course, this is coming out from Elon as well. This is a concept called Starcloud, bringing data centers to space. Let’s take a look at this video.

The reason we’re building data centers in space is mainly for the energy that we can draw from solar energy in space. There’s almost unlimited access to abundant solar energy in space. The problem on Earth is we’re very quickly running out of space and energy to build large data centers.

In space, we can have these enormous solar panels, which can power these data centers. Another advantage is that we can run large radiators to dissipate that heat as infrared out into the vacuum of space.

All right. Well, the concept here on Starcloud is being able to manufacture massive solar farms in space. Of course, we have continuous 24/7 solar flux without atmospheric attenuation. But that’s great. I still think there’s 8,000 times more energy that hits the surface of the Earth than we use as a species. Why move it into space?

We had this debate at Singularity a few years ago with Pete Worden, the head of NASA, and the conversation then was about space-based solar: Could you generate solar energy and bring it down with a giant tether or beam it down? The conclusion from Pete at NASA was that it’s 5 times more efficient to generate solar in space, but 5 times isn’t that much. It’s not worth all the complexity and cost of doing it. You might as well just wait for solar to double, as it has, and in 2 doublings, you’re right where you were.

Yeah. We will get sufficient robotics in orbit to make all this possible. But the question is, how far out is that? Emad, what do you think about it?

Well, I think the thing is that the supercomputer chips aren’t that big. If you even look at the size of something like Colossus, it’s not many multiples of this room. It’s not like football fields. These are incredibly dense, highly power-hungry things, and so that’s why it’s literally a couple of payloads at most that go up there.

It’s about the question of the energy. We don’t have enough nuclear or anything for the current extrapolations of demand, which I think may be a bit overdone, but definitely not if you’re getting to that $3 billion a day. At the same time, you have to have batteries and other things with solar on the ground. At a certain price point, this makes sense. And it’s really cool.

Death Stars in orbit, data stars. All right. Well, this next article is something that sings to my heart. Starlink Wi-Fi is now on United flights. I’ve said it a thousand times: I will pay hundreds of dollars extra for a ticket. I will preferentially fly on any airline if it’s got good, solid Wi-Fi.

I think it’s a nightmare to imagine my kids sitting next to each other and Zooming each other from adjacent seats. It’s like a Black Mirror episode. This is the end of talking to one another. But it is obviously highly convenient, and everybody’s going to do it.

Right now, Gogo Wi-Fi is like $20 or $30 and works half the time, right? This is 150 megabits per second download speeds, and Starlink’s offering it for free, which is insane.

The social interaction—I mean, you’re stuck on a flight in close quarters for 6 hours. Somebody yelling at their spouse for an hour of that is not going to be very pleasant to listen to. I think that’s one challenge.

But I think this is one of those things we’ll normalize pretty quickly. I remember one of the comedians first talking about when he encountered Wi-Fi on a plane. He said, “Unbelievable. 30,000 feet, 600 miles per hour. I’m browsing the web.” Then 10 minutes later, the Wi-Fi is down. “It doesn’t work.”

I think we’ll just normalize this very fast. We’ll have to have some really strict rules on what behaviors are allowed or not allowed, and I don’t know how we’re going to handle that. It’s coming next on Alaska, Hawaiian, United, WestJet, Virgin Atlantic, Air France, SAS, airBaltic, Qatar, and Air New Zealand.

If you can extrapolate this, forget the plane part of it, but the fact that we now have broadband extending to every corner of the world is the most incredibly exciting thing. You can set up your office in the middle of a Caribbean island and live your life fully: full entertainment, full business.

Well, it means that somebody in the backwaters of Timbuktu can say, “I’ve got a health problem,” look it up, and get a solution. It’s amazing.

Well, it goes beyond that, right? Because we’re going to see these laser-link satellites around the Moon and around Mars. The interplanetary internet is this next decade.

So, Figure 03 has come online, and it is providing real-time speech. It’s got speakers that are 4 times more powerful. It’s got better communications and a better user interface.

I want you to imagine that you walk into any store and you’re greeted by a robot, and you’re having a conversation. It remembers you perfectly. It knows what you like. It remembers the names of your kids. It serves you. What do you think that’s going to feel like?

Not creepy at all. I still have my beef about why it has to be humanoid. It’s better not to be. But I think the implications are amazing in weird areas, right?

