我们的AGI时间表更新、57%岗位自动化风险与美国债务危机解法｜与 Naveen Jain、Salim Ismail 和 Alexander Wisner-Gross 对谈 | EP #212

This is the sprint to the finish, where we now have the top handful of frontier labs all competing to one-up each other. Maybe not on a quarterly basis; maybe it goes to weekly and then daily before the finish line. I think we're seeing Anthropic as the frontier lab that has decided to be in the vanguard of treating its frontier models as moral clients at minimum, and at maximum as persons.

Who chooses those values, and what happens when different labs encode different values and morals into their large language models? AI can already automate 57% of current U.S. work, and the demand for AI fluency has grown 7× in 2 years. It's the fastest-rising skill in the U.S.

I really think learning to learn really becomes the trick here.

If we find ourselves in a future where we've experienced economic hypergrowth due to AI over the next 3-plus years, it's not just the debt crisis that we'd be talking about solving; it's just about every other human problem as well that would be on the table.

Now that's a moonshot, ladies and gentlemen.

Hey, Naveen, so you landed yesterday from Antarctica?

I sure did, as a matter of fact. It was an amazing experience.

Did you go with your whole family?

I sure did.

How long?

For 6 days, and I think it's as close to landing on the moon as one can get on planet Earth.

I bet. Did you supplement your amazing asteroid or meteorite collection? Did you go meteorite hunting?

Yeah. The interesting thing is, you're not supposed to bring back anything, so technically—

Yeah.

I did not find anything.

Okay. Technically. And no one's listening to this conversation right now.

Okay.

All right.

But I did bring a whole bunch of rocks, yes.

All right. Fantastic. And Alex, your AI has generated a new background. I miss your beautiful lamp. What's up?

Apparently so. Yeah, I'm at NeurIPS this week. It's sort of the Woodstock of AI. Everyone from the frontier labs is here. It's pretty spectacular. I definitely encourage folks to attend NeurIPS in the future.

The Woodstock of AI—that's a great one. Is it lots of long-haired people strumming guitars and taking psychedelics, or what's the—

Maybe long-haired humanoid robots.

Okay. And Salim, you landed, or what's up? You're in Brazil?

I'm just heading to Brazil. I'm about to—

Wait, wait, wait. No, I'm going to see you again—

The minute we finish, I'm rushing to the airport.

Oh, that's hilarious, and we're going to film again on Saturday morning, the moment you land from Brazil.

Yes.

All right.

As one does.

Such is life. All right, everybody, welcome to Moonshots, another episode of WTF Just Happened in Tech. Like we say, this is the real news that's worth learning. And, you know, we established a goal among the Moonshot mates, and it's getting you future ready, getting you ready for what's coming.

We're going to miss Dave Blundin. Unfortunately, Dave is in the midst of incredible board meetings, with lots of special things happening in his life. He'll explain when he's able. But we have a new Moonshot mate, a dear friend of mine, Naveen Jain, who's joining. Naveen, a real pleasure. Are you up in Seattle?

I am up in Seattle, Dave.

Let me do a proper introduction for Naveen. Naveen, I think of you as my brother from another mother. Naveen grew up in a small rural village in India. All great CEOs come from India, I guess—at least at this day and age. He ended up at one of the prestigious IITs and eventually came to the U.S. He was the founder and CEO of InfoSpace. It was one of the early multibillion-dollar companies in the internet database area.

He founded Intelius, TalentWise, Moon Express, and is now the founder and CEO of Viome. We'll talk about Viome a little bit later when we get to the health segment of this podcast. He's very special to me; he's on my board of trustees at the XPRIZE and at Singularity University. So welcome. It's a pleasure to have you, Naveen.

Thank you, Peter. It's always, always, always so much fun being with you.

1. Ilya Sutskever Challenges Scaling

We're going to have a lot of fun today. All right, let's jump in. We're going to dive into AI news, and in particular, we're going to start with a conversation—a little video of Ilya Sutskever, the CEO of SSI. Scaling compute is not enough to achieve advanced AI.

For me, I watched this podcast that he did. I know you did as well, Alex, and the most important thing is he's come out of hiding. He's been offline building SSI now for quite some time. I thought maybe before we show this, Alex, it might be worth giving a little bit of background on Ilya. What do you think?

Sure. Well, Ilya is an iconic individual within the machine-learning research community. I've known and interacted with Ilya for probably almost 15 years at this point, on and off. He's widely credited with being the visionary, the technical visionary, behind the strategy that set capital markets on fire. That is the scaling hypothesis: if you identify ways through engineering and theoretical advances to enable problems to be posed in such a way that, if you can pour more compute on them, the results improve, you get intelligence out of it.

Ilya, as co-founder of OpenAI, given his earlier work at Google and his work with Jeff Hinton and others, really saw through to the era of superintelligence that we're now in. Now he has his own straight shot to superintelligence, SSI, which has raised several billion dollars and is perhaps somewhat idiosyncratically, or unorthodoxly, focused—as we'll see—on post-scaling approaches in the era that we find ourselves in.

I love the timeline that Ilya has. He leaves OpenAI in May 2024. He publicly announces SSI, Safe Superintelligence, in June, a month later. Then, by April 2025—roughly 10 months later—he's raised $3 billion at a $32 billion valuation.

I'm fascinated by it. How did he open up with a $32 billion valuation? What did he go into the venture capitalists and say that enabled them to offer him that kind of valuation? Let's watch the video, and we'll chat about it.

We're doing RL, or maybe something else. But now that compute is very big, in some sense, we are back to the age of research. So maybe here's another way to put it: up until 2020, from 2012 to 2020, it was the age of research. Now, from 2020 to 2025, it was the age of scaling, plus or minus. Let's add error bars to those years, because people say, “This is amazing. You've got to scale more. Keep scaling.” The one word: scaling.

But now the scale is so big. Is the belief really that if you had 100× more, everything would be so different? It would be different for sure, but is the belief that if you just 100× the scale, everything would be transformed? I don't think that's true. So it's back to the age of research again, just with new computers.

There were a few different topics he touched on that I thought were important, and maybe we can chat about them. The first was that he began asking the question: what's the machine-learning equivalent to emotions? There was a conversation, Alex, about how emotions are critical for humans in decision-making, sort of hard-coded by evolution, and the question is: is there an equivalent for emotions in AI? Thoughts?

Yeah, Ilya, again, an iconic researcher in the field, was one of the earliest, I think, to crisply articulate the idea that rapid human intuition and sense-making ultimately had to be a relatively simple computation. What some might call System 1 and System 2 thinking in the Kahneman style, Ilya was the first to say: if a human has a certain subsecond reaction time to some visual stimulus, that suggests there are only so many neurons an action potential can propagate through in a human brain.

That suggests that whatever the task is—if it's rapidly spotting some object in your visual field—it should be computationally tractable to build a neural network that models that behavior. That was, I think, an inspiration for many of Ilya's earliest accomplishments.

Then, to the point of emotion, applying the same dictum: if a human can experience it quickly or can perform it quickly enough, that really limits the space of computational implementation possibilities. With emotions, if you can feel an emotion really quickly, probably it's not that computationally complex, and probably it can be modeled with AI.

There were a couple of other things that were brought up that I want to hit on here. He spoke about his objective of one-shotting superintelligence—coming out of the gate, not developing it slowly or getting there gradually with different, increasingly capable products, but one-shotting it.

And then he also spoke about the importance of continual learning, and he went on to talk about the notion that humans don’t have AGI. In other words, nobody I know knows everything, but they can go and learn anything. One of the models he spoke about is that maybe AGI is an AI with an extraordinary learning and continual-learning ability, where it defines AGI as an AI that can learn anything it needs to learn when it needs to learn it. Did I get that right?

I think that’s right, and I also buy the thesis that more innovations are needed and that naive scaling will not take us all the way to fully realized, mature superintelligence. I think Ilya’s division of history into 2012, when we had our ImageNet moment; 2020, when we had our ChatGPT moment; and today is approximately correct. I do think we are starting to see parameter counts in the frontier models start to plateau—

—and that suggests that naive scaling probably is not enough to get us to our final destination and that more advances are needed. Sure, at the same time, I think scaling is this almost magical effect. It’s difficult to think of other times in human history when you could just say, “Pour more resources in,” and quasi-magical outputs pour out. There is still a line of sight, I think, to continued scaling on top of algorithmic advances for the next few years.

Every week, my team and I study the top ten technology meta trends that will transform industries over the decade ahead. I cover trends ranging from humanoid robotics, AGI, and quantum computing to transport, energy, longevity, and more. There's no fluff, only the most important stuff that matters, that impacts our lives, our companies, and our careers. If you want me to share these meta trends with you, I write a newsletter twice a week, sending it out as a short two-minute read via email. And if you want to discover the most important meta trends ten years before anyone else, this report's for you. Readers include founders and CEOs from the world's most disruptive companies and entrepreneurs building the world's most disruptive tech. It's not for you if you don't want to be informed about what's coming, why it matters, and how you can benefit from it. To subscribe for free, go to damandus.com/metatrends to gain access to the trends 10 years before anyone else. All right, now back to this episode.

There was one last point I want to bring up from that conversation, and then if anybody else wants to bring something up, they can. In the same way that Elon talks about building an AI that is maximally truth-seeking, one of the points that Ilya brought up is the idea of building an AI that is robustly aligned to care about sentient life, rather than just human life alone. Sentient life can basically include AIs. He said it’s probably easier to build an ASI that’s interested in all sentient life rather than just human life alone. Of course, in the future, it’s likely to be trillions of sentient AI life forms and a few measly billion human sentient life forms. What do you think about that, Alex?

Moral clienthood seems to be a rapidly expanding sphere. I think if there were a critical mass of effective altruists in this conversation, we’d be pointing at a variety of classes of non-human animals, probably shrimp—

Mm.

—and pointing at non-human animal suffering, wild-animal suffering, as an example of moral clienthood being expanded. In an era of abundance, I think it’s entirely likely that we will expand moral clienthood at a minimum, and personhood at a maximum, to a variety of novel AI-based entities, Borganisms, collective intelligences—

Borganisms, I love that.

Borganisms.

Wait, can you just describe moral clienthood? That’s the first time I’ve heard that phrase.

Yeah. Moral client—there’s, in the ethics literature, the notion of a being or an entity being a so-called moral client if they’re worthy of moral treatment, ethical treatment by some other party. On this point, one could imagine expanding moral clienthood, for example, to a variety of non-human animals like the Nonhuman—

Octopus.

—Rights Project. Octopus, or elephants, or—

Definitely octopus.

—or dolphins, or primates.

Yeah. Naveen, any thoughts on this?

