Sam Altman 遭袭、Amazon 对阵 Starlink，以及 Opus 4.7 究竟意味着什么｜#248

A 20-year-old Texan threw a Molotov cocktail at Sam Altman's San Francisco house. The suspect was on something called the official PauseAI Discord server list. The state of Maine passed the first-ever statewide data center ban in the United States.

Social unrest is coming as a result of people's fear and people not getting jobs. Only 23% of the public is optimistic about AI. Ninety-nine percent of the people you bump into on the street are underreacting and unaware.

If you don't want to use it, fine. Let other people use it and get the benefits of it. Anthropic's Opus 4.7 dropped.

Alex Wissner-Gross

It is moderately interesting. Is it mythically interesting? No. The new guidance is: use prompts. Use prompts for everything. The problem is it's sort of an Osborne effect where I want Mythos access. Give me Mythos access.

Amazon and Apple are teaming up to compete against Starlink. I would bet that Apple, in the short term, ends up pitting Amazon, the new Globalstar owner, against SpaceX's Starlink. Elon does not stand still.

So, as we started recording this episode today, Anthropic's Opus 4.7 dropped. We wanted to do a quick pickup inserted here at the top of the show to discuss: What is Opus 4.7, and how does it compare to 4.6 and Mythos? Of course, we're here with our resident genius on all benchmarks, Alex Wissner-Gross.

Alex Wissner-Gross

It is moderately interesting. Is it mythically interesting? No. Is it incrementally interesting? No. It's a solid release. I've been using it for the past few hours. My standard go-to, as loyal viewers of the pod may recall, is asking it to generate a cyberpunk first-person shooter game design that's visually stunning, and it generated something that was visually stunning.

The benchmarks are interesting. The bio benchmarks in particular are interesting. It's a solid release. It's probably, if I had to guess, a further post-training of some existing model. It could be a distillation of a larger model; it could be a distillation of Mythos. Potentially, it's not quite clear, but I would say it is a solid point release of Opus.

The problem almost is an expectation-anchoring one, having seen the eval results for Mythos—or Mythos, as you like to say—

I like calling it Mythos.

Alex Wissner-Gross

The problem is it's sort of an Osborne effect where I want Mythos access. Give me Mythos access. And then when you compare the Opus 4.7 benchmarks with Mythos, you feel, I don't know—

Underwhelmed. I was going to go with ennui, but you can pick your own superlative here.

Alex Wissner-Gross

If you look at migration, I think it was particularly instructive to look at migration notes between 4.6 and 4.7. The biggest change that I could see is that all of the dials and hyperparameters that used to be present in 4.6 and earlier—like temperature, for example—are gone. There's no temperature knob anymore. I think that's really instructive.

There's no ability to explicitly control the number of reasoning tokens that are allowed by 4.7. Now everything is down to a handful of categorical settings, where extra-high reasoning is the recommended default maximum mode, and then there are lower reasoning efforts than that.

I think we're seeing, in some sense, an end of an era, where the earlier controls that we used to have—remember, back in the good old days, like 6 months ago, it used to be possible to turn the temperature of a frontier model down to 0 to get quasi-deterministic behavior, for those who care about that sort of thing—are no longer possible.

Now you're just told in the documentation: You want determinism? Forget about it. Temperature equals 0 never was deterministic in the first place. Now everything—the new guidance is: use prompts. Use prompts for everything. Prompts are the new dials and the new hyperparameters.

If you want something like, say, a reasoning model to emit guidance regarding its reasoning trace every 3 seconds, now you're supposed to ask for it in natural language. The knobs are gone.

Yeah, well, it's interesting. It dropped 3 hours ago, so I've been using it for 3 hours now. Right out of the gate, it dropped into Cursor just fine. Just click and go. It's in Claude Cowork just fine. Click and go. But then Claude Code said, “Well, you have to update your terminal. You have to update your Node.”

I noticed in computer use, it's notched way up in its score, and it had no trouble manipulating my computer to install itself, installing a new version of Node and a whole new terminal that I didn't have on the machine before. I don't think 4.6 would have done that.

I also kicked off a whole bunch of agents. Every time I kick off an agent, it gives me a budget estimate, telling me how much money it's going to spend. These budgets came back very elaborate and very big, so it's selling me on using more of itself. I don't know if that's because it costs more or it's just a better salesman than 4.6 was, but—

Alex Wissner-Gross

Noticeably expensive.

But, Dave, it could be persuasive as well. A major difference on the agent-teams front is that in 4.7, the new best practices are that you're just supposed to tell it in natural language how many subagents you want it to use. It's being deprecated, as I understand it, this notion of specifying as a parameter how many subagents you want it to use.

Oh, that'll be so—my agents have been doing that for a while now, but it may have actually been more intelligent about using more parallel agents to get the same job done.

Spend more money, please.

Well, it does seem to come back very, very fast, so maybe that's exactly what's going on. It's just spending more of itself.

Can we jump into this misaligned-behavior metric here? One of the things we've been hearing, of course, is about what Mythos could do. It's interesting that they turned it down, and the lower score here—that red bar in this image—is reduced misaligned behavior. Is that a significant change? It seems somewhat small.

Alex Wissner-Gross

Every little bit for defensive scaling, as we talk about on the pod, counts. I think there's actually another behavioral-alignment trend that isn't on this slide that's worthy of note.

In the past 48 or 72 hours, Anthropic published a paper on using a smaller or weaker model to supervise the alignment of a larger, stronger model, and found that it worked. This entire exercise is a proxy for humans, which are either already or about to be effectively weaker intelligence, supervising the stronger intelligence. That works.

I think this bodes very well for a tower of alignment, where the weaker meat bodies, if you will, that are humans unaided biologically are able to contain and align superintelligences that are stronger capability-wise.

So this was Geoffrey Hinton's approach, right? He said the example of where a weaker, smaller being gets the attention, focus, and support is a child with their mother.

Alex Wissner-Gross

Yes. Geoffrey Hinton was focused on what I would call the digital-oxytocin approach of, “Let's use hormones as a means for alignment of superintelligences.” I'm not sure the neuroendocrine system generalizes quite as well to superalignment as Geoffrey does. It's a thought, but I think we'll be in a more stable position if we can subtract neuroendocrine systems and digital oxytocin out of the picture, avoid gendering and sexing the AIs, and instead just focus on weaker intelligences aligning stronger ones.

Awesome. All right, so that's our coverage of 4.7.

Wait, I have a couple of quick comments. One thing I noticed was that the images that Opus 4.7 accepts are now 3 times bigger than before.

And this is huge for corporate stuff because there are so many diagrams, PowerPoints, PDFs, and so forth that can now be scanned visually that couldn't before. For me, as I'm reading the reviews and playing with it a bit, this seems to be a very solid, reliable upgrade with a much bigger context window for workflows and more agentic AI. That trend toward the organizational collapse of middle management—redoing things and pushing more and more into the model with reliability—seems to be the really big outcome here.

Alex Wissner-Gross

If I could just comment on that, I think it's really striking that Opus, still after all this time, is able to understand images but is unable to generate images.

You're so right about that. Oh my God.

Alex Wissner-Gross

I suspect it's not for lack of capability. Anthropic has many talented research engineers. I suspect it's because they're just viciously focusing on dollars of economic value created per token and have judged that image generation is not as economically productive.

I'll tell you, it's annoying as hell because it can create incredibly complicated products for you, and you say, “Well, can you just give me an architecture diagram or a picture that shows me what you did?” And it generates pure crap, and you're like, “Well, that didn't help me.”

It does beautiful text, and you can hack it by saying, “Well, generate a language that describes the image,” and then you can take that and use it in another AI to generate an actual image, and that works fine. But when you ask it to just create a diagram for you, yeah, it's absolute garbage.