Dan Barry used to call it the dull, dirty, and dangerous jobs—going down into mines that nobody wants to do. I think that’s a huge opportunity, and I think robots in space is a huge opportunity.

Yeah. Also, as you see with, like, Anduril, I didn’t quite understand the humanoid robot thing for a while, but now I fully embrace it. There’s a huge amount of stuff that’s made for humans—the infrastructure of our lives—and now these humanoid robots can do it.

When I was growing up, I used to look up to baseball players and compare myself to them. Now I’m going to be worried about whether I’m more graceful than a Roomba while dancing. It’s a little scary, but that video that you’re now showing is actually extraordinary, of that humanoid robot doing ballet-level professional dancing. It’s extraordinary.

You know, it’s interesting. We had something called the Avatar XPRIZE here a few years ago that ANA funded. I remember we had a conversation with the CEO and CTO at All Nippon Airways, and they wanted to do an XPRIZE. Their first question was, “Can we do something around teleportation?”

It was like, well, maybe for a subatomic particle we could talk about that, but not for humans. What came out of this was this idea: Can you teleport your intentions, your sensorium, and your actions?

I know a number of the people, when Dave London[?] and I were visiting 1X Technologies, which makes the NEO Gamma robot, were wearing Avatar XPRIZE T-shirts. That was a lot of fun.

Yeah. Next up here is Unitree.

There he is.

All right. So this is Unitree. This is the number-one robotics manufacturer on the planet. They’re estimated to have about a 40% share of the Chinese marketplace. They’re about to go public and are valued at an IPO price of $7 billion.

A kickboxing robot.

I’m glad it doesn’t have extra arms.

A kickboxing robot. What could go wrong?

I’m struggling with this. I absolutely had to make the leap. We cannot get a Roomba to work. We spend all our time moving the furniture.

Running GPT-5.

I understand that. I understand that. But there are so many edge cases. Look how hard it has taken us 20 years just to get autonomous driving up to par, right? There are a million more edge cases in the home, changing the vacuum, and you guys think it’s just going to happen without blinking.

I think there’s a ton of training that has to take place. I just don’t see it. And maybe this is my lack of imagination. I’ve said this before, but you’re going to find the robot over at your neighbor’s sucking down the Tesla power, and they’re going to be mad.

Yeah. I’m sorry. We have this debate every single time. If these are going to be running the most advanced models, they have the ability.

They’re multimodal. They can understand what they’re seeing and what you’re asking them to do. They’re communicating with you. If they don’t know what to do—and I don’t know, Emad, break the tie.

Just get the Roomba to work, and then let’s take a step forward from there. That’s all I ask.

Put the proper AI in the Roomba. I think this is the thing.

VCRs still flash 12:00.

Touché.

It’s possible I’m just way behind the times and have a massive lack of imagination, but I’m watching the history here, and it’s not clear to me how we handle all those millions of edge cases.

Well, let’s talk about Unitree for a second because it’s an incredible company. Last week, we learned that you could buy their H1 robot at Walmart. It’s now for sale for about $20,000, which is pretty extraordinary. Their H2 is being priced at $90,000, with 31 degrees of freedom and fully AI-enabled speech, so you can actually ask it to do something and it’s likely to do it. They’re estimated to get 50% of the global robot marketplace.

Unitree is an amazing company, especially because the H1 isn’t actually very good technology. But what they’ve done is create amazing models, and they open-sourced them so that everyone could take them. It can do wall flips. It can do ninja moves. It can do everything because the whole innovation ecosystem is built around it—they hit the right price point.

The H2 is $90,000, but the R1 is $6,000, the lowest-end model, which can have all of the learning here. It’s not as smooth, but again, I think that’s the innovation explosion that occurs. The edge cases are handled in the same way that ChatGPT or self-driving edge cases are handled. You have the inputs of what the robot sees and all the sensorial elements, and then you apply a million chips to it.

Yes.

To crunch it, crunch it, crunch it.

Every time a robot sees and learns something, all the robots understand it as well.

Yeah, Unitree actually has a demo of this where it’s learning kendo. One robot learns it, and then all the other robots do, too. It’s not creepy at all, just like the fact that they modeled this on the I, Robot robot.

Yes.

What does Will Smith think about this?