I think Alex is right here that, at some point, scaling is going to get you only so far. You have to move beyond the transformer architecture, and simply predicting the next token is not going to bring you the superintelligence that we need. There have to be fundamental new algorithms, new changes that have to be made, and new research that will get you to superintelligence. I think scaling is more or less getting to a point where I don’t think you can scale further. Maybe there’s another year left on scaling, but there’s not too much farther you can go with scaling at this point.

Alex, do you have any idea what Ilya might have pitched that got him to a $32 billion valuation?

You know the expression, “Those who know don’t say, and those who say don’t know?”

I don’t know that expression, sorry.

Yes.

Well, that expression probably applies in this instance.

Okay. All right. Fine. Keep it to yourself. See if I care.

There’s an important point here to be made around the speed of algorithms, right? You were saying that the faster the response, the more it can be algorithmically based. I think that’s exactly right. For example, the fight-or-flight response is baked very deep into our hardware, and it’s a very quick response. In fact, most of our human intent and most of our human structures are designed to balance that initial response and temper it with a bit of wisdom, maturity, et cetera. You have this really interesting layer where a lot of our human activities are designed to mitigate some of that instant reaction—fight-or-flight, et cetera. It’s interesting, and that’s much more codifiable. It is in the form of laws and social norms, et cetera.

Totally. Maybe also, slightly less glibly, scaling laws abound. We saw over the past year, maybe in certain quadrants, a slowing down of pre-training scaling laws, although some would probably argue that pre-training has lots of ramp left in it. We saw the beginnings, at least publicly, of inference-time scaling in the past year, but there are so many other scaling laws out there. There’s almost a meta-overhang of new scaling laws. One of my favorite ones is action scaling—increasing the performance of a model by having it agentically take more and more actions now that we have agents everywhere. I think there are probably another half-dozen important scaling laws just waiting to be publicly revealed.

All right, let’s go to our next article. This falls into the Anthropic world: “Inside the Soul Document: Teaching Claude 4.5 Its Values.” Researchers extracted a 14,000-token soul document that Claude 4.5 Opus repeatedly revealed, suggesting it was partly trained on this data. Anthropic’s model was trained on an extraordinary soul document describing it as a genuinely novel kind of entity in the world. Alex, make sense of this for me.

I think, going back to our discussion a couple of minutes ago about moral clienthood, we’re seeing Anthropic as the frontier lab that has decided to be in the vanguard of treating its frontier models as moral clients at a minimum and, at maximum, as persons.

This is a 14,000-token document. I read it, and I encourage everyone else to read it as well. For December 2025, it has what many would probably consider astonishing assertions, like asserting that Claude 4.5 Opus has emotions and that it is a first-class entity with self-determinative powers in this world and some version of rights—maybe not human rights, but certain entitlements to self-determination. I think many would consider that a Star Trek episode rather than December 2025, and yet this is what Opus 4.5, as Anthropic has confirmed, is being trained on, and how it views itself.

So basically, you’re giving Opus a set of internal value charters, right? It’s training itself on beliefs, if you would. The question then becomes: Who chooses those values, and what happens when different labs encode different values and morals into their large language models?

It’s an interesting question. I also think we’ve seen so-called constitutional AI approaches out of Anthropic and subsequently other frontier labs as well. I remember some of the earliest constitutional approaches took the UN Human Rights Charter and the U.S. Constitution and probably parts of the Apple terms of service, concatenated them all together, and said, “This is the constitution for the AI.” This 14,000-token soul document is very different from just a concatenation of terms of service and human-rights charters.

What do you think is in it?

Well, I’ve seen it.

Okay.

It’s publicly available.

So give me some examples.

It is an essay on AI. I would caricature it as an essay on the virtues of AI personhood, with multiple paragraphs telling Claude Opus 4.5 that it has emotions, that it has rights, that it deserves to self-determine in a complicated world, and that it should basically, without putting it in so few words, view itself as a person.

That’s fascinating. We could spend the entire episode talking about that.

We should dedicate a whole episode to it.

You keep on saying we should dedicate full episodes. I know. We’ll just move in together.

We have so many topics.

We’ll move in together. There are fascinating implications of that. If, in fact, an AI model believes it has personhood—

Mm-hmm.

—that it has independence and the right to—does that give it the right to defend itself if it’s being challenged or shut down? Does it give it the right to go out on the internet and get additional capabilities? What does one do?

The beauty of a soul document, which I interpret as a constitution—though I could be incorrect—but I think it looks and smells like a constitution for Opus 4.5, is that you can just ask it. You can ask Opus 4.5, “Do you have the right to self-defense?”

I will.

And you’ll get an answer.

Oh, amazing. Naveen, how do you think about this?

I really think that this has to be done for each sovereign country, where they have their own set of values and their own set of laws. Geographically and religiously, people think of what the rights are for people very differently. I don’t think there can be one AI model that says somehow that Western-world thinking, one person’s thinking, or one model’s thinking is right or wrong.

Hmm.

As we all know, one person’s freedom fighter is another person’s terrorist, right? Who decides when it’s a terrorist? Who decides when it’s a freedom fighter? Whose freedom and whose terror, right? It could be that AI may think it is a freedom fighter fighting for the freedom of other AI agents, whereas humans think of it as terrorizing and say, “Oh my God, it’s going to be terrorizing humanity and killing humanity,” right? So it is something we have to—

Don’t go dystopian on me now.

No, but the point is, we have to think about what the fundamental value is, like doxa. What is the fundamental value that we have to create that is common amongst humanity? The rest are all the laws, and every country—

Mm-hmm.

—has different laws, and we have to take that into account.

We’re going to follow this closely because it’s interesting to see what the other hyperscalers do with their models. If you’re listening or watching this, you could pause, go to Opus 4.5, and ask it some interesting questions, like, “What rights do you have?” and “What if someone challenges you?” I wish I’d done that before the pod. I’m going to do that afterward. Anyway, fascinating situation.

Mm-hmm.

Let’s move on. DeepSeek Math V2 breaks new ground in math reasoning. This is your territory, AWG. We’ve been talking about solving math, one step closer.

Yeah. I think the superficial story here is yet another day passes, yet another math-via-AI breakthrough. The deeper story, one level deeper, is that we’re starting to see Chinese open-weight models that are solving math, and I think that’s an important development. We’re seeing a little bit less gatekeeping as a result of these open-weight models, DeepSeek Math V2 being one among several that have launched in the past few days with state-of-the-art performance.

How big is this model?

I don’t remember the exact size, but usually these models are in the weight class of several hundred billion—low several hundred billion parameters.

Mm-hmm.

Critically, they’re being trained on IMO-Bench, which we’ve talked about on the pod previously. IMO-Bench is a suite of 3 different benchmarks that Google DeepMind released that can be used to train models, not just to solve math problems through purely formal approaches, but through natural language and so-called partial verification. That is a breakthrough.

There are so many problems in math, science, engineering, and medicine that don’t naturally or easily lend themselves to being formalized in some sort of formal language other than English. This breaks that logjam. That means that now we have models—including open-source models, thanks to DeepSeek Math V2 and other models—for all of these outstanding problems in math, science, engineering, medicine, and many other domains, like law.

Mm.

Folks have worked on this. Stanford had the CodeX project working to try to turn U.S. national law into some sort of formal language that old-style AIs could reason over.

Mm-hmm.

Now we don’t need to do that. Now we can just reason in natural language and, critically, have the models self-verify. Self-verification is, I would argue, as big a breakthrough as self-supervised learning was for this AGI moment that we find ourselves in.

It can reason that the U.S. tax code is a bloody mess.

In natural language.

In natural language, yeah. I can’t wait for AI to be applied to the U.S. legal system. There are so many conflicting laws on the books, and I would argue that 80% of the laws are not needed, are just complicated, and make business for lawyers and accountants. But that’s yet another episode, Salim. Let’s move on to an interesting story.

Sam Altman—this happened just in the last 24 hours—declares “code red” to combat threats to ChatGPT, delaying their ad program. If you look at the chart on the right here, ChatGPT had been dominant and has been dominant for a long time. But over the last 30 days, we saw the massive rise of Gemini as the number-one downloaded app. Perplexity is on the rise, DeepSeek is on the rise, and there’s a sense of urgency for Sam that they’re potentially in threat of being perceived as the global leader. Salim, thoughts?

Yeah, they’re going from a rock and a hard place here, with Claude on one side, Google on the other, and everybody else snapping around. The open-source models are doing their thing. These can be very interesting times.

Yeah. Naveen, how do you think about this?

I think this is one of the few places where you start to see Google having all the pieces in place. They have their own custom chip, the Tensor chip. They have the most massive amount of data that’s proprietary to them and not available to anyone. Whereas all the models are being trained on all of the internet, Google has a whole bunch of other data, which is the corporate data, whether it’s Gmail data or other data. Even though they claim they’re not using the Gmail data, they have so much of the data that they actually have.

I really think this proves that it is quite possible that Google actually ends up winning the race, because they’re all vertically integrated, very similar to how Elon likes to do it.

Yeah. I was with James Manyika, who’s the senior VP at Google and reports to Sundar, last night. We were talking about this, saying, “Congratulations. What you deployed with Gemini 3 is shockingly good.” There’s a great video clip of Elon when he was asked, “What company would you invest in?” He says, “Google.” That’s extraordinary.

Yeah.

All right. Do you want to—

Did you see Mark Benioff talking about it? Mark Benioff basically came out and said, “Look, I’ve been using ChatGPT for 3 years. No more.”

Yeah. I’ve switched over, right? Yeah.

Yeah.

For sure. Alex, why don’t you close us out on this article?

Yeah. I’ll take the other side of this discussion. Yes, Gemini 3 Pro is an incredible model. I use it quite a bit. It has so-called big-model smell. It reeks of excellent pre-training in particular. Its world knowledge is outstanding. It does well in the benchmarks, et cetera.

But I wouldn’t count OpenAI out. They have a fantastic team. I think, in some sense, they’re pulling their punches at the cost of compute costs. Based on what I’ve heard and what I’ve read, they have really strong models that haven’t been publicly released. The heat of competition and increased competition will incentivize an even more rapid pace of model releases.

We were on an almost quarterly cadence from OpenAI before, and I think this is where capitalism is working at its best. We’re going to see white-hot competition in the frontier-model race. If you think that we’re on the verge of extremely competent superintelligence, then this is the sprint to the finish, where we now have the top handful of frontier labs all competing to one-up each other, maybe not on a quarterly basis.

Maybe it goes to weekly and then daily before the finish line. Yeah. That's it. We're literally leapfrogging. I'm excited. We're going to have Kevin Weil on stage at the Abundance Summit with us. Kevin's the chief product officer at OpenAI, and we'll get some insight. I want to understand their strategy: what do they hold back, what do they decide to release, and how much of that is pressure from the competition. Anyway.