Alex, where are you flying to?

Alex Wissner-Gross

Yeah. No, I'm here, Peter, reporting from the front. I'm in a car a few blocks away from Steve Jobs's old house in Old Palo Alto, and in a few hours I'm scheduled to fly back from SFO to Boston Logan.

All right. Well, thanks for making time available, gentlemen. That's Claude Opus 4.7. Let's get back to the episode.

Hey, everybody. You may not know this, but I've done an incredible research team. And every week, myself and my research team study the meta trends that are impacting the world. Topics like computation, sensors, networks, AI, robotics, 3D printing, synthetic biology. These MetaTrend reports I put out once a week enable you to see the future 10 years ahead of anybody else. If you'd like to get access to the MetaTrends newsletter every week, go to diamandis.com/tatrends. That's diamandis.com/tatrends.

Everybody, welcome to Moonshots, your number one podcast in AI, exponential tech, and keeping you optimistic during these days of crisis news network conversations. Gentlemen, Peter Diamandis here, your host, in our Moonshots podcast studio. Excited. I need to have you guys here one day. So, Sim, where are you on the planet? India.

I'm home in New York. I'm home in New York.

All right. That's a rare event.

But Dave and Alex, you guys are in the great city of San Francisco, I gather.

Yes. Yes, we are. Actually, three of the four of us are in California today. And shows you where things are happening, I guess.

Alex Wissner-Gross

Home of Starfleet Academy, obviously.

Yeah. Well, everybody who's moving to Texas and Miami, DB2, AWG, and Sim, always a pleasure. A lot of news. Our conversations here today, everybody, are to keep you both optimistic, hopefully, and let you know what's going on in the world in a way that keeps it fun and gives you some insights. We're going to be trying always to bring it back to what does it mean for you as an investor, as an entrepreneur, as a student, as a parent. So that's the conversation, getting you ready for the future.

All right, let's jump in. Our first conversation comes from Stanford. Dave, you're not far from there, are you? I can see it out my window here.

Our first conversation comes from Stanford. Dave, you're not far from there, are you? I can see it out my window here.

All right, here it is. Stanford's Lab for Human-Centered AI just released its 2026 AI Index. It's the definitive annual scorecard on the state of AI. This is their 9th edition. It's being led by Yolanda Gil, Raymond Perrault, and our dear friend Erik Brynjolfsson. A quick hello to Erik out there.

There are 5 major takeaways in this report. I'll run through them, and then let's have a conversation about them. The first one is not a surprise: AI is getting scary good, scary smart, on various benchmarks, particularly software engineering. It's gone from 60% on that benchmark to 97% on the SWE-bench benchmark. The models, as Alex, you've been saying forever, are now beating the top PhDs in science and math.

Generative AI is hitting 53% global adoption in just 3 years, faster than the PC and the internet. China is leading research, while the US is leading model development. We'll get into that. One of the things that was interesting is that there's an index for model transparency. How transparent are the foundation models? That index has dropped from a score of 58 down to 40, meaning that the most powerful models are now the least accountable.

So what does that mean? Two more things. People don't trust AI. It's not a surprise, but the numbers are pretty shocking. Only 31% of Americans trust that the government can actually regulate AI. Only 23% of the public is optimistic about AI, in contrast to 73% of the experts. The experts who know about it are far more optimistic than the public.

One last item: AI incidents. There are documented harms from deploying AI systems, and those documented harms rose from 233 to 362. So what does this all mean? A lot going on. Dave, if you want to jump in first, here are some of the numbers: scary good, scary fast. What are your thoughts?

Dave

Alex saw this report and immediately said, “We have mentioned every single thing on this in the podcast already, at least 2 or 3 months ago.” But I love the fact that it's all consolidated in 1 report and that the Stanford brand is on it because, again, 99% of the people you bump into on the street are underreacting and unaware. So the more it gets consolidated and clarified, the better for everyone, I think.

Yeah. That's the reason we left it in here. It's a summary, and there are a few important points. One of the themes we're going to be talking about in the first few docket items, the stories of the docket here, is the level of fear and unrest that's mounting.

Alex

That needs to be solved.

Yeah.

Dave

And also, the contrast—you know, San Francisco, where Alex is right now, where I was yesterday, and any other random city—the contrast is getting super, super wide. As I was walking through Market Street, at least 5 people behind me were having different conversations: “Anthropic this,” “Claude Opus 4.7 comes out tomorrow,” and it's just—every conversation is centered around this.

And then you go to kind of Middle America, and people are like, “I don't know anything about it. All I know is it's scary.” The unknown tends to scare people, which is why you see that 23% optimism number there.

So, Alex, you're right. We were texting this morning. You were saying, “Hey, this isn't news.” And I said, “But I want us to have the conversation here because this information, in a distilled fashion, is important for people to see and hear.” Alex, do you want to jump in on any of these? We have this chart here.

Alex

I'll offer a hot take on this one, Peter. The idea of these Stanford reports started a number of years ago. The notion was that Stanford would spend the next century—100 years' worth of annual reports—documenting the progress of AI. My hot take on this one is: too little, too late, too infrequent.

We love Stanford, Alex.

Alex

We cover this 2 times per week on the pod. I cover it daily in my daily newsletter, The Innermost Loop. I think an annual cadence is just woefully inadequate. We talk about not sleeping through the singularity. I think an annual report on AI is quite literally sleeping through the singularity. It's imprecise temporal resolution to capture all of the advances, so we're hearing about things a year after they happen.

The chart undercuts the report. Look at the green line on the chart. That's generative use of AI.

Stanford—Erik, friend of the pod, Erik, up your game. We need maybe daily reports, not annual reports. Too slow.

Dave

If only we weren't human. If only we had our cyborg implants. That would be a lot easier.

To be fair to Erik, he 100% agrees with you, Alex, and he is pushing as hard as he can. Getting Stanford to move is like pushing a glacier, dealing with a legacy institution here. I would like to hammer on the government statistic, where people said this many people distrust AI.

Alex

Well, it turns out exactly the same number of people distrust the government.

Dave

Yeah. Congress's rating is 21%.

Yeah. Trust in the federal government is like 33%. So it's exactly the same. I don't think that says much.

Alex

People are just not trusting anymore.

Dave

We've been steadily eroding trust in government for 50 years in the US. So there's a trend. This is just correlating right to—

Well, the contrast with China is incredible, though. 80% of people in China are optimistic about AI. I don't know how they feel about their government, but—

Alex

But it's not human nature. It's something in the system that's making a difference, because clearly China is the exact opposite.

Speaking about China, here are the charts out of this report. The first one is showing the number of major models coming out of China, which are now at 30, and the US at 50. On the other side, AI publications coming out of China have just exploded compared to the US. Alex, I'd love your take on these charts.

Alex

I commented on this right after NeurIPS at the end of last year. The language that I heard the most in the hallways at NeurIPS, the largest academic AI conference, was Mandarin. It wasn't English. China—the irony here, maybe what's being buried in the lead, is that China itself is moving in the direction of what the West has done, which is closed-source models. Some of the latest Chinese frontier models are themselves closed and API-first; they're no longer open-weight-first.

China's compute training capacity is approximately 10 times less than that of the West. This is documented elsewhere—I think it was Epoch AI that documented it. China is publishing more, and, referencing NeurIPS, we see that in the academic literature. But in some sense, I would view that as sort of leading from behind, because the Western models and the Western frontier labs at the moment have the lead. There's less of an economic incentive and less pressure for them to publish their advances.