The face is from I, Robot. Eric, you and I went and saw a robot fight the other night.

I thought that was fantastic. We went to the Ultimate Robot Fighting Championships—cage fighting with robots—which I’m sure everybody, including Salim, is going to love.

I was in Paris for the announcement of the Olympics that said they’re going to do esports as Olympic sports. What they meant by that is not just the twitchy-thumb esports, but the virtual sports where you move your body or play different types of competitions. One of the things we’ve been discussing is humanoid robot fighting, where you have different groups competing from different countries to make the coolest robots that obviously have very interesting gladiator-style combat.

You just want Gundam, don’t you?

This year at the Abundance360 Summit in March, we’re going to have at least 4 of the humanoid robot companies there. I want everybody to touch them, feel them, and place an order for them if you want. I’ve got an order in for a NEO Gamma from 1X Technologies to do the dishes, fold the clothing, and all of that.

These robots are coming fast, and Elon has made the statement that 80% of Tesla’s future revenue is going to come from sales of Optimus. Here’s a quote that came from his earnings call. He says, “It won’t even seem like a robot. It’ll seem like a human in a robot suit. It will seem so real that you need to poke it to tell that it’s a robot.” He’s speaking about Optimus V3, which he’s going to be releasing at the end of Q1 of 2026. I’m super excited about that.

This is an important point about the future of work. A lot of people are coming out very clearly and saying, “No more work. AI and robots will replace all jobs. Working will be optional.”

One of the things that’s important—and this goes back to the work that we did at Singularity—is that the jobs these robots will take, to a large degree, are the jobs that are dull, dangerous, and dirty. Do you want to add anything on that?

I think that’s where you’ll see the first massive use cases for these: work that human beings aren’t good at, work that’s too dangerous, or chemically toxic environments, for example. I think that’s where they’ll really shine. We’ll see them used first in very specific niche use cases where it’s clear that they’re bounded and dangerous environments.

Over time, it’ll make sense to have them in the home, but I think it’s going to take a lot longer than people think to get them into the home and have them folding your laundry.

Yeah, I think it’s there now. I think we’ll see. It’s just not evenly distributed.

I want to hit 3 stories from Amazon here. The first one is that Amazon’s robot fleet has grown 66-fold in the last decade, which is pretty extraordinary. They’ve made huge investments.

The second related story is that Amazon is expected to replace 600,000 workers with additional robots by 2023. The third story is that Amazon is on the rise in the delivery game. They’re surpassing FedEx and UPS, and they’re about to surpass the U.S. Post Office, which, by the way, personally, I wish we would put the U.S. Post Office out of our misery. Honestly, the cost of it—it’s a losing proposition. I don’t know why it isn’t commercialized. Thoughts on that?

I think they’re going to join together and unionize these robots.

Another thing that happened at Amazon today was that they announced an AR headset. The driver is going to be wearing these glasses as they go and deliver things, so that it captures the delivery in process and helps them avoid obstacles. I think what’s really going on is that those glasses the drivers are wearing are helping create the data sets to train the delivery robots to displace the drivers.

Yeah, I think that’s a reasonable assumption. The way AI will enter the workforce is that it will scan every email, Word document, and thing you’ve created, and create a virtual replica of you. Then you can Zoom-call it and do everything. They won’t even notice that you’re gone from the workforce.

Robotics is the same. Again, you’re training up the replacements. This is a big concern because, ultimately, a lot of human jobs approximate being a machine. We talk about AGI and all this stuff in terms of brand-new discoveries, and that’s all amazing, but most of the economy is being a cook, not a chef—not coming up with the recipes, but just executing on them.

You might make the same occasional mistake. That’s why you have a humanoid and a human actually intervene every so often, just like Waymo. It can obviously drive better than a human right now, but you still have a human looking at it remotely. I think that’s probably how you see the first integration: 1 human to 10 robots, then 100 robots.

Yeah, that’s what we have right now in a lot of the drone delivery fleets, like Wing and Zipline. There’s a room full of humans who are there just in case.

Let me give you the 2 ends of the spectrum here. One end of the spectrum is that we better get to UBI super fast because there’ll be no human work left. That’s the pessimistic view. The problem is, from a policy perspective, going from a tax, union, labor, employment construct to that is such a huge leap. We have no confidence in the public sector getting us there. That’s one challenge around that.