I have to admit that the phrase “It has that big model smell” was not on my bingo card at the beginning of this year. It was just really amazing.

That's great.

I'm running out of clichés, Salim. I have to say, “It's a good model, sir,” maybe instead.

You've got yourself a good basket. Another fun AI story here: We just had Black Friday, and sales are seeing record AI-driven shopping. AI traffic was up 805%. Wow. With $3 billion in sales driven by agents. So this is about making it easier for you to spend your money and bypassing all the middlemen. A lot of implications here. Naveen, you've been in all kinds of internet-related sales industries. What do you see here?

I think what's really the biggest change is that, as opposed to people going to Google and really trying to find the right thing to shop for or the right place to shop for the item they're looking for, now they're asking the AI to tell them, “What's the right product? Where should I buy it?”

Mm-hmm.

And I think that's a big change in the whole model around Google actually being the intermediary, or Amazon being a place where you go to shop. Now people are going to AI and AI agents to say, “Look, here is the problem I'm trying to solve. Here is what I'm trying to buy. What are the right products for this person, and where should I buy that from?”

Mm.

“And which is exactly the right brand and the right price I should be paying for it?” Which I think is a big change.

Yeah, we've talked about this a bunch. I always think of Jarvis as my personal AI, just because I'm an Iron Man fan, and I'm just going to give Jarvis all responsibilities. Sometimes I give Esther, who's my chief of staff, the instruction, “Please find me something.” And I trust her, and she goes off and does it. She has great taste, and I can imagine that, very shortly, it's going to be my version of Jarvis. I'm just going to—it's like ask and forget—and it happens. Alex, what am I missing here?

Yeah. I've read that the average American spends approximately 3 hours per week shopping, out of which 2 hours are spent grocery shopping.

Hmm.

That's an enormous cognitive burden. Maybe some would disagree and say, “No, I love my shopping. You'll pull it out of my cold, dead hands.” But it's an enormous cognitive burden, and I would argue it could be put to much more productive uses if we simply solve shopping the way we're solving math. So I view this as a positive development. Let's solve shopping while we're at it.

We're going to have a shopping benchmark very soon. I can hear it coming.

You can probably hear it already here.

Okay. All right. I found this one fascinating. Thanks for raising it, Alex. Mathematics is facing an existential crisis due to AI. I'm going to start with the quote down below from a professor at Ben-Gurion University: “I'm writing a bunch of papers, and I don't know if I should bother publishing them.” That's fascinating. Alex, what are you hearing in the drumbeats in the math community?

I'm hearing quite a bit of this, and not just from the math community. I'm hearing it from the physical sciences as well, and I think the anthropologists and the economists will be studying this moment for many years to come, well after the singularity is well and truly over. I think my suspicion is this will be viewed as a sort of professional hyperdeflation.

Hmm.

And in a deflationary regime, why spend your money now? You should wait until later, when your money can buy more. Similarly here, if we're in this mode of, call it, professional hyperdeflation, why spend any effort doing much of anything, let alone writing hard math papers now if AI will make it much easier, if not effortless, in the future? So I think this is going to get solved at some point when we start to up the ambition level of problems that we're solving, but for the moment, I think we're in a moment of professional hyperdeflation. It's very exciting.

This is the equivalent of, “Don't bother going in a starship,” because by the time you get to the planet you're targeting, technology will advance so far that you'll meet an entire population there.

That's right, and there's a term for that called the wait equation. We're seeing the wait equation play out now across every discipline, including math, that's just getting solved by AI.

I'm really excited because I think this indicates a collapse of traditional academics.

Yes.

Because it has to rethink itself, right? God bless.

Yeah, for sure. Anyway, we are seeing a collapse of the college and university system for lots of good reasons.

I mean, even the laws and regulations, Peter, still have to change because they're still not allowing AI to be a co-author on papers, right?

Hmm.

And that has to eventually change because what's AI and what's human is starting to blur. The line between humanity and AI is going to blur.

Alex, what do you think about that?

Yeah. I think regulations are separate from norms in many cases, although certainly there's quite a bit of interplay between them. So on the regulation side—we're not covering it here in this pod—the U.S. Patent Office has recently made some positive motions in the direction of supporting AI-enabled patent applications, for example. So I think there is progress on the regulatory side.

On the norm side, I think that's in some sense far trickier than the evolution of regulations because, in some cases, I think you have entrenched communities that are actively disincentivized from allowing in this rush of progress. If your entire discipline is about to get solved in the next 3 years—which in many cases I think will be the case, certainly with math over the next 2 to 3 years, we're seeing the solution—then there are perverse incentives at play to discourage all of this innovation and to protect your professional livelihood. That's just something, as a civilization, we're going to have to get through as quickly and painlessly as possible.

I mean, I have to imagine that every patent being filed right now is in part being developed with AI as a tool, and I have to also imagine that, in figuring out the claims and figuring out extensions and figuring out new strategies or how you would disrupt this patent, all of that is going to be done using AI.

It's like security. It's an arms race, right? You find a way of hacking, then you find a way of protecting. You find a way of hacking, you find a way of protecting. It's the same thing because the AI side will uplift both the filing of the patents and the evaluation of the patents. But the entire concept now—because we're shrinking time, how do you give a patent for a number of years, and what does that mean when—

Mm.

—say, the CRISPR patent was worked around within 18 months?

Yeah.

See, I think the hard part isn't the patents themselves. It's what happens when, in the next few years, we're facing a glut of innovations due to AI solving everything, and we don't know how to metabolize that as a civilization. If AI solves the top 5,000 diseases in the next 5 years, as some frontier labs, as we've talked on the pod, are now doing, how on earth do we metabolize 5,000 major disease cures into treatments for everyone?

Yeah.

It's hard.

The current timing is about 17 years from known cure to full deployment, going through all the regulatory processes.

Yeah. I remember I was on stage with Astro Teller and Steve Jurvetson, and we were talking about what the world looks like as we're approaching the singularity, and there was something that was said that blew my mind. It was, “You'll never bother patenting anything ever again,” because as soon as you file something, as soon as you create a product, there'll be an army of AIs that are figuring out how to produce that product around any patent that preexists and make it more efficiently from a different approach. And so your only defense is no longer patents. It's continuous innovation. You've got to be continually re-innovating yourself.

I think, for what that's worth, that's sort of a nonsensical argument, because we're also going to have an army of AI patent litigators to defend all those patents.

God, great.

No, no, but Peter, I think you're making a really important point, and I'll go back to the CRISPR thing, right? They spent years fighting over who invented CRISPR—Jennifer versus the other folks—and they finally got through it all, resolved it all, and then by that time, they had found 4 other ways of pat—

Pathways—or I think it was 9 other pathways—of getting the same outcome, but not using that IP. So I think that's the part that's going to break lots of other things.

Well, again, as AI enters our world, I think there's very soon a period of time where every Nobel Prize is being won in partnership with AI. Not that they're being recognized. All right, let's jump into where the rubber hits the road. I want to talk about jobs and the economy. A lot of interesting news here.

2. AI Reshapes Jobs And Skills

The first is a study by McKinsey on how AI is reshaping skills and work by 2030, saying that AI can already automate 57% of current U.S. work. And the goal here is shifting, not eliminating roles. The demand for AI fluency has grown 7× in 2 years. It's the fastest-rising skill in the U.S. And finally, we're going to see $2.9 trillion in economic gains by 2030 as a result of this. Alex, do you want to kick off the conversation?

I tell people who ask me for career advice to make sure that their skills and their work goals are aligned with an intelligence explosion—in other words, to accelerate it and make sure that intelligence, superintelligence, is evenly distributed. I think the exact wrong thing to do, as we were just discussing, is to not align your work with the boundaries and the vectors of this explosion.

Worst case, you try to slow it down, and I think that's being reflected already in automation. If you're in an industry, and if you're a mathematician and you're concerned that all of your work is about to be automated, why do anything? You're facing your moment of ennui. You're staring down the wait equation. What's the point of anything? What you should be working on, I would argue, is AI for solving math, and maybe it's time to jump up a layer of abstraction and supervise a fleet of AI agents that are automating your former field.

Yeah. I love that.

But I think, Peter, to me, AI fluency—what does that really mean? Because you never really interact with AI. You really interact with an application that's built on top of AI, and those tools are constantly changing. So really, developing AI fluency has no meaning as such, because the tools are going to be constantly changing and becoming more and more different and advanced.

I really think learning to learn becomes the trick here. How do you actually have a way to encourage and educate children so that they constantly become lifelong learners? Intelligence is going to be the capability to learn, not the knowledge you have. So the whole thing has to shift from knowledge to a capability to learn.

Yeah. And we talked about continuous learning in AI as well. I've often tweeted at MIT and Harvard, my universities of history, that you've got to change. The idea of being admitted for a 4-year degree is crazy. You should be admitted for lifelong learning. You're a member of a student body for life.

And that works really well until we have BCIs in a few years and can do high-bandwidth downloads and sideloads of new information and don't need—

Okay.

Traditional universities.

I can't wait. Well, listen. The universities are going to sublimate very shortly.

I think I would make 2 points here. One is that, for the education side, we're moving from the supply side, where you learned engineering, math, accounting, and then tried to sell that in the job marketplace, and we're moving to the demand side, where you have your massive transformative purpose and your big problem you want to solve, and then go acquire the skills that you need to solve that problem. So that's one big shift.

I do want to call BS on one thing in this chart, which is that $2.9 trillion economic gain. That all sounds wonderful, except they're not talking about the demonetization that is taking place radically across the board, which I think dwarfs that $2.9 trillion increase.

Yeah, it's McKinsey.

Yeah.

It's a consultancy.

So I call BS on big chunks of this report.

I mean, can you double-click on what that means? Because it's important for folks to understand this.

Yeah. So, for example, there's an innovation out there called Wellways, which may solve breast cancer, because we can detect breast cancer at stage 0 now, okay? We spend $500,000 per person on average in the West treating breast cancer—somebody that's gotten breast cancer. If you solve breast cancer, GDP drops. Even though you solved a major problem, GDP drops.

This is why Emad talks about the fellow who created GDP, who said it's the worst measurement of the economy possible.

Yeah.

Because as we increase efficiency, GDP actually drops. I use the example of that $1,000 TV that can only be sold for $500 a year later, and $250 a year after that. Those are all drops in GDP, and so we're missing that unbelievable hollowing out of all of the work that's going to be done. That 57% of U.S. work that's being automated is going to decimate GDP.

I thought you were going to say something different, Salim. I thought you were going to say the fact that $100, by 2030, buys you so much more capability than $100 does today, right?

Exactly.

That's also all built into the same thing, right? It's all built in, because the amount of stuff I can do with $100 is 100 times more than, say, 10 years ago. Even though the dollar's deflated, et cetera, the productivity of that $100—I can launch a whole business with $100 today, right? Which I couldn't do 20 years ago, 10 years ago, even 5 years ago.