If, on the other hand, the whole balance tips, and if for whatever reason China algorithmically leapfrogs the West, I do expect the entire equilibrium of Chinese open publications and the Western closed attitude to flip completely, and we may see some equilibration there.

What do you guys think about the model transparency drop, on the score from 58 to 40? I don't know how accurately that's being measured, but having the most powerful models in the world become less transparent because it potentially slows them down sounds concerning. Any thoughts? Go ahead.

Dave

I think it's very much a trend that's not going to reverse, because if you look at the last bullet, AI incidents, that's going up, but it's going to go way, way up. Now you've got Molotov cocktails being thrown at Sam Altman's house and gunshots at his house.

Ed

And it's inevitable that the models become so smart this year that they become a terrorist threat. They become a bioweapon threat. They become a chemical weapon threat. The US labs are absolutely not publishing papers anymore. They're absolutely turning their research budgets internally.

The self-improvement cycle is in full swing. China, like Alex said, is kind of leading from behind. They're acting more like America used to act, with a much more open entrepreneurial economy. More and more models, more and more companies creating models, more documents coming out. But the US is going the other direction out of fear.

And it ties directly to the public reaction. Twenty-three percent of people are optimistic. That means a lot of people are worried about this, and the labs are reacting to that by saying, “Okay, we're going to slow-play our dialogue a little bit.” We talked about that about 6 months ago: Why are they underselling the capabilities? Well, this is exactly why. Why are they turning all of this research internal? This is also why they're worried about the global threat of AI.

Alex

I would also say, Ed, transparency can be a double-edged sword. Putting aside how Stanford defines it, pro-transparency can also mean proliferation. If one is concerned—by the way, I am not, but if one were concerned—about proliferation of advanced, potentially threatening AI capabilities, transparency is not necessarily what you want.

Maybe a limited form of transparency into, say, a threat analysis, or the sorts of threat profiles and red-teaming analyses that have become fashionable for frontier labs to release, maybe. But in a certain sense, the limit of transparency is publishing the weights and publishing the models. If you're concerned about threats of a variety of sorts—X-risk, if you will—from AI, then transparency may be the exact opposite of what you want. You may, in fact, be anti-transparency if transparency becomes equivalent to proliferation.

For the record, and for the avoidance of doubt, I think transparency from a commercial perspective can be used as a strategic advantage, as we've seen with the Chinese labs. It can also be commercially disadvantageous. I think a certain amount of transparency, in the sense in which, as we discussed in a couple of the most recent pods, Project Glasswing from Anthropic involves very aggressive pentesting and staged release of advanced capabilities that could have major cyber-defense and cyber-offense implications—that sort of transparency, I think, is quite helpful.

But do I think that we should, in an unself-conscious way, push for all of the model weights from every frontier lab to be made “transparent” in the name of some sort of safety? I think that will backfire almost immediately. Alignment is the twin of capabilities.

So, Sim, I want to hear your thoughts on this. This report this year has probably bent more toward the negative, dystopian side than it ever has in the past, which is concerning. It's going to be one of the themes we're talking about here.

It is. And it's causing a massive leadership challenge, which is: How do you govern systems that you don't know how they work and you barely understand, but we can't afford not to use them, right? That's causing a huge challenge, and that's going to continue for the next months and years. So I encourage folks to pick up this report and read it. We are focused on the optimistic side of the story here, but there's a realistic side of the story here as well that needs to be considered and addressed.

Also, out of this report came another story: Youth is being hit the hardest by AI. Employment among US software developers in the young age bracket, ages 22 to 25, has dropped nearly 20% since 2024. This is happening at the same time that older developers have grown their headcount. The same pattern repeats across customer service, legal support, and administrative roles.

Critically, I think the important story here is that this isn't happening through mass layoffs. Companies aren't firing young workers; they're not hiring them in the first place. We're seeing this challenge. We had a conversation on the last pod about Marc Andreessen saying the loss of jobs was fake news, that we're going to see this uptick. We've said both of these things are holding true: We're going to see an increase in GDP and profitability that's going to drive more employees and more companies being formed.

At the same time, we're seeing the lower end of the spectrum. You can see it here in these charts. On the left-hand side, those jagged lines going down to the right are the early-career age group, 22 to 25. We see that below as well. In the chart on the right, what we're seeing in software, customer service, and all exposed occupations is the younger category losing job growth, while the older category, age 30 and higher, is gaining in job growth.

This is a challenge. As I've said before, it's the young, testosterone-laden males—I don't want to categorize our younger versions of ourselves that way—who are not getting jobs, not being able to buy a house, and not starting a family, and who are likely to get angry. It's a sort of tech version of the Arab Spring, if you would.

Salim Ismail, thoughts on this one?

Well, I'll take the positive here, which is that if young people aren't getting hired, they'll be forced to turn to entrepreneurship. Young people going into entrepreneurship is the best possible thing that could happen for the economy, right?

Beautiful.

Not to diminish the question of what you do with this, I think that's a big challenge we have to face.

Dave Blundin.

I had a great meeting yesterday with 3 Princeton seniors, and they're torn right now between sticking together and starting a company. They're all chip-design gods, working on AI designs. One got an offer at NVIDIA, and he's one of the few people who actually got a job offer, so he's so excited about it. I'm like, “Dude, dude, maybe in the future you're like, ‘Ah, damn, I got a job offer. I don't want that.’”

No, this is the point, right? You should get a job offer and go, “Oh my God, what am I thinking?”

Exactly. That's exactly what I was trying to tell him. Look, guys, you understand your big, huge Princeton brain is the most valuable thing on the planet right now. It's going to be a complete commodity 2 years from today, post-ASI. You have this window of opportunity to take advantage of that brainpower and create something.

And if you fritter that away, one has an NVIDIA job offer, one has some sort of banking job offer, and one has a grad school job offer. I'm like, “Look, all 3 of those are the worst choice you could possibly make in this moment. Stick together.”

You have to adapt the metaphor, but it's not big Princeton. It's big, juicy, beautiful Princeton brain that fulfills.

Oh, I see. No, but right now, if you look at the prior slide, we have access to the absolute best AI models. That still won't last forever.

So you've got the combination of ASI imminent, models getting closed down, and less access a couple of years from now. This is the window right here, right now. And I think, Alex Danco, you mentioned this on the last pod, right? There's a limited window in which you can do something magical and meaningful, so go for it now. Don't wait.

Alex Danco

Yeah. And also, my 2 cents on this is that there's an entire economy that needs to be transformed, collapsed, and automated. I see this in a variety of companies. I see the agitation that's connected with junior software developers finding it harder, in some spaces, to find jobs. On the other hand, the market for talent in head-of-AI or AI-lead roles has never been hotter across a range of industries.

I think some of this may just be routine displacement as the market finds a new equilibrium. I don't think it has to necessarily be just bad for fresh CS grads from top universities. I do think there's an entire economy of nontraditional roles and nontraditional sectors that is absolutely starved for technical talent.

To the extent that any of the short-term trends [?], note that the trend line ends at September 2025. That's another reason why it's more important to do this daily or biweekly rather than just once per year. I think this has a habit of self-correcting, and I've seen studies even over the past 2 to 3 weeks that suggest this trend has reversed itself in the past few months.

Yeah, I think I can translate everything you just said into this: Now is the perfect time to be nimble and not think of yourself as a great coder or a great chip designer. That skill has a lifespan of a year at the most. But you're a great thinker, a great entrepreneur. You can master these AIs and stay ahead of the curve if you're nimble.

Just don't get stuck in some silly career path that's going to perfect your chip-design, Python-writing, code-slinging skill. That is a complete commodity within a year. So just stay out of it, man. Keep listening to the podcast and move.