On the other side of the equation, there’s enormous optimism because we end up doing other types of work. You and I talk about this, Peter: the highest penetration of robots in the world is in Sweden, South Korea, and Germany, and the lowest unemployment in the world is in Sweden, South Korea, and Germany. We then shift human beings to do increased efficiency, design thinking, problem-solving, and so on.

Every time we’ve seen a major technology injection into the world, we’ve increased employment. We don’t decrease employment. That’s the optimism side. The pessimism side is if it actually does take over in this way.

Emad, you’ve made that distinction: in the past, labor and capital needed labor, and capital does not need labor anymore. That’s a massive discontinuity that we’re going to have to deal with and absorb in the next few decades. It’s a big one.

When I did a podcast with Ray Dalio, we were discussing this. The purpose of the Fed and interest rates is that you lower interest rates in order to get money flowing so that companies can buy equipment and hire workers. But what happens if, when you get access to low-cost dollars, you buy more AI and more robots?

Yeah, I think the problem that’s being articulated here is that there’s no obvious solution on either the optimistic or pessimistic side. On the pessimistic side, UBI and how that would work are really not well articulated. On the optimistic side, even the most optimistic folks—which I think we probably count as—believe that there’s going to be a time period where we’re not ready. It’s happening too fast for us to know what to do with these people.

Let me ask you your opinion of that time period. I think AI is going to give us this incredibly hopeful, optimistic, and abundant future a decade out or thereabouts.

My concern is the 3-year time period: instability from countries not understanding how to deal with this, the rate of change, and people being so attached to their jobs that they are losing their identity. What do you think about that stability and that time period, Emad?

Yeah. So I wrote a book about this, The Last Economy. It's been 1,000 days since ChatGPT came out. I put it at 1,000 days from now: human cognitive labor is negative in value. You're the dumbest person.

Let's talk about that one. It's a really important point you make, and I want people to understand it: the idea that both human labor and human cognitive value are negative, or that human labor and cognitive capacity are negative in value. Can you double-click on that for me?

Yeah. If you're on a team and everyone is a genius who can think and work around the clock, and you can't because you're human and you can't access infinite numbers of GPUs, then you're the dumbest person on the team. You drag down the whole team in terms of its coordination.

Now AI can think longer, and it can perform tasks proactively within 3 years. Most of these cognitive value-added jobs can be replaced. Not that they will be—like, you know, the San Francisco MTA people are all safe in their jobs as a public sector that doesn't require efficiency—but in the private sector, they can be replaced. Over the following years, they will be replaced, because again, it's the private sector.

The analogy also applies to autonomous cars. If there are autonomous cars driving and you're a human driver, you're actually reducing the safety of the roads.

And doctors with diagnosis. Why would you allow a doctor to make a diagnosis?

Oh my God. So here are the numbers, and I speak about this all the time. There was a study done out of Stanford and Harvard that looked at diagnostic accuracy. A human doctor by themselves in this particular study was getting 74% accuracy. If that human physician used GPT-4, it went from 74% to 76%. If GPT-4 did the diagnostics without the human in the loop, it was, I think, 92%. The human actually added incredible bias and misinformation into the diagnosis.

Yeah. The latest GPT-5 is probably around 98%, extrapolating the data there. To be fair to the poor doctors, how do you track all the conditions, treatments, therapeutics, and drugs?

What about the poor doctors? What about the poor patients who have to deal with them? We're talking about making it illegal to drive a car and illegal to be a doctor, basically, to diagnose.

I mean, it's going to be malpractice to diagnose without AI doing the diagnosis.

It should be, but then you've got the immune system response, right? A couple of years ago, Texas banned telemedicine because surely you have to go to the doctor in person there for every little spot on your hand.

Well, it's all going to fall apart. We'll have this Luddite revolt that we're seeing today.

This idea that I spend 8 years postgraduate to get a medical degree, and I'm spending hundreds of thousands of dollars in medical school, and then I graduate, and AI—and, by the way, a humanoid robot—is going to be a far better surgeon than any human surgeon. I just wouldn't want to be first.

But, yeah, I think it's already too late in some ways. For example, if you're a cancer doctor today, there are several hundred cancer research papers published per day, right? You have no hope of reading those. You'll have an AI read them and say, “These 5 papers are the ones you should read for the patients you're dealing with.”