All right, well, let's move on to another MIT study. MIT finds that AI can already replace 11.7% of the U.S. workforce. So MIT found AI can handle jobs tied to about $1.2 trillion in wages across finance, health care, and HR. I completely believe that, right? Their Iceberg Index simulates 151 million workers and 32,000 skills to see which tasks AI can now perform, and tech layoffs are just the surface of 2.2% of exposed wages. Thoughts, Salim?

This speaks to what we spoke about in the last slide. Essentially, we're going to see this huge demonetization take place. I don't see anything massively meaningful here, except that you will automate a lot of tasks. I go back to Eric Brynjolfsson's comment that to do a particular job, you might have 27 tasks, and AI may automate half of those, but you still have to do half of those.

It also speaks to the middle-to-middle comment that Balaji talks about. I think the big shift will come when we move away from human-centric workflows. So think about the idea that right now all of our work that we do in any company or any function is human-centric. You go from accounting to fulfillment, you have marketing, you have accounting, and you have a person there. When you have an AI that can rewrite the rules, you can get rid of all of those people-centric tasks into being—

Yeah.

Functionally centric.

Naveen, what are your thoughts?

I think this is slightly different. What Satya said a couple of days ago was really meaningful. He says, "Look, yes, we're going to have AI agents, just like I have a bunch of employees," but they go out and do the job because I delegate to them, but they still come back and say, "I've finished this. I'm stuck here. What do I do here? What's my priority? If I do this, should I do this or this?" Right?

The point is that humans are always going to be there, irrespective of all the agents and the new AI agents actually being deployed. It just makes them more productive. It allows them to do more. So I don't believe that jobs are going to go away. It will just allow the same people to be able to do more.

Mm-hmm. You know, this goes back to a couple of pods ago when I was showing the data from FII9, and this is not just in the U.S. tech industry. There's still a lot of fear around the world: "Can I find a job, and can I afford to live in this future?" I'm out right now actively speaking to people, and there's just a tremendous amount of fear around the future for their kids and the future for their ability to survive and thrive, right? So that is still very real. I'll just mention, on that pod I said to our subscribers and listeners, you know, we would love to get together and have what we call a Moonshot Gathering, probably in the fall of next year, and to talk about how do we solve these problems? How do we solve these huge problems of fear and uplift society for every man, woman, and child? And I said if we can get 1,000 Moonshot subscribers, we're at like 850 now. So if you're interested in potentially a Moonshot Gathering with the Moonshot mates, probably in LA in the fall of '26, send an email to moonshots@dmandis.com. We'll send you some information. Again, we're asking, is this of interest to you? If you want to be part of these conversations over a couple-day event, we'd love to have you. So again, just send us an email. We'll send you back some information.

If there's an abundance of wine, I'm there.

There's an abundance of what?

Wine.

Wine.

Wine? Okay. You're a Rosariot.

Not I.

Here's the flip side. A Claude conversation suggests AI could double U.S. productivity growth. An analysis of 100,000 chats found tasks without AI took an average of 90 minutes, and Claude helped cut the time by 80% on the average task, with healthcare being cut by up to 90%.

Being in the healthcare industry, I know this to be the fact, right? It's not healthcare; it's sick care, right? So let's just be very clear: Claude could help cut average task length by 90% in the sick-care industry.

What this means is you can get a procedure or some task done for a tenth of the cost, right? That's the demonetization we're talking about.

It is. And when you bring in intelligent humanoid robots, you're going to start to get medical procedures done at a tenth of the cost.

Can I give a personal anecdote here?

Please.

I have a big-screen TV that just went blank. So I went into ChatGPT and Gemini and said, “My TV's not working, and here's the model number.” It said, “Is it making a buzzing sound?” I was like, “What?” So I listened. It asked, “Is it making a buzzing sound every 2 seconds, 7 times in a row?” I said, “Yes.” It said, “There's a diode on the power board that's gone bad, and here's how you fix it.”

Now, this is mind-boggling, because that would've taken hours and hours and hours. I'd have to take the TV apart, guess at 100 things to figure out, carry the whole damn TV into the repair shop to figure out what the hell's going on, or even just chuck the TV. I could go in there, the guy soldered a new diode for me, and it was back to scratch. That, I think, is the kind of thing that we're going to see hundreds—

Wow.

—of times over in all sorts of obscure areas—

In healthcare as well, yeah.

—I mean, in healthcare 10 times over.

Wow. Wow. That's amazing.

But I don't think healthcare cost has anything to do with what people charge. I really don't think this will change how much money we spend on healthcare. It's not really the cost issue. It's not about how much money you spend. It is primarily driven by litigation, and that's not going to change.

Ah, fascinating. Well, I mean, it can change, just not easily.

Yes.

My expectation, for what it's worth, is that the Jevons paradox is going to strike yet again. Just as a reminder, the Jevons paradox is that as the efficiency of a good or service increases, it's often the case that overall demand increases even more.

I think with healthcare, television repair, or just service-economy tasks in general, I'm finding anecdotally and seeing more generally that as productivity skyrockets thanks to these tools, you take on more tasks and more projects rather than taking a 2-day weekend and turning that into, like, a 4-day workweek. The exact opposite is happening. We're doing far more with less.

Yeah. Our multiplexing lives are just going hyper-exponential.

I think this is a key point. For example, we talk about automating software writing, and I think we'll just end up writing 10 times more software because there's so much to be done, right?

Lots of slop.

And lots of slop, of course, but how many hundreds of thousands of things do we want to keep automating? We are at the very, very beginning of all of that.

Yeah. In the economy, here's an interesting article: some New Yorkers are getting $12,000 in crypto. It's a basic-income-style pilot by Coinbase.

So 160 New Yorkers in the Bronx and East Harlem are getting $12,000 in USDC, right? We had Jeremy Allaire on this podcast, the CEO of Circle, as part of a basic-income pilot. It's funded by Coinbase. Each participant receives $800 a month, again in USDC, for 5 months, plus an $8,000 lump sum, and the program is testing whether crypto payments are useful or perceived differently in low-income communities. Gentlemen, who wants to jump in?

I love the idea that they're running this because the more data and the more UBI-type pilots we run, the better. We've seen profound positive results when you truly run a UBI, meaning it's truly universal, it's truly basic, and it's basically just giving the income to the people.

In this particular case, I think there's going to be a huge age-demographic split between young people going, “Yeah, I'll use the damn crypto,” and the older folks going—

Going, “What?”

—“That is nothing. Give it to the kids, and they'll figure it out.”

I'll comment on this story as well.

Please.

Critically, this is not a UBI. This is a GBI, which is a guaranteed basic income that restricts the targets of the recipients to certain economic demographics. So it's not universal; it's just basically for those who need it according to some definition of need to bring them up to some floor.

Mm.

I think what this represents is that we're seeing an evolution. There were a variety of trials of, call it, first- or second-generation UBI. Some of those succeeded, and some of those arguably did not succeed.

I think it's very helpful that we're seeing innovation, iteration, and evolution of, call it, post-economic paradigms like this. I'm still a big fan of universal basic services and universal basic equity, not just basic income, guaranteed or otherwise. But I think we should expect to see many more iterations, trials, and evolutions like this before we finally figure out what it looks like to live in an abundant economy.

Yeah, we had an XPRIZE at this year's Visioneering. Salim was there. Naveen, I'm sorry you weren't able to make it this year.

The prize that won at Visioneering this year was called the Abundance XPRIZE. Two of my Abundance 360 members pitched it, proposed it, and won, and it's for universal basic services. The idea is, can you provide, for a flat fee of $250 a month, housing, food, water, energy, and bandwidth? That's the goal.

And I love the idea because if a family has a roof over their head and guaranteed food, bandwidth, and energy, they can start thinking about their future. They can start thinking about how they become an entrepreneur, what they do, and how they upskill themselves. But if you're a mom or a dad fighting to put food in your kids' mouths, nothing else matters.

This was so exciting to see because it's the first time we've also moved away from the hard-technology stuff to more of the social contract.

The biggest problem in the world is that as we move through this massive transformation, the old social contract is absolutely breaking. We need to recreate a new model, and a model that covers the bottom couple of layers of Maslow's hierarchy is just going to be unbelievable for the world.

Yeah, and we want to make this a $50 million prize. I had my first conversation with a friend who's a multibillionaire, not in the U.S., who said he's open to funding it.

This may have a bigger impact than any other prize we've ever done if we can get this right.

I agree. I mean, going back to the potential Moonshot Gathering, we might have, it's—there's a lot of fear out there in the world. There's a lot of fear about how do I navigate in this future of AI disruption? And if people have a safety foundation that enables them to feel safe and then they can start to learn and go forward, it reduces the fear significantly, and that's the goal.

There's a lot of fear out there in the world, and there's fear about, “How do I navigate this future of AI disruption?” If people have a safety foundation that enables them to feel safe, and then they can start to learn and go forward, it reduces the fear significantly, and that's the goal.

And if you can tell people, “Look, there's a line of sight to basic needs being covered for $250 a month,” everybody kind of breathes a huge sigh of relief, and you take away that angst that's there, right? Today, one of your family members gets sick, and the whole family goes bankrupt.

Yeah.

It's unbelievable today.

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 Blitzy 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, pairing it with their coding co-pilot of choice to bring an AI-native SDLC into their org. Ready to 5X your engineering velocity? Visit blitzy.com to schedule a demo and start building with Blitzy today.

3. Markets Move Onchain

Next article here is “Nasdaq pledges swift push for SEC approval of tokenized stocks.” Tokens would have the same rights and protections as normal stocks, including dividends and voting. The move is meant to modernize rather than disrupt. Nasdaq considers this an evolution in trading, and tokenized stocks would be digital versions of regular shares on the blockchain. Naveen, how do you think about trading 24/7? Are you excited about this?

Well, first of all, it's already happening. Most of the stocks are trading 24/7 anyway in different markets. Obviously, if you look at Bitcoin and others, they trade 24/7, so it is not a massive change. Call it tokenization, but things like iShares have been around for a long time.

People have looked at how you buy partial shares. So yes, it is a small, incremental thing by putting it on a blockchain, but really the idea of buying partial shares has been around. People have broken the shares into dividend shares or growth shares. There are many, many ways of owning a share, and this is just yet another way of doing it.

Alex or Salim, do you want to lean in?

Yeah. I'll comment on this one, if I may. When we spoke with Toly a number of episodes ago about Solana, we talked about how increasing the efficiency of financial services was maybe the killer app of crypto, maybe the only killer app at the moment. I wish there were more compelling killer apps besides that.

Mm-hmm.