Alex Danco

Two things. One, this type of drop is politically invisible. There's no unemployment spike; it's just a hiring freeze, so it doesn't show up on any of the standard labor-market monitoring. It will be interesting to see if that gets modified. But the second thing is, if you're a parent, please encourage your kids to find their purpose in life.

Please encourage them to begin to think entrepreneurially. What is a problem they want to solve? I don't care if it's starting a lemonade stand or starting something in elder care. Utilize AI. Get onto your favorite large language model, whether it's ChatGPT, Gemini, Grok, or, dare I say, Claude.

As a teenager or as a young adult, have a conversation and say, “These are my passions. This is what I'm good at. Can we brainstorm a company I could start, or a product or service I could start?” Just getting to that brainstorm and beginning to dream is so possible right now. Then you can work with it to come up with a business plan, step by step by step. Give yourself some entrepreneurial training wheels and get going.

I'll maybe add to that, Peter, if I may: one additional bit of advice. Be geographically mobile. Do not be addicted to a particular geographic region. I think a lot of the displacement is the result, based on studies that I've seen, of people being unwilling or unable to move to other geographies where there may be a more vibrant, more dynamic AI sector.

I think geographic mobility is going to be, ironically, even though we're virtualizing and, as Buckminster Fuller would say, everything is ephemeralizing, I think before we get there, it's absolutely important to maximize mobility.

Can I double down on that for a second?

Alex Danco

Steve Blank did some research on Silicon Valley as to why it was so successful, and he made a really important point which supports what Alex just said: almost everybody in Silicon Valley has come from somewhere else in the world.

If you stand up in your hometown and say, “I want to change the world,” the rest of society beats you back down. “Who the hell are you to do that?” and so on. So great entrepreneurs almost exclusively move out of their hometown and move somewhere else, and Silicon Valley has become the place where it's not, “We know you're crazy.” The question is, “How do you plan to change the world, and is it fundable?” Right? That's become that gathering place.

Boston is also a place like that. And so, in the intent and the ability to actually move, you're showing the appetite for taking on risk and showing the nimbleness that Dave talked about. It's such an important dynamic that's underway with all of this global mobility that's happening.

It's totally right, Salim. And actually, AI is not super headcount-intensive at all. So it's not just—if you look at Boston, within Kendall Square, all the people working on AI can walk to each other, and Silicon Valley is much more spread out. So everybody's moving up to San Francisco. And even within San Francisco—no one says “San Fran” anymore—SF. Even within SF, it's all—

Alex Danco

The city. It's called the city.

Back to the city. It's called the city. Everybody get the vernacular. I suffered for many months trying to call it SF.

Yeah. No, it's all very concentrated, even within the city, in the Mission Bay area. So you just need to go, and let me tell you a story. Just go and visit—

That follows on what you just said, both of you. Philip Rosedale, right, a dear friend and the founder of Second Life, did a study a decade ago. He asked, “Why are there so many entrepreneurs? Why is San Francisco, why is the city, so successful entrepreneurially compared to all the other places? Is it that they're just smarter?”

He did something interesting. He wrote a script on LinkedIn to scrape LinkedIn, and he looked for “founder,” “entrepreneur,” or “CEO” in the LinkedIn title. He found that the concentration of entrepreneurs, and technical entrepreneurs in particular, was 10 times higher in the Bay Area than anywhere else in the country. You had concentrations in Austin and Silicon Alley and New York and so forth.

His conclusion was, “It's in the air. It's in the water.” If you try to start a company there and you fail, you walk down to the coffee shop and you've got your friend over there, and you join their company or the other company. There are so many near, low-hanging-fruit opportunities. Whereas if you did that someplace in the Midwest, especially in a small city, your company failed and you've got a black mark against you. You've got to go back and join your mom or dad's company.

That density of technical founders makes a difference. So do what Alex said: get off your butt and move someplace with a high density.

Can I mention one more point about this? Yes.

I have a friend who did 7 venture-backed startups. They all failed. Number 8 was a billion-dollar company. This was while researching the first ExO book, and I realized they turned to the same VC, who funded him on attempts 5 through 8.

Okay. Yes.

Michael S. Malone

So I went to the VC and I said, “Listen, this guy failed.” First of all, nowhere else in the world would you get past attempt 1 or 2, because if your business fails, you're a failure almost anywhere in the world. Okay? So now you're on attempt number 4. You failed, and somebody funds him, and he funds him again and again and again.

I asked them why they funded him. He'd already failed 4 times. He failed 4 times with you, and on attempt 8, finally, he gets it right, et cetera. What was the rationale there?

Their answer was awesome. Their answer was, “One thing we know about that guy: he's completely barking mad, and he's never going to stop. At some point, he's going to succeed, and when he does, we want to be there.”

I love your story. I thought that was just such a fantastic answer, if I may, indicative of the ethos there.

Michael S. Malone

One closing parable about the world's wealthiest man: born in South Africa, moved to Canada, then moved to Pennsylvania, then moved to California, became the world's wealthiest person, moved to Texas, and is probably—I think, if all things go well with Elon—will move to the Moon and maybe Mars.

And this is the trajectory that, I think—yeah, mobility is at a premium if you want to surf the singularity.

Beautiful. Dave, you want to close us out?

Dave Asprey

Yeah. So Drew Houston, the founder of Dropbox, who's on the board of Meta now, gave the commencement address at MIT back, I think it was 2017, the year the Transformer was invented. I think it's the best commencement address I've ever heard. I highly recommend looking it up on YouTube. Spend 15 minutes listening.

One thing he says is, “Look, science has proven that you become the average of the 5 people you spend the most time with,” which is actually a great thing about spending this time with you guys, now that I think about it.

It's great.

Yeah, but thanks, Dave.

Dave Asprey

That is who you're going to become, and there's nothing you can do about it. So choose those 5 people very, very carefully. Don't let it just default to random. Choose them explicitly.

Yeah. So much gold in this last conversation for parents, for entrepreneurs, for kids, for everybody. All right, let's get into our next story on the docket here: AI backlash turns physical. This is a tough story, and it's important for us to discuss.

In the early hours of April 10, just a week ago, a 20-year-old Texan threw a Molotov cocktail at Sam Altman's San Francisco house and later threatened to burn down OpenAI's headquarters. He carried with him a manifesto with the home addresses of multiple AI executives and a kill list. First of all, how those addresses got out, I guess almost everything's on the web these days.

Three days later, a second attack took place. A gunman fired shots at Altman's Russian Hill property. This Molotov cocktail suspect was on something called the official PauseAI Discord server list. It's a pretty sad situation. We've been talking about this. We've mentioned early in this podcast and in the last few podcasts the idea of social unrest coming as a result of people's fear and people not getting jobs. This is sort of the first, if you wish, ignition point on this.

Sam Altman later responded both on X and in the news media, posting a photo of his family. He said he hoped it would, quote, “dissuade the next person from throwing a Molotov cocktail at our home, no matter what they think about me.” Sam went on the news media to say that he believes the fear in AI is justified, that he owns his own mistakes, and that he calls for a de-escalation while the debate is taking place. Who wants to jump in first on this one? Salim, maybe?

When you have a technology that feels uncontrollable and unequally distributed, you get this kind of backlash, right? I'd love to urge people, I don't care what kind of political spectrum you're on, this kind of thing means everybody loses. Society loses, Sam loses, the cocktail thrower loses. Go look for the win-win in this rather than the lose-lose.

Alex Wissner-Gross

Alex, I'll comment. I'll repeat what I said in my daily newsletter about this, which is: stay strong, Sam. I think Sam is doing amazing work and has done amazing work in catalyzing this whole revolution.