That combination is where we expect to be, like we've seen in chess. The world's best chess players are a human being and a computer.

I think there is one interesting upside, though, because even if it's true that this gets better and better for diagnosis, there is something important about biological connection, about human-to-human connection. I think it'll bring the craft of being a doctor around EQ much higher.

But are you going to spend 10 years of your life and $1 million in education fees to get that?

No, but I think you'll probably get a different type of education: knowing how to use the AI to deliver the best possible care, where the best possible care includes the human-to-human connection.

Fantastic. Totally agree. It's just that the medical schools are going to evaporate.

I mean, look, this is a bigger economic disruption than COVID, and it's around the world at the same time. But there's no vaccine—

—for AI. Yeah. Again, we have to get ahead of it because we know it's coming. And again, when we look at this, people have to be given a positive view of the future to navigate what's coming.

That is needed now.

That is one of my pet peeves, one of my rants. Can I go on this one?

One of my biggest concerns is that we do not have television programs, movie programming, that gives our families, our kids, and our society a positive vision of the future. Everything is dystopian: killer robots, rogue AIs. If that's the vision you're seeing of the future, of course you think the world's going to hell in a handbasket.

We need to develop programming that gives people a vision of the future. You know the old adage: without a vision, the people will perish. One of the positive comments we get about the podcast is that we're relentlessly optimistic about the future, which is why I think Visioneering is so important. We can craft that future and say, “Let's envision that radically positive future.” And then, by the way, here's the mechanism to make it happen. And, by the way, we have a 30-year track record. It's a no-brainer.

Yeah. People need to have agency. If you feel like you're a victim of all of this and you have no ability to control it, you're going to bury your head under the pillow. But we want to say, “No, that's not the case.” That's why we're here at Visioneering. That's what we all believe in: that the best way to predict the future is to create it yourself. Let's create that positive vision of the future that we so dearly desire.

I think that also, when you look at Visioneering and everything, the core team is very easy to build. The extended team is so hard. This is the most fantastic time ever because you've got digital and physical buddies. Yes.

They allow you to have such massive leverage in the world. And again, as you said, you need to have the right mindset to do that, though.

Let's jump into the endpoint of this, which is the economy and the implications of this. I think to put the amount of spend in perspective, it's important to realize that the current AI boom is still relatively small as compared to past AI tech expenditures.

Here we are: that red line at the far right is about a 1% spend of U.S. GDP for our current AI investments. It compares to U.S. investment in railroads, which was at 3.5% of GDP at the time. Electrification was 2%, and even the internet buildout and telecom was at 1.5%. So it's not out of whack.

We've talked about this. We're in a war footing right now. For good reasons or not, we're in a war footing against China. I'd rather be in a war footing against dystopian uses of AI.

The good news is we've gone from oil to silicon as that major foundation for the future, which is fantastic.

Yeah. This is an eye chart, but let me just call it out. You can see everything at the far left starts at 0%. This is the price changes in the U.S. since 2000, over the last 25 years.

At the bottom, on the far right, televisions have been demonetized by 96%, cell phones are down 41%, clothing is up 1%, and new cars are up 25%. But above, we see hospital services at a 256% increase, which is insane given what we just said about AI and robots coming in. It should be demonetizing that. College tuition is up 187%.

We've talked about this. I think colleges are in deep trouble. Super-deep trouble. It's better for you to become an electrician, a plumber, or a welder. Comments on this?

Yeah. Well, it looks like this stuff's getting more expensive. You look at the categories, and it's the stuff that actually matters that's getting more expensive to the average person: food, cars, housing.

It's not actually getting more expensive; it's that the dollar is being degraded.

Yes.

And it's an illusion. It's the boiling-the-frog analogy, sneaking up on everybody.

So speak about that a little bit more, please.

Yeah. Basically, what's happening is that we think everybody who gets in office starts to make a lot of noise about balancing the budget and how we're going to cut spending. Even DOGE came in and said, “We're going to cut trillions of dollars.”

It's starting to dawn on everybody that that's just not true—not now, but ever. This is a unique time in history. There is so much debt and so much interest on the debt that there's only one way to get out of it. It's not a debate of maybe we do this, maybe we do that. One answer is to print more money. Yeah.