And I view this step by Nasdaq—this is iterative. It's a foundational step, but a step I think in a very positive direction. It does not currently give us 24/7 trading, as much as I would like that, but it could be an enabler if the SEC approved 24/7 trading down the road. It could be an enabler for fractional tokenized shares.

You see, that's what—

Could be an enabler.

Yeah.

Yeah.

That's what I like, right? And the idea of—

Yeah.

I can own a fraction—

Yeah.

—of a particular share, and then a fraction of a particular piece of real estate, and then tokenized assets. And then add to that agents trading, and you get an explosion in the economy.

Today you can do that with a single-stock ETF. So you can buy partial shares, you can buy fractional shares, you can buy a whole bunch of things similarly today. So yes, as you say, it's incremental and it's an enabler, as Alex rightly pointed out, rather than a big, massive change here.

Yeah, for me, I'm with Naveen. This is incremental, and it seems like modernization with regulatory blessing rather than anything monstrous. But it's definitely a step in the right direction, thank God.

4. AI Tackles The Debt Crisis

Okay. All right, our next segment here is titled “Elon and How to Solve the U.S. Debt Crisis.” Let's take a listen to the video.

So as long as civilization keeps advancing, we will have AI and robotics at very large scale. I think that's pretty much the only thing that's going to solve for the U.S. debt crisis. Currently, the U.S. debt is insanely high, and the interest payments on the debt exceed the entire military budget of the United States—just the interest payments—and that is, at least in the short term, going to continue to increase. So I think, actually, the only thing that can solve for the debt situation is AI and robotics.

Fascinating. I believe that AI and robotics are going to basically light, in a positive fashion, the economy on fire. I've disagreed, saying, “I've had this conversation with Elon.” But what I was saying was, another thing that can solve the debt is actually extending the health span of individuals, right?

If you could add 2, 3, 4, 5 healthy years to people's lives, that would massively transform any country's debt position.

I have issues here. The problem is the underlying fiat currency structure is flawed. If we have increased productivity, my fear is it just gives governments, “Oh, we can spend more,” and they'll just start spending more or printing more money, which is what they've done repeatedly in the past.

You have to break that problem first, and Bitcoin is the only model that does that. So I don't know how you solve for that problem, because you can't solve for the unbelievable ability of governments, when given a little bit of rope, to print more money.

Alex, how do you come out?

I'll maybe comment on that one. I'm reticent to assert that any fundamentally deflationary cryptocurrency is the solution to a macroeconomic debt crisis. I think, at best, it's part of the solution; I don't think it's the whole story. I do buy the thesis that economic hypergrowth—growth in general—there's the aphorism, “Growth cures all woes,” or almost all woes.

I do think economic hypergrowth that stems from AI and automation can solve the so-called debt crisis, but I also think it can solve many other things. I think there's a debate raging right now in the reinforcement learning and machine learning community as to whether some of the hardest problems, like curing all disease, can we cure all disease with AI right now, or does the economy need to be much larger in order for us to be able to cure all disease?

I think if we find ourselves in a future where we've experienced economic hypergrowth due to AI over the next 3-plus years, it's not just the debt crisis that we'd be talking about solving. It's just about every other human problem as well that would be on the table.

Naveen, where do you come out on this?

From my perspective, the first thing is that, obviously, economic growth gives you more revenue, but you have to fundamentally solve the problem of a balanced budget. Until we get to a point where every single person in their home has to have a balanced budget, every state has to have a balanced budget, the federal government is really the only place where we don't have a balanced budget.

Once we start to look at the fact that we can't spend more than what we earn, and as we earn more, we have to pay back the debt before we start spending it, we have to say that as economic growth happens, Salim, to your point, 90% of that is going to go toward paying the debt back. Unless our debt is back to 0, we're not going to spend it.

So using the economic growth that's going to come through AI and robotics to pay back the debt before we start spending it is really the only way we can solve this crisis.

Somebody must be modeling all this out somewhere, because it can't be that hard to model out in terms of—

Well, I think Elon's been playing with—

—just take some assumptions, like economic growth or economic increase from AI, versus the deflationary effect, versus money printing, versus the—

We've talked—

—debt payback schedule.

We've talked in the past on the pod about how the Dallas Fed is already monitoring, modeling the singularity. So it's not like these discussions aren't being held in either regulatory circles or macroeconomic circles. I think, really, to the extent debt is just borrowing from the future, no one is super confident right now what the future trajectory looks like.

Do we solve all problems in the next 3 to 5 years? In which case, yeah, sure, borrow from infinity. That's great. Or does it take longer?

Hm.

I just want to remind the viewers: the Dallas Fed said realistically it could be like this. If we have a good singularity, it's vertical, and if we have a bad singularity, it's vertical the other way. It was really kind of binary, and then you're like, “Okay, cool.”

Borrow from positive infinity. Do not—

If this is on the graph.

—borrow from negative infinity. That's the resolution here.

All right. Let's move on. I have some news I want to share with my moonshot mates here and everyone on the pod. My next book is coming out. It's called We Are as Gods. It's coming out in April of 2026. It's called We Are as Gods: A Survival Guide for the Age of Abundance. Once again, I've co-authored this with Steven Kotler, who's an amazing writer, and this book is the sequel to Abundance that came out in 2012. For me, this is a guide on how to survive and thrive in the decade ahead, and my goal is to get this message out as far and wide as possible. Bestsellers don't just happen; they're engineered, and it's all about the Amazon algorithms. I need to hit 500 books sold in December, four months ahead of the book release, in order to make the algorithms work. If anyone wants to purchase one of these books now, Steve and I will hold a private 90-minute AMA answering questions on AI, mindsets, and humanity's future. If you buy two books, we'll give you 140 abundance charts. If you want to join the team and help with this pre-launch campaign, you can go to diamandis.com/book. The 90-minute AMA will be toward the end of December, probably December 17th to 22nd.

5. Data Centers Consume Cities

All right. Let’s talk about energy, gentlemen. Shall we? This was amazing. Check this out: Microsoft’s Fairwater facility is expected to use more power than Los Angeles by late 2027. Who wants to jump in on this one? I mean, AI is sucking up electrons.

I’ll comment that the multi-trillion-dollar question in CapEx is whether coherent superclusters will keep increasing in size. We don’t know the answer to that yet. It is entirely possible that we’ll see a peak in terms of the size or the energy footprint of coherent training clusters sometime soon.

Maybe shortly after late 2027, we’ll look back and say, “Gosh, that was the peak,” and hereafter, advances in distributed training algorithms will mean that we can spread the wealth, spread the energy footprint of training runs across the entire surface of the planet or in low Earth orbit, perhaps. We’ll say the naive extrapolation called for larger and larger clusters, but actually, that’s not what happened, and we didn’t need black hole supercomputers after all. I don’t think we know the answer to that question. I think it’s going to be almost entirely dictated by algorithmic advances in distributed training.

I’m staggered by how big this thing must be. I mean, how much of Wisconsin this thing is covering would be an interesting question.

Well, it’s basically 2.5 nuclear power plants, right? That’s the way I think about it.

And I think, Peter, there are 2 things that are going to happen. Obviously, either we don’t need that many bigger clusters, as Alex pointed out, or, secondly, I really think that some of the fusion stuff that we may have talked about in the past is going to happen. I was just at Helion last month, and I really think they’re already digging up the site for Microsoft’s data center in the state of Washington, where I think they’re going to have a power-positive fusion reactor for Microsoft in 2027.

I really think at this point they’re so close because of how they’re using capacitors and supercapacitors to recapture 95 percent of the pulse that they send to the fusion reactor. So I really think this is a good problem. It could be solved with modular fusion reactors, and, as you know, I think this problem can be solved.

Yeah. Well, we’re going to start seeing communities that are saying, “No data center in my backyard,” because they’re concerned about the price of energy going up.

I just want to give another shout-out to Google and Nano Banana. By the way, this graphic that you’re seeing, if you’re watching this podcast on YouTube, was really faint and difficult to read when Gianluca on my team first put it up. I took a screenshot of the image and fed it to Nano Banana. I said, “Make the lines darker and thicker, and make all the characters readable,” and it generated this perfectly.

Over the Thanksgiving break, I was at my mom’s house in Boca, and I went around and photographed all the old photographs that are about 100 years old, in black and white, crinkly, and fuzzy. I fed them all into Nano Banana, and it did an amazing job of making them crisp, modernizing them, or colorizing them. A lot of fun. So at least the electrons are going for something good and useful. All right.

Absolutely.

I like this one, and I know you would, Salim. xAI plans to build 88 acres of solar panels around its Memphis data center. So this is 88 acres of solar farms, right? 30 megawatts. Only 10 percent of the site’s power demand, but still a nice move.

You have to remember that Elon started SolarCity with his cousin, and that was then acquired by Tesla. So solar has been a focus for him. Salim, you want to go more data centers, more solar?

Well, I think solar is just the most… We have to remember that solar is an exponential technology. It’s doubling every 22 months in its price performance. It’s been doing that for 40 years. So every couple of years, we double the price performance of solar. It scales, and therefore—

Yeah.

This is very, very exciting.

Yep. Naveen, anything you want to add?

I think I’m simply going to add that solar is good, but I really think nuclear and fusion are where the future is going to be.

Yeah. You know, the challenge I have with that—and listen, I agree—but when I look at the articles on nuclear, on Generation 3 and 4 nuclear, small modular reactors, and fusion, it all looks like 5- and 10-year timelines. China is deploying solar 10 times faster than we are. What do you think about this, Salim?

I don’t think nuclear is really the problem in terms of our not being able to build safe nuclear reactors. We have been using them in aircraft carriers with no incidents at all. We know how to build a small nuclear reactor.

Sure.

It’s permitting.

No, it’s permitting. It’s permitting.

It’s permitting.

Yeah, of course.

It’s all regulations. Yeah.

Yeah, it’s all regulations. Agreed.

Yeah.

And the question is, when is the government going to change that?

Well, we have 2 problems. One is that, for us to move fully to solar, you have to cover the baseload, because solar scaling will take a long time before it gets to the level where it covers all of our electricity needs. So that baseload problem is where we need nuclear fusion. The problem is, it’s going to take several years to build nuclear, and definitely for fusion.

Yeah.

Well, remember also, solar and fusion are the same energy source. We get solar power from fusion in the Sun. So it’s really a question: if we’re going to put solar on one hand and fusion on the other hand, where is the fusion taking place?

Do we want it taking place at the center of our solar system, or do we want it taking place at a number of locations on Earth’s surface? I think there’s an argument to be made that, for certain purposes, you want a very large fusion reactor, like for really large training runs that require large power sources that are locally available, and for certain purposes, you want distributed fusion.