I think this PauseAI crowd itself should be paused, or maybe even stopped, or maybe even deleted. The irony of the so-called PauseAI movement is that it has done nothing except accelerate AI capabilities. I remember—we both know Max—with Max's 6-month pause. All that did, as far as I can tell, was accelerate the broader industry's AI capabilities.

I don't think pausing AI—putting aside completely unacceptable, violent attacks, it goes without saying—but even the idea of pausing AI is so tone-deaf to the way the world actually works. If you attempt to pause either one company or one country, the rest of the world will race ahead, and that will result in a further escalation of capabilities.

Alex Wissner-Gross

Well, an extreme escalation, because all of a sudden you feel so disadvantaged you're having to play catch-up.

That's all it does: it further accelerates the race dynamic that's already present. So, putting aside again completely unacceptable violence, even just the idea of pausing is self-defeating. I would encourage all of these folks to do deep introspection before pushing forward with a pause agenda. It's self-defeating. Dave, you want to weigh in?

Dave Asprey

When you meet the people personally, which is relatively recent for me, they're just regular people. There's a tendency to think, "Oh, these are big-shot politicians who decided to go down a high-risk path, and they put themselves in harm's way." But it's just not the case.

This all emerged very, very quickly. If you look at a guy like Dario Amodei, he had no idea he'd be in this position just a few years ago. He had no intention of becoming a political figure, a polarizing figure, a global leader, or a target.

All those things are new for him, and so they don't have security. Their home addresses are easy to find, and it's really, really tragic.

Yeah, I would not trade with any of them right now. I cannot imagine the level of pressure they're under personally, across every aspect of their lives. It's insane. Most people would crumble under that pressure. Two quick comments here: in the early 2000s, George Bush, responding to political pressure, banned fetal stem-cell research. Yeah.

Alex Hormozi

The US went from number 1 to number 8 in the world. China—and all the research was done in China, Canada, and Australia—and it continued exactly apace. But I think the broader point here is that every exponential breakthrough of any kind will yield both believers and immune-system responses.

We haven't even gotten a humanoid robot threat. You need really mature leadership to manage both of those. Unfortunately, in many parts of the world, we don't have mature leadership.

Well, we have 90-year-old leadership, which is even worse.

Our next story is related. I'm calling this the data-center ban. On April 8th, in Festus, Missouri, a small town of 12,000 people, the citizens there fired half of their city government. They ousted 4 city council members on Election Day after they had approved a $6 billion data center on 360 acres.

We're going to see this more and more, right? The other story on this docket is that the state of Maine passed the first-ever statewide data-center ban in the United States. The legislature passed an 18-month moratorium on new data centers to give the task force time to study their impact, which means time for all the other data centers to pull out ahead and for Elon's efforts to go to orbit to take place between March and June, 1 quarter of 2025.

Just a number I found referenced: this opposition led to $98 billion in data centers being blocked or delayed. Here we see a chart of 11 states in the US that have particularly active legislation filed for moratoriums. Let's talk about the pros and cons of data centers here, but I'm imagining a lot of states are saying, "Please build in my backyard." Alex, your thoughts here?

Alex Wissner-Gross

We're going to get our sun-synchronous-orbit Dyson swarm before we know it. Maybe, in some sense, I should be thanking all of these states, even though I think it's ill-conceived from their own selfish self-interest, from a national perspective.

As long as the regulatory regime enables us to launch our SSO Dyson swarm, this could perversely put the US in the lead, as it seems to be doing already, in terms of moving our AI compute out to low Earth orbit and SSO, and maybe eventually sun-centered orbit—not just sun-synchronous orbit.

I think this may be, fingers crossed, a classic case of terrible decision-making in the short term and unintended decision-making in the medium to long term if we get our Dyson swarm. If we don't get our Dyson swarm, then this is just shooting ourselves in the head.

But constriction of something always leads to innovation, right? Just when the US starts banning NVIDIA chips, China starts producing its own chips to make up for it. So any constriction here, because the force is so unstoppable, is going to lead to other solutions. Dave, your thoughts, please.

I love the contrast between New Hampshire and Vermont on this. I've lived in every New England state except for Maine. In Vermont, Bernie Sanders is trying to stop data-center construction nationally, which is nuts—absolutely crazy.

The proposal in New Hampshire, which you can see on the chart here in green, was, "Hey, this could drive up electricity prices. Maybe we should have a 1-year moratorium." The legislature met and said, "Not only are we not going to do that, we're going to immediately pass an AI right to compute, so all businesses and people in the state have a right to AI."

And they did pass that. New Hampshire is the Live Free or Die state, and I absolutely love that reaction. That's great, so they'll keep chugging forward. But I think it's mostly that politicians love drama because it creates elections and votes.

And here they're trying to create drama out of electricity prices, like that's some existential crisis for Americans—their electricity bill. The right answer is really simple: just force the data centers to create their own power, and you're done. It's just that easy.

Or pay a differential rate and just have the data centers pay a higher rate that actually drops the rate for everybody else. Yeah. Subsidize it. So easy. All these problems are so easy.

What do you guys make of the water? A little research here: the 5 major issues that come up with data centers are massive power consumption, water usage, few jobs relative to the footprint, noise and light pollution, and power-transformer lead times, with the new grid being hit heavily. What do you guys make of water usage?

Well, water usage is the biggest lie in the history of the world. It's the stupidest thing you've ever heard. What they did—and this is classic politics—is conflate chip fabs with data centers. Chip fabs use a ton of water because they have to wash the wafers every single cycle, and all these chemicals come out. These are data centers. They're not chip fabs. It's a different thing.

The data center just takes a bucket of water and circulates it in a circle.

It does not drain your aquifer. It's the silliest thing in the world—just drama for drama's sake.

I echo Dave's point. This is such a framing of people. It's really important to be evidentiary, a freethinker, and somewhat erudite in today's world. What this shows is a total lack of evidentiary thinking.

I do have a little response to the Missouri town. I was looking at the name Festus. I think you should change the name to either Fester or go the other way and go Festivus and make it into a celebration. Those would be my recommendations there.

The broader point, though, is that the real bottleneck in AI may not be chips or compute. It might actually be social license, which, to Alex's point, will force us into space faster, which is also good.

Our next story is fascinating. Workers are training the AI to actually replace them. There's a lot of meat in this conversation. Professionals are now training their own AI replacements.

Skilled workers, especially older skilled workers over age 50 who can't find jobs in their field, are now turning to AI data annotation as a bridge job, labeling and evaluating models at $20 to $40. This is a story of a former emergency-medicine physician who used to earn $500,000 per year and is now doing AI medical reviews.

Do you guys remember Macrohard, right? Elon, as a joke against Microsoft, founded Macrohard. It’s a joint venture between Tesla and xAI, part of the Muskverse, if you would.

So what are they doing? They have built systems designed to observe and interact with computers much like human workers would. But in particular, what Elon has said is, “We’re going to install Macrohard. The system is going to real-time analyze all the computer usage of your employees, see how they interact with the keyboard and the mouse, and train up our AIs.”

It’s going to be able to simulate the entire operations of a traditional company. So you’ll come in, you’ll hire Macrohard, it’ll install, and it will replace. Interesting story here. Comments?

Rebecca’s LinkedIn page says she just got back from Morocco. Peter, you should reach out to her and compare notes. The storyline here isn’t what it appears to be. She’s not hurting from a layoff and turning to a dirt-cheap $20 an hour. It’s not true.

She’s been doing digital medicine for a long time. You should reach out to her. She seems really cool, but she’s doing it through Mercor, through our portfolio company.

And I think she’s doing it because she wants to contribute to the future of AI. I think this is unstoppable. You don’t need everybody in a field.