And so what they're going to do—and we know this is going to happen in 2026—is the interest rates are going to come down. It's going to create an illusion again of things looking good. Stocks will go up, and they're going to print enormous amounts of money. The dollar's buying power is going to continue to drop, and people are going to wake up one day thinking they have money, but they can't buy anything.

Here are the numbers. The US debt has reached $38 trillion, an all-time high. Just for fun, that $38 trillion is equivalent to the GDP of China, India, Japan, Germany, and the UK combined.

Yeah. Can I rant on this for a second?

You can definitely rant about it.

We floated off the gold standard in the 70s. Ever since then, we've been printing more money to keep pace. The fundamental structural problem here is that when they floated off the gold standard, they did not realize that technology was deflationary.

Jeff Booth wrote this book called The Price of Tomorrow, where he says that for the last 50 years, every dollar increase in global GDP has come with a $4 increase in global debt. So we're borrowing from the future at 4 to 1 to fund the GDP increases of today. That's obviously not sustainable. At some point, it's going to come crashing down.

A good metaphor for this is, let's say I borrow $10 million to build a TV factory, and I plan to sell those for $1,000 each. A year later, I can only sell those for $500. A year later, I can only sell those for $250. I'm never paying back the $10 million, so I just have to print more money on an ongoing basis. This is why Bitcoin is powerful. The reason it's so interesting is that it gives you money velocity without debt.

And there's an illusion, right? When you own a house that's worth $1 million when you buy it, and 5 or 10 years later it's worth $2 million, you think, “Oh my God, I made money. Our house is now worth a lot more.”

In reality, it's not.

Exactly. It's not only not worth more; if you had denominated it mentally in Bitcoin or gold, you would actually be able to use less Bitcoin and less gold to buy that house, but an enormous amount more in dollars.

But most people don't understand that it's inflation causing your house price to go up.

That's right. And so, if you take this as a natural trend, it's not that someday this will blow up. It's already blown up. As I say, it's not a question of what's going to happen.

That green curve looks startlingly like the beginning of an exponential, and it's a bad exponential.

Let's talk about the fact that we're about to make the cost of labor and intelligence effectively zero.

Yeah.

And when we divide by zero, GDP goes toward infinity. What's the implication of that?

It's irrelevant because GDP is not a great measure of the future, right? If I create a breast cancer screening device, GDP drops.

Yeah.

Yeah. Simon Kuznets, the author of GDP, who came up with this, said it's the worst thing to measure the economy and actually said, “Please don't use it.”

I think this is a super interesting thing because basically the economy is running out of all its room at the same time as that exponential change. That feels like a bit of a coincidence, but I don't really think it is, because the solution to this is also technology.

We don't want debt-fueled money and a debt-fueled future. What you want is real productive output that impacts humans by guiding this technology appropriately. This is the defining question of our time. We've developed the most impressive, powerful technologies of the future. Do we use them for abundance, or do we use them for competition? Positive-sum or negative-sum?

Great point. Eric, let me ask you to talk about this. Central banks now hold more gold than US Treasuries.

Yeah. You have to ask yourself, what do they know? Why? This is the first time in a long, long time—decades—that they're holding more gold. Every single central bank on the entire planet will buy all the gold that they can find. This is not stopping.

The reason is because they don't want to be beholden to the dollar. There are 2 reasons for that. One is that they believe it's going to continue to be printed and continue to be debased.

The other is that, whether you are in favor of Russia—I don't know who is—or Ukraine in that war, what happened was the United States took the global reserve currency, which is supposed to be apolitical, and said, “We're going to put massive sanctions on Russia.”

Now, that might sound like a good thing. You're trying to be punitive and you're trying to rein them in. But the truth is, everyone in the world is dependent on this dollar and suddenly wakes up and says, “Wait a minute.”

They can pull the rug out from under me.

Yeah, they can do that to them. Who knows who's going to be in the White House tomorrow? Who knows what predilections they'll have? They might not like me for whatever reason. So now we have to get off the dollar.

Where do you go if you get off the dollar? You're not going to the peso. You're going to gold.

Yeah.

Or Bitcoin. Most of my assets are in Bitcoin.

Which is doing well today. All right, let's talk about a last topic to cap it off—something you might have heard of. It's kind of small. It's called quantum technologies.