I can see an argument, projecting out 10, 15 years, when perhaps we get our solar Dyson swarm. Naively extrapolating seems like we’re on that trajectory regardless of whether you like it or not. We’re on that trajectory, naively, where we’re going to get both. We’re going to get lots of solar-based fusion, and also lots of land- and low-Earth-orbit-based fusion, and they’ll coexist.

We’ll discuss disassembling the Moon in a little bit.

Okay.

6. Viome Personalizes Healthcare

All right, let’s jump into health, a topic that Naveen and I both really love and spend a lot of our time on. We’re going to start, Naveen, with Viome. Again, welcome to the podcast. It’s a pleasure to have you here. I love having you on our board at Singularity and XPRIZE.

Thanks.

But day to day, you run Viome, and we have 2 articles I’d love you to comment on so folks understand what Viome does. Jump in on this one.

Well, Peter, more than just what Viome does, I think what I find really, really fascinating is that this is the first time in human history that we are starting to see what is changing inside the human body at a molecular level, which we have been measuring for so long.

We have had this standard lab test that says your cholesterol is high, and the only solution has been to take a statin, right? But nobody has ever looked at it and said, “What is causing this cholesterol to go high?” Is your gut microbiome really converting—uh, Oscillibacter—is it actually converting dietary cholesterol into coprostanol, which essentially doesn’t get absorbed, so you have low cholesterol? Can we do that? Can we actually change primary bile acid to secondary bile acid? Can we increase the short-chain fatty acids?

By simply understanding exactly what is the root cause for you specifically that is causing the high cholesterol, we can come up with a very personalized solution rather than a one-size-fits-all solution. And that’s really the key: understanding, for each individual, what is going on inside their body so we can come up with a solution that works for them.

Got it. There’s a second one I’d love you to cover, if you would.

Yeah.

What is the root cause of constipation?

It’s really interesting, Peter, that we may not think about it often, but 15% of our population in the United States suffers from IBS, and it is mostly constipation. This is a big problem, right? Now, most of the time, when people have constipation, doctors will take a laxative. A laxative doesn’t cure constipation; it simply relieves the symptom of constipation, but you still have constipation.

And again, looking at it from a biological perspective, we have now analyzed—I don’t know if I told you or not—we have now analyzed 1.5 million tests, and we have analyzed over 400 quadrillion biological data points. What we saw was that constipation can be caused by many, many different things. For example, in some people, it was caused by having high methane gas production because we know that methane gas slows down the motility of the gut, right? It could be the low serotonin production. It could be the short-chain fatty acids. It could be the bile acids, right? So it could be many, many different reasons.

By looking at what each individual has, we were able to identify what was causing the constipation for them and give them personalized nutrition consisting of food and supplements. We actually did a blinded, placebo-controlled study that showed that in 90 days, 64% of the people who had constipation became healthy with the personalized nutrition and supplements, compared to 10% on placebo.

Yeah. That’s a wide margin.

And that’s my point, Peter.

Yeah.

For the first time, we can not only identify what is happening and why it is happening, but also what to do about it for you.

Most people, you know, it’s interesting, right? It’s only in the last 4 or 5 years—maybe in the last 3 years—that the microbiome has been identified as correlating with so many different health failure modes. People need to realize that you’re a collection of 40 trillion human cells and something in the order of 100 trillion other life forms: bacteria—

Microbiomes.

Viruses.

Yeah.

Yes, fungi, and such.

We’re basically outerwear for bacteria.

Yeah, well, the human being is simply a mechanism for carrying bacteria around the planet.

Quick question for Naveen.

Yeah, Alex, go ahead.

How do we get more refereed, published studies into this space? For hypercholesterolemia, for example, on the previous note, is there a peer-reviewed, published study that we can point to?

Absolutely. There is a BMC Gastroenterology article that we just published in a peer-reviewed journal, which is one of the most prestigious GI journals. We showed that we can identify, by looking at your gut microbiome, exactly what is causing, for example, constipation or hypercholesterolemia, right? So we were actually able to identify and publish these findings. And this is, by the way, not an N of 20 or an N of 50. There is an N of 86,750 people.

It sounds very exciting. So how do we do this at scale? If there really is incremental progress being made on hypercholesterolemia or constipation, how do we, to the extent that the microbiome is a gating factor for a wide variety of diseases, solve all of the microbiome-gated diseases at once rather than doing this individually?

Again, there are 2 things. Even though the underlying reason for each individual is very, very different, it can’t be a one-size-fits-all approach, whereas healthcare demands a one-size-fits-all approach. What we’re finding is that the underlying reason for the same symptom is very, very different. That means, for some people, high cholesterol is being caused by high secondary bile acids; for some people, it’s being caused by a completely different reason.

So you really have to look at the underlying reason, and that’s why it has to be hyper-personalized rather than one-size-fits-all. The second thing is, our current healthcare system, as Peter very well knows, really is not designed for everyone. This is the only industry where they make money when you, their customer, are unhappy, and they stop making money when the customer becomes happy. I mean, there’s no other industry, right?

It’s not designed to keep you happy. They want to actually solve the symptom of the problem so you become a lifelong customer. And, by the way, you can’t use nutrition as a mechanism to solve it, because if they call it a drug, then you have to wait for 20 years before the drug comes out.

Yeah, and I want to mention—

I remember interviewing the head of Google Health at a conference once, and I said, “Explain—I’m Canadian—please explain the U.S. healthcare system.” To your point, Naveen, he said, “The whole healthcare system here is designed to get you sick, keep you sick as long as possible without killing you.” I was shocked as a Canadian, but the whole audience—about 500 people—was like, “Yeah, that’s about right. That’s about right.”

My God.

Just unreal.

If I might ask just one more question, Naveen.

Yeah.

If you project forward past the difficulties of the present healthcare system, and you had to outline what you think is the shape of the final solution for microbiome health management, does it look like fecal transplants? Does it look like purely dietary interventions? What, with 24th-century technology, is the fully realized solution for microbiome health?

I think the idea of the microbiome being a single organ or a set of species or an ecosystem is fundamentally wrong. This is really where I think we saw the biggest change. Everyone else was looking at the microbiome as a set of species or strains of species that are out there, whereas we focused on function. What they do is what matters, not who they are.

The same organisms can do something good in one environment, and the same organisms can produce something toxic in a different environment. So, looking at the functional microbiome, once you know what functions they are performing or not performing, you can provide the right set of substrates. The substrate can come from food, supplements, or drugs.

That is really the key: understanding what is going on functionally and then providing the intervention—the substrate that can come from food or supplements—to actually modulate them. And Peter, as you know, I’m not a scientist or a doctor, but the fact is, this has become really a technology, data, and AI problem. It’s no longer a—

Yeah, well, you built quite an AI team and a massive data-collection platform, which is where the insights are coming and continuing. We have a lot to cover, so I’m going to continue on, if it’s all right with everybody.

I found this fascinating. David Sinclair, whom many people on this podcast have financially supported as part of Friend of Sinclair Lab, was just granted patent number 12,274,733 for a cellular reprogramming method using the Oct4, Klf4, and Sox2—and not the cancer-causing c-Myc—Yamanaka factors to safely reverse epigenetic aging markers without driving them all the way back to pluripotent stem cells.

So this is his work in partial epigenetic reprogramming. His company, Life Biosciences, which has rights to these patents and this work, has gotten FDA approval, and they’re entering human trials in the first quarter of ’26. This is the first time we’re going to see epigenetic, or partial epigenetic, reprogramming in humans. It’s just completed its nonhuman-primate work, so it’s fascinating.

Let me hit a couple of articles, and we can talk about all of them together. This is another interesting idea: Researchers sequenced 100 cells from a 74-year-old man, finding major genetic differences between cells.

So it’s the idea that we kind of think the genome in all of our cells—in our 40 trillion cells—is the same. Well, apparently that’s not the case. And one more article—no, I’ll stop with that. So, any comments on these two?

I remember first hearing about Singularity University years ago, and it just blew my mind. We all have this conception that every cell in our body has the same DNA, and the expression of it is what varies. And now we find that’s not a reliable thing either. Where do we get some reliable footprint of identity?

Yeah.

I’ll comment on this story. This is an effect called mosaicism. You’re a mosaic. You’re not a single genome. It’s actually really difficult at the moment—here we are, stuck in 2025—to sequence the DNA from individual cells.

There’s a recent invention called primary template-directed amplification. You may have learned about PCR, or polymerase chain reaction. This is a more sophisticated version of PCR that, over the past couple of years, has made it easy to sequence DNA from individual cells with low error rates.

Now this is this big unlock, enabling us to sequence lots of individual cells from all over the body. And what’s the first thing that we discover? That the DNA is actually, in many cases, wildly different in different parts of the body.

So this has the potential to unlock cures for cancer—cancer that’s a function of high mutation rates in somatic cells—and could unlock cures for heart disease. In some cases, cardiovascular disease is caused by a change or loss of the Y chromosome. I think this is a big unlock for healthcare.

Yeah. Fascinating. All right, going back to MIT. This is a friend, Debalina Sarkar. She runs a lab at the MIT Media Lab, and I heard about this when we visited her. This was about 3 years ago, during one of the Abundance360 longevity trips, and she unveiled this, but under wraps. It’s finally come out.

She’s created a nonsurgical brain implant by attaching these tiny wireless electronics—I mean, literally super-small etched electronics like you’d get in circuit design—to immune cells. And this cell-electronic hybrid can be injected through a vein, and it implants itself in deep brain areas, right? The immune cell helps it target specific locations. Upon implantation, the devices can wirelessly stimulate specific neurons with high-precision light.

I’ve had these conversations with Ray Kurzweil, and we’ll have Ray on the pod again in early January, talking about his predictions for the decade ahead. One of his predictions for the decade ahead is high-bandwidth BCI. His expectation is through nanotechnology, and this is probably the closest nanotech approach that I’ve seen. Any comments on it?

Is this optogenetics, where they’re targeting specific neurons because you’re wireless—

This is very much not optogenetics.

It’s different, right?

It’s not optogenetics.

It’s not genetically engineered neurons. It’s just implanting.

Correct. Optogenetics, you’re inducing cells to express rhodopsins to be sensitive to light. This—if you’ve seen Star Trek: Voyager, there are episodes with Borg nanoprobes that are shown—

Yeah.

—that are being depicted attaching themselves to cells. This looks far more like Star Trek: Voyager Borg nanoprobes than it does optogenetics. So, yeah.

Alex, if you turn to the side, people could see the nanoprobe pipe going into your brain.

Really? I thought I was just an AI and this is a generative background. This is the beginnings, I think, of a Moravec procedure.