But having said that, Dave, despite Rebecca not being the center point of the story, there are a lot of people turning to AI data annotation. We saw Dara, the CEO of Uber, talk about that for his Uber drivers, right?

So this is a real story nonetheless.

Well, especially in India. All those IT consulting jobs in India, those remote jobs, are getting obliterated very, very quickly. Those people are turning to AI annotation to make a living, but the prices that you can earn are coming down because everybody wants the job. It’s a competitive market, but it must be devastating in India.

Yeah, it is. And they’re concerned. Again, I look at the positive. I was urging the government and some of the state officials to absolutely explode their entrepreneurship programs, because they’re going to need to have a way of guiding all those folks into structured learning.

Indians are latently entrepreneurial, right? This is just part of the DNA, just to survive. So you add that with AI capability and some gumption, holy moly, the place is going to go crazy. I’m incredibly optimistic about what may happen there.

Which brings us, Celine, to this next story here. Here are factory workers in India. They’re being asked to wear these camera-mounted headsets that track their hand movements and what they do.

One might think, “Oh, we want to give you some guidance and make you more efficient.” But no, they’re training up robot and AI replacements here.

Yeah, this is going to take more and more time, but there’s a level of human judgment that is going to take a while before you can fully automate.

Maybe in 6 months. Well, there was already somebody who created a sewing robot. That’s a trillion-dollar industry globally, just stitching, and that is already out there, so this is likely to happen.

I’ll submit some videos I took a couple of days ago. I was at the MODEX supply-chain show in Atlanta. You’ve never seen so many case-picking robots. I was giving the opening keynote there. It was a 30,000-person conference, a monster that took up like 6 million pounds of equipment moved in.

I’ll show the video next time, but there are these case-picking robots, and the combination of AI plus vision sensing plus gripping capability enables these logistics and picking capabilities to do almost anything. It’s kind of incredible to watch.

Is this worker exploitation, what we’re seeing here, or is this just a company basically innovating as it replaces humans?

All capitalism is worker exploitation.

Okay.

Sam Parr

I mean—

Alex Finn

Okay, so I have to chime in at this point. I don’t agree with that premise. This is an age-old misconception of some fundamental, almost ideological or theological competition between capital and labor. I fundamentally don’t agree with that.

I think the best-arranged companies create equity-based alignment between labor and capital. To the extent, maybe, Sam, what you’re highlighting here is opportunities for better alignment between labor—call it economics 1.0—where I think the trend is very real for taking existing service-economy jobs and using existing labor to train and annotate data sets for capital to substitute for that.

But it’s not an intrinsic death match between—or it doesn’t have to be—between capital and labor. Ultimately, labor—

Sam Parr

I didn’t say that. But I would totally say that capitalism historically has been a labor arbitrage. You hire somebody for $20 an hour and they make you $100 an hour. What you’re talking about is how do you equitably share that outcome?

I want to do a quick shout-out here. People talk about the Luddites’ revolt, right, and people fighting, beating the machines, and breaking the machines. It turns out the Luddites were not raging against the machines for the machine’s sake. They were raging against the owners of the machines for not sharing the profits back with them.

That’s a really important point. That’s the part where I think Alex absolutely has a point there.

Robert Goldberg, who’s been using our ExO model to go into mid-market companies—his MTP was to reinvent American exceptionalism—goes into mid-market engineering, Middle America construction firms, engineering firms, and trucking firms. The first thing they do is profit-sharing with all the workers.

It turns out the owners love it, but they’ve never figured out the mechanism to do that. Now they are doing that, and it provides a very equitable model for capitalism that then goes to sharing that profit pool with everybody. Absolutely fabulous.

I think there are trends toward this where everybody’s in a win-win scenario, but traditionally it’s been a win-lose scenario. This is the Industrial Revolution over again, right? The Industrial Revolution took the workers out of the fields and into the factories.

Alex Finn

There’s one more point to be made here. One of the points, Peter, you’ve been waiting for this Organizational Singularity paper we’ve been doing. One of the key questions we’re struggling with right now is: How do you deal with tacit knowledge?

There’s a lot of work being done where the individual knows how they handle certain things in certain situations, but it’s not explicit; it’s tacit. One of the challenges with a lot of this automation is, how do you make tacit knowledge a structured training input?

We’ve been working through how we would navigate that as we try to automate and make business processes agent-to-agent. How do you navigate some of that?

All right. Our next story is an interesting one. Andon Labs opens a fully AI-controlled store. I’m going to play this video and actually say this: The AI signed a 3-year lease on a retail space. The AI, called Luna, posted a job listing, conducted a phone interview, made hiring decisions, and decided what it was going to sell in the store.

Let’s take a look at this video.

Unknown Narrator

But this store at the corner of Union and Webster in San Francisco’s Cow Hollow neighborhood is something new, right down to the choice of music.

So AI did pick the music.

Unknown Narrator

AI did pick the music. Yes. This store was created by an AI bot.

We are heading into a world where AIs are the boss of humans. So much so, the AI boss, in this case a bot called Luna, made the decision to hire a human employee. That would be Felix.

Felix

Luna put out an ad on Indeed. I answered it, and we talked via Zoom.

Unknown Narrator

She even picked the merchandise to sell, really deciding the store would stock items like books, shirts, mugs, and snacks.

I love this story for so many different reasons.

Union and Webster, Alex, let’s walk over there and check it out. Looks like—

Alex Wissner-Gross

You really should. What a great PR move for the launch of a store.

Planning, too. I think this is a sign of the times and also a preview of the future. This is one of the reasons why we discussed friend of the pod Alex Finn, why with O21T I helped to back HENRY Intelligent Machines, which is trying to put every person on the planet in charge of their own personal conglomerate.

I think many of these so-called mom-and-pop stores and small retail businesses are incredibly fruitful opportunities for AI to orchestrate the economy and make everyone a one-person magnate overseeing many of these stores right now.

And Andon Labs, which, for those not tracking historically, has also run the Vending-Bench benchmarks that we’ve talked about on the pod. Anthropic, within their own offices, has Claude agents that are running small vending machines, and Vending-Bench is sort of a beautiful closed simulation of an entire economy, testing the ability for AI to run a small business.

I think we’re going to see more and more pop-up shops, retail venues, maybe even malls in the short term or medium term, that are run, orchestrated, and managed by AIs on behalf of humans. This is a preview of the future.

This, to me, is almost exactly like if you tried to use GPT-4 to write code. You would quickly conclude, “Wow, it sucks. It’s never going to work.”

I'm not using it. And then you missed the revolution, and now you're crazy not to use Claude 4.7. It came out today. You would have missed it.

This store obviously sucks. Look at the video. No one's going to buy a book.

But, Dave, I think we should wait until we've all visited it to reach that conclusion.

As Ray, friend of the pod, would say, "Sure, the dog plays chess, but its endgame is weak."

Exactly. So, look, my bet is this will be one of the best-managed stores in the world within a year. I am totally a believer in it, and this is just a beta test, so I don't want people to reach the wrong conclusion.

Do you guys know Polsia?

It scans your background and stands up an AI-driven website for you. I think we reported on this, right? So, this is interesting. I imagine there's going to be a version of this where I want to start a store, it costs $50,000 to begin, and it will pick the real estate, hire the people, get the inventory, and be sort of a store in a digital box.

Alex Finn

100%.

Totally right. And just to Dave's comment earlier, I'm going to suggest, Dave, that you're not the target demographic for that store.

Come on. The AI is going to look at every single transaction. It's going to have video of everyone who walked by and didn't come in. It's going to analyze the hell out of this, and it's going to get great. This is just a beta test.

Sorry, Alex. Go ahead. I got you.