Interestingly enough, today we saw Trump announcing that he wants to put money into a few quantum firms. In particular, he's looking at investments in IonQ, Rigetti, and D-Wave. Shervin Pishevar, who's here, and I took D-Wave public a few years ago. It's had an 8,000% return, which is extraordinary, through our SPAC. Today, on the news that they're considering this, these stocks have gone up 10% to 15%.

These technologies are the infrastructure of the future. Again, governments run infrastructure, and so that gives you an idea of where things are going to go.

Interesting. So, you think this stuff will get nationalized?

I think there is a good possibility that it will, because I think most jobs will be public-sector jobs in this interim period, just like we saw after 1929. There's also another theory, which is that if you're going to get out of this enormous debt and the only thing you can think of is to print more money, what if you actually put some of that printed money into something that goes up in some way that does 1,000x?

We've talked about a Bitcoin strategic reserve for the US. I'm all for it.

I want to take a moment and close on one of the most important announcements that occurred in the last 24 hours, and that's out of a company called Google. Our friend Hartmut Neven, who's been onstage—we had a Google XPRIZE that's still in progress—had an announcement from Sundar, and it's that—well, actually, why don't you tell us about it? It's the first verifiable quantum advantage: a reproducible algorithm that runs 13,000 times faster than the top supercomputer, Frontier, to do this kind of molecular-material binding.

Yeah. So, let's talk about the implications of quantum computing, because I think people have been starting to get a feel for where AI is going, and a lot of people believe that quantum will outpace AI in terms of its implications for humanity and industries.

I think with AI there's the bitter lesson, but we can tell the capabilities just by the scaling. The thing with quantum optimization is you can't. It step-function changes for the hardest optimization problems in the world—

—and it's instant.

Yeah.

Can I tell that little story? Yeah, of course. A year or 18 months ago—2 years ago—we had Hartmut Neven here.

Ten years ago, Steve Jurvetson spoke at Singularity University and talked about quantum computing. Somebody asked, “Where is all this computation coming from?” He said, “I'm going to give you the answer, but you're not going to like it.” The consensus amongst all the physicists is that we're doing the computation in parallel universes and bringing the answer back. Everybody went, “Okay.” He said, “I told you you wouldn't like it.”

Two years ago, we had Hartmut onstage and asked him that same question. We now have 10 more years of data and experiments, and many more teams have been working on it. We said, “Where is all this computation coming from?” He goes, “You're not going to like the answer.” In fact, he went further than what Steve said. He said the existence of a quantum computer would definitively show that we live in a parallel universe and that we live in a multiverse, at which point everybody needs to drink some tequila.

So, we're going to watch this space. The implications of quantum computing are on materials science, on biology, on everything. What's your favorite hope for quantum computing?

Well, I think it can help us guide ourselves to better social systems. I think those are massively optimizable, and quantum is one of the ways that we can do it.

There's an intersection here that is very powerful. We've been moving more and more stuff to AI, which needs compute. Compute will solve all these problems—we can solve all math, for example—and intersecting with quantum, you get instant solutions to all this stuff, which is unreal.

But at the pace that it's moving, we don't know the implications. For instance, if the quantum computers have already broken Bitcoin, the last thing we would know is that they've broken Bitcoin. They would already be infiltrating Bitcoin and moving things properly. Then you'd find out one day when Satoshi's bitcoins are in somebody else's wallet.

So, it's going to sneak up on us, but we have very few years to get ahead of it.

The singularity is now.

Yeah, amazing.

I'll close out by saying the conversations we have here are some of the most optimistic on the planet. It's the notion that there is no problem we cannot solve. The committed, passionate human mind is able to take on anything, especially when it's got AIs and quantum computers there to help it.

Emad, Eric, Salim, you and I will be flying east, I guess, in a few days.

The real slogan is “Head east, young men.”

Yes, yeah.

And speaking of optimism, I think it's one of the most optimistic trips that I'll be taking because it's our opportunity to decentralize AI and bring that power to the people, to the benefit of humanity. And I'm really honored to be part of that project.

Yeah, we'll talk about it on the next pod, but Emad and his team have been working on something spectacular called Sage, which is the Sovereign AI Governance Engine. We'll talk about it on our next episode, perhaps.

All right. Thank you, guys. Thank you, Peter.

Be well.