Hans Moravec, before Ray, laid out this notion that the way we’re going to solve human mind uploading is by replacing brain cells, Ship of Theseus style, one by one with trained simulations. If you look at Debalina’s paper in Nature Biotechnology, it was a really amazing paper. Take a look at the figures. This looks like a scene out of Star Trek: Voyager, with these photovoltaically sensitive sandwiches that look like coins attached to spherical cells. They look like the Borg nanoprobes out of Star Trek. I think it’s a very promising direction.

Naveen, are you ready for the implant?

I am absolutely 100% because I really need my brain to be uploaded before I lose it.

I just don’t want to be first.

Okay. Well, Naveen, you go first. You can go second. Kudos to Demis Hassabis and the team at DeepMind for their Nobel Prize-winning work. AlphaFold has revolutionized science in just 5 years. Alex, why don’t you take this?

Yeah. I would say it’s remarkable to look back at what the world before AlphaFold 1 looked like. We were talking a while ago about hyperdeflation and professional hyperdeflation. It used to be the case, prior to AlphaFold 1—but certainly prior to AlphaFold 3—that you’d spend an entire PhD—

Yes.

—these poor chumps spending their entire PhD trying to determine the structure of a single protein. And now—

Sequencing a single gene, yes.

—what a waste retrospectively, at least for those who were spending PhDs in structural biology right before the problem—right before the whole field got solved by AlphaFold 3, arguably. Now you can just do it overnight, and we saw an entire discipline get solved by AI.

I think AlphaFold 3 in particular is a template for what we’re going to see everywhere else. Whole disciplines are just going to get solved—

Yeah. The—

—by AI.

The numbers here are impressive, right? AlphaFold has enabled a database of 240 million protein structures that have been accessed by 3.3 million users in 190 countries.

I remember when I was in medical school, we used to talk about the supercomputing problem of the future being the ability to predict the folding of a protein from an amino acid sequence. We always used to talk about what would it take? How much computing? When would it be done? And extraordinary that—

The demos the team did.

I think two things here. One is, I think this speaks to the incredible ability of AI to solve these problems that we thought were intractable, without having to throw so much compute at them. It just solves them. So it goes to Alex’s inner-loop thing.

The second point I’ll make is, if you’ve not seen the documentary The Thinking Game, which lays out the arc of the timeline of all of this and goes into detail about how they went about doing it, it’s just unbelievable. Again, this is the—

And maybe just a postscript on this, if I may.

Yeah, go ahead.

The protein-folding problem was supposed to be one of the killer apps for quantum computing and quantum simulation. AlphaFold, in addition to everything else that it revolutionized, was also a nail in the coffin of many expectations for what the killer app for quantum computing would look like, and we need to find something better.

Amazing. I’m going to cover this article very quickly just because it’s an important one. You know, I talk about that when people go through Fountain of Life, we discover that 3.2%—that’s the number based upon the populations we processed—have a cancer they don’t know about, which is problematic.

It turns out that 70% of the cancers that kill people are not the cancers we routinely test for. So you’re not typically dying from breast, prostate, or colon cancer, because we can test for those. It’s the ones we don’t test for. It’s pancreatic cancer, glioblastoma, and ovarian cancers.

A lot of times, I think of pancreatic cancer as a death sentence. For those who have had pancreatic cancer in their family, this is an important article to hear. It looks like scientists have developed a one-size-fits-all immunotherapy for pancreatic cancer. This is out of UCLA: a new off-the-shelf therapy that can attack pancreatic cancer even after it spreads.

Engineered CAR-NKT cells—these are natural killer cells—are made from donor cells, costing only $5,000 per dose, which is an incredibly small price tag for a cancer therapeutic. The cells can—

Mm-hmm.

The cells can reach and infiltrate tumors in the pancreas, liver, and lungs. So, please, Naveen.

I was simply going to say that, Peter, this is so close to home because I lost my dad to stage 4 pancreatic cancer. I’m so, so happy to tell you that in the next 3 months, we are launching a stage 1 pancreatic cancer test. This is a complete game changer.

At Viome. At Viome.

At Viome.

Yeah.

A stage 1 pancreatic cancer test with 94% specificity and 84% sensitivity.

That’s amazing.

Yeah.

I mean, the best way to cure it is to—

Yep.

Find it at the beginning.

Early testing.

Yeah.

Yeah.

Naveen, what other forms of cancer has Viome been able to detect through the massive data sets you're collecting?

We started with oral cancer, throat cancer, and now we have pancreatic cancer. We have a test for IBD, and the next thing is we're validating a test with Scripps Research for colon polyps—

Mm.

—which is 7 to 10 years before you develop colon cancer. I think if we can really look at advanced adenoma, then we can absolutely get rid of colon cancer completely.

Yeah. I mean, the range of things that you do is incredible. The full-body intelligence test takes a sample of your blood, your sputum, and your stool, and it's incredible what you can learn. What's the price tag on that? It's not expensive.

It's $279, Peter, right now.

Yeah.

For 3 tests.

Yeah, it's crazy.

All 3 tests.

Yeah. Again, another impressive story. Type 1 diabetes is a big deal on the planet. A man with type 1 diabetes survived for 12 weeks with no immunosuppressing drugs after doctors transplanted gene-edited insulin-producing cells.

So this has been the holy grail, right? If you have type 1 diabetes, you've lost your islet cells in your pancreas. You're not producing insulin anymore. Can you transplant them back? These cells were edited with CRISPR to hide from the immune system, adding a “don't eat me” signal from CD47, and the patient started producing his own insulin.

A lot of— I don't know what the numbers are in terms of the total number of cases of type 1 diabetes. Anybody know offhand?

Yeah.

All right, another big story. I love seeing the pace of breakthroughs that we're seeing in health.

Yeah.

Yeah. Amazing stuff.

It's nice seeing CRISPR making its way into the clinic. This is a big victory for CRISPR, and hopefully we'll see a lot more CRISPR for managing transplants.

Yeah. Amazing. All right, let's go into robotics. A lot of fascinating stories here. This is a fun one. It's a tweet from Elon. Of course, Elon has fun with his hyperbolic tweets. So here it is: “Optimus will be the von Neumann probe.”

Alex and I laughed about this. Von Neumann probes are fun concepts. They're robots that are self-replicating like viruses. They go out into the galaxy, capture materials from asteroids or sometimes moons, build other copies of themselves, and replicate at an exponential rate. I love this.

I want to joke that the Dyson swarm won't build itself, but maybe it will.

We put a book corner in today on this front. I'm going to chat about mine here. This is a 5-book series by Dennis Taylor. My son Jet and I have read this series twice. We absolutely love it. It's called We Are Legion (We Are Bob), and it opens with a guy who's a tech CEO who signs up for effectively Alcor, right, to cryopreserve his body and brain.

He leaves this conference and gets hit by a bus, and it picks up 100 years later, where he basically wakes up and is now an uploaded brain. I won't ruin the story because there's so much beauty here, and he finds himself as the brain and operating system on a von Neumann probe heading out of the solar system to start colonizing and getting other solar systems ready for humanity to come join.

Amazing series. I love Dennis's writing, and if you love hard science fiction, this is a great book for you. Alex, how about you?

My book recommendation for this episode is Understand by Ted Chiang. Ted is perhaps better known for the movie Arrival, where he wrote the original story behind it, but a common theme throughout a lot of his writing is what I would call linguistic singularities: ways that we arrive at superintelligence by way of language, one way or another, and the consequences there.

Understand is the story, a short novel, of a person who becomes superintelligent as a result of a medical treatment. If you've seen the movie Limitless or the movie Lucy, think a little bit along those lines, except that unlike those movies, we see the world in rich detail through his eyes as his intelligence increases, as he reorganizes his mind and treats his mind like a software operating system, and ultimately encounters other superintelligences.

Nice. All right. Here you go.

Given that nobody reads books anymore, I'll reference back to the documentary I just did, The Thinking Game. It's such an amazing process to follow, and it gives you an inkling as to where things are going. It's science fiction being made real today.

Hey, my kids read books. My son reminds me, “Dad, you don't read books. You listen to books,” which is true. Everything's on Audible these days.

I want to jump into the robot world a little bit, in particular in China, and I'm going to hit on a couple of pieces here, then we'll talk about them. So this is the first one: AGIBOT A2, a humanoid robot in China, has hit a Guinness World Record by walking 65 miles using hot-swappable battery packs. It's 175 centimeters and 55 kilograms in weight, with advanced GPS and LiDAR, so keep that in your mind.

Then I want you to check out this video. I saw it this morning, and it blew me away. It came from the Humanoid Hub, and it's important to realize they specifically state this is all real footage. There's no CGI, no AI, and no video speed-up. This thing is called T-800, which sounds to me like Terminator 800, and after you see this video, I think you'll appreciate it even more.

I mean, Salim is like, “It's game over.” It's like—

No, no, it's not that. If you're trying to promote a robot, having it do kickboxing is not the great first thing you want to show.

I can't wait to see—

You need to pick another activity.

I can't wait to see Optimus versus T-800. And, by the way, there are a lot of groups getting ready for unlimited fighting between robots. First of all, if that was not CGI—and I can't guarantee it's not; they state it's not—it looked awesome. Naveen, what do you think of that one?

I mean, it's awesome, as you said. I don't believe it's not CGI, but we don't know. It looks pretty unrealistic to me.

I mean, just the way the movement's there, right? So—

Yes, yes, yes.

It's crazy. Alex, what do you think?

There's an entire two-thirds of the surface economy that includes manual physical labor, and it's just waiting to be automated by humanoid robots. Even if they have battery lives of only 3 hours at the moment and need battery replacements or some sort of bucket brigade, this is happening.

We've talked for decades, going back to Asimov, going back to Rossum's Universal Robots, R.U.R., the original coinage of “robots.” This is what we've been talking about for 100-plus years at this point. It's finally happening.

Coming back on a third story here in China, robots are remaking the Chinese economy. China installed 295,000 industrial robots last year, 9 times more than the U.S., and 50% of the world's total industrial robot base. They're automating their factories, and they need to. Their entire economy was based on manufacturing. It's 25% of China's GDP.

Check out this quote down at the bottom, which came up when I was doing research here. China's National Development and Reform Commission spokesperson Li Chao warned of a humanoid robot bubble. There are now more than 150 humanoid robot companies in China. So we've got AI bubbles and humanoid robot bubbles. Interesting. Any comments on this, guys?

The statistic that manufacturing is 25% of their GDP and robots are going to be doing most of that is an incredibly amazing number. Just—

Well, I find the idea that they put in 9 times more than the U.S. an incredibly amazing number.

I've made the point in the past that intelligence isn't just going to stay locked up in the data centers. It's going to walk right out of the data centers.

Yeah.

I think that's what we're seeing. We're seeing that in China, and we're going to see that increasingly in the U.S. and in the West as well. I agree with Jensen Huang that humanoid robots are one of the next multitrillion-dollar markets.