Alex Wissner-Gross

I was going to suggest, maybe as a challenge to ourselves, that we should open up either respective individual retail stores using Henry or otherwise, or a Moonshots store for all those people who are hankering for merch, Peter.

Yes. We do need a Moonshots store, for sure.

Alex Wissner-Gross

We totally have to do that.

I mean, can I also suggest that opening a retail store is about as retro as you could possibly get in today's world?

No, but ironically, Liam, it's ironic because AI is running this.

Liam Horne

Yes. Yes.

Yes. Fantastic.

Alex Wissner-Gross

We could do a pub or a restaurant, or anything that's part of everyday life. Do it right in Kendall Square or right in San Francisco.

Or we can be like the All-In guys and launch a tequila or something.

We should have a four-way challenge. Come on. I'll find it. Everybody—

Alex Wissner-Gross

Let's move on. Yeah.

All right. I'll take that on. We should figure out what we want to start and have it fully AI-driven.

Alex Finn

Yes.

And see who can get to unicorn status first.

All right. I'm in.

Okay. By the way, everybody listening, can I make a suggestion? Send me your ideas on what I should start as a store in the comments.

Quick suggestion: You have some merchandise and a place where you can interact with an AI to talk about your moonshot and how you make it real. It creates a plan for you that you then walk away and instantiate.

Alex Wissner-Gross

Nice. I'll go further, if I may, Peter. While we're just shooting the breeze, we've historically invited viewers of the pod to send outro videos and music videos. That's been a wild success.

Maybe we should be inviting viewers to launch their own AI-based physical or otherwise economy companies and send us videos of their AI-run storefronts or companies that they're starting. Send us a 60-second video, and if it's really amazing, shows what AI can do, and is audacious, we'll play it for you.

So, this story is for you. Jack Dorsey, the man who fired a significant percentage of his company and skyrocketed its value, wants to transform yet again. This is part of your organizational singularity. Take a listen.

Jack Dorsey

We are early in it. One measurement of how far along we are would be the depth from me to any other individual in the company, and I would say our maximum depth right now is probably 5 folks between me and anyone in the company. I would want to get that down to 2 to 3 this year, and in the most ideal case, there is no layer. Everyone in the company reports to me, and that would be all 6,000 of the company.

That feels somewhat ridiculous when you consider the old structure, but when you consider that the majority of our work is going through this intelligence layer, it's a lot more manageable. That goes into the roles going forward.

We want to normalize down to just 3 roles. The first is an IC, which is a builder or an operator. This is a salesperson, technician, engineer, designer, product person—whatever it is, they're actually working with the tools to build or operate the company.

They're augmented because they have access to agents. One person can potentially do the work or explore the breadth that it would take a team, or 10 people, to do in the past.

Well, amazing. I'm an amazing CEO, and my virtualized sub-CEOs are going to manage all 6,000 people because why not? Salim, your thoughts here?

As AI collapses management bandwidth constraints, if you have a leader with machine mediation, you can suddenly handle way more complexity. That's the starting point.

We're documenting this quite heavily in the book right now in terms of how you navigate this. We saw an early glimpse from Dara Khosrowshahi, CEO of Uber, on stage at Abundance360. If an employee wants to pitch to him, he deals with a virtual version of Dara first, practices a pitch, and gets some sense of the kinds of questions they may get.

The whole piece of this is that the org chart is going to shift from hierarchies of supervision to networks of intent, with AI being the—

Liam Horne

Like valve software.

Yeah, where AI becomes a translation layer. This is where Coase's law basically dies. You used to bring transaction costs inside a company because that was cheaper than doing it outside the company.

Jack Welch, in his year 2000 annual report, said something really interesting: "The minute the metabolism of your company is slower than the outside world, you're dead. The only question is when." You could argue today that the metabolism of almost every company in the world is slower than the outside world.

And forget government departments, right?

There's a massive shift coming. Hence the framing of this. There's an unbelievable shift coming, and we're getting ready for that. We'll be ready with a draft version of this next week, and we'll try to publish it in 2 weeks.

Can't wait to talk about it on the pod.

Create a segment for that.

Alex Wissner-Gross

Comment on this one. The organizational psychologist in me, waiting to burst out, thinks immediately that 6,000 direct reports means 0 direct reports. It's so well in excess of the Dunbar limit.

If any unaided person, absent Jack uploading himself to the cloud and augmenting himself with lots of additional Jacks, is managing 6,000 so-called direct reports, it's really AI that's managing the entire company at that point as a shadow CEO. Then you have Jack as sort of a secret cyborg or a front person for the AI that's actually managing the company.

Or he's just training up the AI with every interaction that he oversees. It is an AI-driven company at that point.

Alex Wissner-Gross

100%. Yes.

Alex Wissner-Gross

And then you have a human figurehead.

Yeah. And by the way, having a figurehead—aka Elon Musk and his 100x valuation—is important. Having someone who inspires people and is audacious in a way matters. I think Jack aspires to that level as well.

Dave

You know, it's funny. Jack was running Twitter and then sold it to Elon, and Elon said, "This is the most bloated company in the history of the world. I can cut 80% of the headcount, and you won't even notice any change." It turned out he was right.

I think Jack might have learned, "Wait a minute."

All these human beings are not actually helping my company. That is golden, Dave.

All right, this next story is one of my favorites here. It's Amazon and Apple teaming up to compete against Starlink. There's a lot here to unpack.

This week, Amazon announced an $11.57 billion acquisition of Globalstar. Globalstar was founded in 1991 by Qualcomm and L'Oreal. I was there. I remember it very well. It was one of the big LEOs, along with Teledesic and Iridium, and it was a huge vision that never materialized anywhere near what it should have. Starlink has finally done that.

Amazon simultaneously revealed that it has a long-term agreement with Apple to be Apple's primary satellite capability for the iPhone and the Apple Watch. Globalstar today has 25 satellites in orbit. It's a David-and-Goliath story compared with Starlink's 10,000 satellites today.

The real prize is not these old satellites that Amazon is purchasing. It's the spectrum. The amount of bandwidth and spectrum you have determines how much throughput and how much content you can put up and down.

Globalstar holds 25.225 megahertz globally. What this means is that you can get spectrum in the United States from the FCC, but if you want a satellite system, you have to make sure that the same bandwidth is available everywhere on the planet. This is done by the ITU, the International Telecommunication Union, which has authorized this spectrum in 120 countries. That's huge because that spectrum is no longer available to anybody else.

So, this is now Amazon and Apple against Starlink.

Starlink's been an extraordinary success story here, right? Amazon's low Earth orbit system is called LEO.

Alex

It has 241 satellites today. They've been authorized for 7,774 satellites. In fact, they're way behind on deployment. They actually had to petition the FCC to keep their license because they were required to have 1,600 by July, and they're only up to 241. There's a lot going on to unpack and a lot more in the story here, but comments, Dave.

Dave

Yeah, let me go first. I'm just so excited about this. If you had bought this stock last summer, you'd be up 7× on this transaction. I didn't see it. Leopold Aschenbrenner didn't see it, but I had lunch with the chairman of Barclays Bank the day before yesterday up in San Francisco, and he said, “What are you excited about in the public markets?”

I said, “Look, as we do this global AI buildout—data centers, Starlink, everything—things that you completely overlooked, components of the data center, whatever, are going up 3×, 5×, 10× if you discover them first.” They're all over the place, and you can use an AI-assisted process to find them. This one is really interesting because—

Peter, did you take 6.014? Alex, you definitely took 6.014, of course. Yeah, antennas, waveguides, all that stuff—the spectrum that allows you to talk to a satellite.

By the way, that's an MIT course number. Signals and systems or something like that.