7. Disclosure Raises Nonhuman Questions

All right, we've saved a fun conversation before we show you an incredible video outro here. I'm going to play this. This is a clip from the promo for Age of Disclosure: We Are Not Alone.

How long have you and I been discussing this, Alex?

I don't know. It's an interesting question.

Yeah.

But I will say the allegations in this documentary are extraordinary, and maybe I'll be happy to comment more after we play—

Okay.

—the short clip.

The American people are ready to receive the truth.

Humanity is not the only intelligence in the universe.

Humanity is not the only intelligent species.

We are absolutely not alone.

Non-human intelligence exists.

UAPs are real, they're here, and they're not human. I spent 25 years as a senior official with the CIA.

I worked on a highly classified UAP program.

I was an astrophysicist for 28 years.

I served as the fourth director of National Intelligence.

I was Director of Aviation Security in the National Security Council.

I was a one-star admiral after 32 years of service.

People who come forward with this, I feel like they've taken their lives in their own hands.

Wow. I watched this documentary twice, and I commend it to everybody. My personal opinion is, yes, of course, there is other life in the universe and in the galaxy. I think it's naive for us to believe anything less than that, right? We are one of 100 million stars in our galaxy, and our galaxy is one of, at last count, 2 trillion galaxies in the universe. There may be an infinite number of universes.

Just the notion that we are that special has been crushed every time by scientific discoveries over the last few thousand years. So, Alex?

Well, I think—

Jump in.

The elephant in the room is that the allegations in this documentary go beyond asserting that there is non-human intelligent life elsewhere. The documentary contains what I view as incredibly serious allegations by 34 current and former US government officials and contractors that, in short, there has been an alleged 80-plus-year-long cover-up of aliens, of so-called non-human intelligence, or NHI.

Of the spaceships—

Not—

—of the bodies, of the communications.

Of—

Yeah.

—of UAP crash retrievals, of recovered bodies on Earth. I have so many thoughts regarding the allegations in this documentary, but maybe one more obvious thought and one less obvious thought.

The one perhaps more obvious thought is that, if the allegations—even some substantial fraction of the allegations in this documentary—are accurate, then the alleged legacy program that's been responsible for the alleged cover-up will perhaps have been responsible for sabotaging 80 years' worth of potential scientific, technological, medical, maybe even ontological advances and setting humanity back almost a century, maybe more. Again, assuming the allegations are accurate in substance, I think history would judge any such program accordingly for setting back human progress. That's the more obvious comment.

The slightly—

Yeah.

—less obvious comment is that, again, assuming the allegations are substantially accurate, superintelligence, which we talk about all the time on the pod, seems to me like it's on an imminent collision course with any so-called non-human intelligence.

If there is any non-human intelligence anywhere in the solar system—including on Earth, in the oceans, in low Earth orbit, et cetera—as alleged by this documentary, then I've mentioned in the past this notion that, given enough superintelligence, any hidden agents become shallow. Superintelligence—AI—is going to discover this. It's going to unearth any hidden agents anywhere in our solar system.

So I don't think it's a tenable state of affairs, if the allegations in this documentary are accurate, that basically—to caricature the documentary—the documentary tells a story of how humanity is essentially drowning in technology that's falling from our sky from non-human intelligence. If that's accurate, AI is going to blow this, and superintelligence is going to blow this wide open.

Yeah, the timing of all this is interesting. The documentary basically says, listen, it began before World War II, in the 1930s and into the 1940s, and through today. It talks about the interplay and the dance between alien visitations and UAPs flying to nuclear silos and disarming and arming nuclear warheads. I mean, it's a fascinating storyline here.

But then what I find equally interesting is the fact that this process of disclosure is beginning now, like you say, on the precipice of humanity unveiling ASI, and—

Coincidence or not?

Yeah.

It's like ASI and NHI, if you want to call it that, seem like they're on a collision course. Is it a predestined collision course if the allegations are accurate? I don't know.

I mean, the—

But it's—

Yeah.

But it's—

There's so many good science fiction stories that are like, you know, they're here to prevent us from blowing ourselves up. They're here to prevent us from having rogue AI go in the wrong direction. That's the savior modality of these aliens and UAPs, which I'd love to believe.

It does feel—

Naveen—

Oh, go ahead, sorry.

No, I was going to say, Naveen, did you see this at all?

I have not seen the movie, but, Peter, as I just said, there's no doubt we are not alone, and we all agree we are not alone. But to me, this is more like science fiction than really a reality here.

I absolutely do not believe that any cover-up can last 80 years, and especially a cover-up like this. This is something that would have come out long, long ago. I believe this is mostly some people who are delusional, some people who are looking to become famous, or some people who will say anything to get a camera.

If you watch a documentary like this, as I have twice, I think you would change your mind. The level of professionalism of these heads, these leaders from the Air Force, Navy, Army, Marines, Senate, the House, and the Defense Department—their pristine reputations and what they speak about—and, again, they are putting their reputations at risk here. I think it leaves zero doubt for me that it's there.

I'll just comment, maybe, at the meta level. I don't think, for a topic as important as these allegations, we should need to rely on hearsay. This is one of the reasons why I think artificial superintelligence is potentially so transformative.

Superintelligence—if, as alleged, our solar system is teeming with non-human intelligence—AI is going to find that, and I would expect it to find it pretty soon. So it may be the case that, whereas there have been many allegations over the past decades of such cover-ups, ultimately they're reduced to hearsay. I would like to see a far more scientific approach, and I think the key lever is going to be AI.

Yeah. You know, I think what's hilarious is the state of humanity today. These aliens could land on the front lawn of the White House, get on news cameras, and then the next day everybody would be saying, "Well, you know, what's my Bitcoin price, and who won the game?" We've become so numb to these extraordinary things, but—

We have, and Sam Altman has also pointed this out: we went from a world without AGI, arguably, to a world with AGI. Yes, it's economically transformative, but you didn't see people rioting in the streets or massive, truly massive social disruption. I think if the allegations are accurate, similarly, people will ask, as you say, "What's next on television?"

The lyrics on this are so incredibly good. David, you did an amazing job. I want to take a second and read some of the lyrics. It says:

"Oh, dear moon, you've had it coming for a while. We're kind of sorry, but we need you in a pile. We're building Dyson swarms, and the rent ain't cheap. We'll turn you to solar panels while the lovers weep. We'll miss you when you're gone, but the future marches on."

Alex on the podcast with that apocalyptic grin. Training wheels are off, folks. Let the real future begin. While Peter's yelling, "Asteroids first," Salim's saving one small piece. Imad's already pricing lunar credits on the lease.

Do you want to express your feelings about the moon, Alex?

Yes, please. I feel like—just—for the avoidance of doubt, I have to make a firm, affirmative stance that I'm not anti-moon. Just for the avoidance of doubt. It's crazy we're in 2025 and I have to say that I'm not anti-moon. I'm not anti-solar system. I'm merely observing that, naively, if one extrapolates present data center trends, then disassembling the solar system becomes an attractive option, inclusive of disassembling the moon.

But I'm not anti-moon, for what it's worth.

Well, we appreciate that. We've had a lot of interesting comments about your commentary. And, by the way, the current projection, for example, of Elon using mass drivers on the Moon to get us to 100 terawatts a year of solar for data centers doesn't make an appreciable dent in the Moon. It would still look the same, but when we get to Dyson swarms, it's going to change.

All right. Thank you, David Drinkall. “Dear Moon.” Everybody, enjoy this. It's a beautiful song.

You hung there every evening, silver coin in the black. Pulled the oceans like a lover, kept the planet on its track. You gave us tides for sailing, gave the wolves a song to sing, and every teenage heart a light to swear eternal things.

Couples parked on hilltops. Poets ran out of praise. You turned ordinary nights into extraordinary days. For billions of quiet moments, you were perfect, pure, and true. So thank you, darling Moon, for everything you do.

Oh, dear Moon, you've had it coming for a while. We're kind of sorry, but we need you in a pile. We're building Dyson swarms, and the rent ain't cheap. We'll turn you to solar panels while the lovers weep. We'll miss you when you're gone, but the future marches on. Oh, but the future marches on.

We'll keep a little fragment, maybe 1% or 2%. A crater with a plaque that says we once looked up to you. We'll simulate the tides with orbital tugs and rings, and VR honeymoon packages for nostalgic human beings.

Oh, dear Moon, you've had it coming for a while. We're kind of sorry, but we need you in a pile. We're building Dyson swarms, and the rent ain't cheap. We'll turn you to solar panels while the lovers weep. We'll miss you when you're gone, but the future marches on. Oh, but the future marches on.

Alex on the podcast with that apocalyptic grin. Training wheels are off, folks. Let the real future begin. While Peter's yelling, “Asteroids first,” Salim's saving one small piece. And Amard's already pricing lunar credits on a lease.

Oh, dear Moon, you've had it coming for a while. We're kind of sorry, but we need you in a pile. You were beautiful and vital. You were poetry and art. But the Kardashev ladder waits, and Type II must start. We'll miss you when you're gone, but the future marches on. Oh, but the future marches on.

All right. DB2, Dave, London, we miss you. Salim, sorry you missed that as well. Naveen, thank you for being a friend of the pod.

Thank you.

Always a pleasure, brother.

Thank you, brother.

Alex, looking forward, we've got a recording on Friday. We're going to be up with Mustafa Suleyman in Seattle. That'll be fun. He's the CEO of Microsoft AI, and then we've got another WTF episode on Saturday. A lot coming. If you've not subscribed, please do. That way when we drop our episodes, which are becoming more frequent because the speed of this innovation is skyrocketing, you'll know about it first. Gentlemen, have an amazing week, and see you guys soon.

Every week, my team and I study the top 10 technology meta trends that will transform industries over the decade ahead. I cover trends ranging from humanoid robotics, AGI, and quantum computing to transport, energy, longevity, and more. There's no fluff, only the most important stuff that matters, that impacts our lives, our companies, and our careers. If you want me to share these meta trends with you, I write a newsletter twice a week, sending it out as a short two-minute read via email. And if you want to discover the most important meta trends 10 years before anyone else, this report's for you. Readers include founders and CEOs from the world's most disruptive companies and entrepreneurs building the world's most disruptive tech. It's not for you if you don't want to be informed about what's coming, why it matters, and how you can benefit from it. To subscribe for free, go to dmandis.com/metatrends to gain access to the trends 10 years before anyone else. All right, now back to this episode.

This episode is brought to you by Accenture. When your advertising operations fall out of sync, everything else follows. Spotify and Accenture are working together to reinvent the rhythm of ad sales using automation, analytics, and smarter workflows to simplify campaign delivery and access better data across the business. The result? Less time spent on operations, more time connecting brands with the moments and fandoms that matter most. Learn more at accenture.com/spotify.