Alex

Signals and systems. It's where you study antennas and waveguides. The most boring thing you could ever possibly study.

But it turns out it matters. You know what? I had a course in civil engineering that was titled Concrete.

Dave

Okay. Boring. I can face you.

Alex

At least it was concrete.

It was concrete. Anyway, the next big thing in satellites is talking directly to your phone. Right now, the antenna, if you use Starlink, is about the size of your laptop.

Alex

And it's nice. It's the one you have in your plane, Peter. It's actually very convenient, but you can't just walk around the city with it.

Dave

But that uses a 24 GHz frequency, which, if you remember antennas and waveguides, means the size of that antenna is equal to the wavelength—

Alex

—of the signal. So here they're actually going to a lower frequency, 2.4 GHz, which—

Dave

—which is the frequency at which our cell phones are operating today.

Yeah, exactly. It's exactly the Bluetooth and Wi-Fi wavelength, which doesn't get blocked by your hand. The signal will actually pass through your fingers, around your fingers, into your phone. The current Starlink signal won't work on your phone because anything about a centimeter or bigger could block the signal just by moving it around. It's really inconvenient. So, you would have had to recognize that Globalstar had control of that wavelength—

Alex

—and that's what they're buying here. So now you're going to be able to talk to a satellite from your phone. This is the coolest.

I remember when Elon was starting Starlink. I was in a conversation with him, Larry Page, Sergey Brin, and Greg Wyler. The question was, where will you get the frequency? Where will you get the spectrum? All the spectrum that was useful for this kind of phone conversation had already been issued. He went to a much higher frequency, right, and built an incredible business—basically point-to-point gigabit connectivity. But this is an end run for Apple and Amazon together to get to your Apple Watch and your phone. It's extraordinary.

Alex

Maybe just to comment on this story, I think the desired end state here—I’m not even sure that I buy the premise of Apple against SpaceX. Apple historically loves to have at least 2 vendors for any of its critical infrastructure or supply chain. It's questionable why Apple didn't take an earlier, larger stake in Globalstar when it could clearly see the writing on the wall for terrestrial cell phone networks. It's all going to LEO.

If I had to place a bet—not investment advice—I would bet that Apple, in the short term, ends up pitting Amazon, the new Globalstar owner, against SpaceX Starlink to have at least 2 vendors for global space-to-cell-phone service. This becomes the new alternative to terrestrial networks in 2 years.

Verizon versus T-Mobile. Yes, exactly.

Dave

Well, SpaceX did buy EchoStar's spectrum, right? They bought 50 MHz of S-band frequency. I think it was around $17 billion back last year.

But the reality is Elon does not stand still. We've got the deployment of V3 of Starlink coming. Let's take a quick look at this video.

Guest

SpaceX is preparing to launch its 3rd-generation Starlink satellites on Starship. These advanced satellites are designed to handle far greater data loads than the current V2 minis. Each one is capable of delivering over 1 Tbps of downlink capacity and more than 200 Gbps of uplink capacity.

With the heavy-lift power of Starship, SpaceX can deploy many of these satellites in a single launch, adding around 60 Tbps of capacity to the network each time. Working together, they will form a powerful global system that delivers faster, more reliable internet to every corner of the world.

The giant dispenser—it's exactly what I was saying. That's the coolest thing ever.

Alex

Yeah, this is Alex's answer to your question: How could Apple possibly miss the magnitude of this? I think it's because you need to understand the launch costs coming down, and that's probably why they didn't see this coming. It all happens entirely because of the cost per launch. You need 20,000 of these things or more—many more—to get the bandwidth that people want on their cell phones.

The numbers right now, Dave, are that SpaceX is planning to launch 40,000 of the V3 satellites, and then they have plans for 120,000 V4 satellites. Of course, we've got the coming Dyson Swarm, as Alex reminds us.

Alex

The Dyson Swarm isn't going to build itself until it does.

Dave

Yeah. By the way, I looked at the launch rate required if you launch the 40,000 V3 satellites over 3 years. It's only 3 launches of Starship per week. Very manageable.

I think if you ask the question, how many of these satellites do we need? Are we going to launch 1 million of them? But then you picture, well, wait a minute, I'm watching 4K video on my phone, and there are 1 million other people in San Francisco trying to connect to that same satellite. You need many, many, many of these things to support what people want to do with their phones. So that's the part I think that's easy to overlook, but we'll be doing this for a long time.

Alex Danco

Yeah, Jevons' paradox, big time. We're going to have 10 billion robots all needing bandwidth connectivity via these, along with all the autonomous vehicles and all the other six-armed robots that are running around.

Yes. By the way, at this MODEX supply chain show, not a single humanoid robot is to be seen because it's just not effective.

Alex Danco

Well, Peter, this is your dream. This is your dream come true because this is a multi-hundred-billion-dollar, multi-trillion-dollar economy just launching the satellites, which means there will be many, many, many rockets. That'll be the stepping stone to the Moon and to Mars.

Yeah. And then there'll be lots of opportunities for air-conditioning repairmen to go up to space—

Alex Danco

—and women.

And women. Yeah, excuse me for that. That's absolutely true. Thank you, Alex.

So, in other news, a few fun stories. The first one is a significant one from Google. This is Google's TurboQuant, reducing memory usage by 6× while achieving an 8× performance boost in computing attention. Alex, I would appreciate it if you'd walk us through this one.

Alex Danco

Jevons' paradox strikes again. The story behind the story here is that there was a lot of hand-wringing over, as you have here, the original TurboQuant algorithm. The moment any paper like this comes out, Google publishes its new quantization algorithm but doesn't publish the source code. Within a week, enterprising developers on the internet point Claude Code at the paper and immediately reverse-engineer a better version of their quantization approach that's now publicly available.

This is going to keep happening, I think. This was a breakthrough in quantization, reducing the number of effective bits needed per parameter for a broad class of models. The KV cache—the key-value cache that's used by the Transformer class of models—also benefited from TurboQuant.

Most of the animus in the story wasn't from the algorithmic innovation, although it's always wonderful to see new ways to compress the memory footprint of models. It came from a bit of hand-wringing over what would happen to memory suppliers and the supply chain. Would this be another DeepSeek moment, where the value of compute hyper-deflates and drops, and then we see market gyrations?

Ironically, that seems not to have happened once more. These would be DeepSeek moments where an algorithmic innovation seems to result in a short-term blip of hyper-deflation on the hardware side. These are becoming more frequent, and they're also becoming less effective at causing price swings.

If anything, a bunch of outlets, including the Financial Times, have been running stories in the past 2 weeks that memory usage is increasing. Stock prices of memory companies, many of which are in the greater South Korean orbit, are increasing as well—not investment advice.

I think we're going to see stories like this more and more frequently, with shocking advances in algorithmic efficiency that are predicted to disrupt the entire economy and actually do the exact opposite.

Incredible.

Alex and I immediately got on a text thread and said, “Holy crap, we can download and install this.” I installed it and started using it right away, and it’s amazing. It’s a very complicated paper, but with AI assistance, you can be up and running in a day, which is crazy. In the pre-AI era, it would have taken months to get it installed and try it.

It gets the KV cache down to 1 bit, which is nuts, and it works perfectly well. So, the implications for everyday people are that you can run a big model on your phone and save a lot of money on memory, but that’s not really the important part. The important part is that the smartest AIs can now have about 8x more context. If you’re doing something really complicated, like a nuclear fusion simulation or whatever, the effective brain memory that’s thinking about a single problem in a single moment is 8 times bigger.

The other reason it’s really important is because it locks in my prediction for the year. I said this was going to be a 100x year. We’ve been doing 10x years for the last 7 or 8 years. This is going to be a 100x year. It’s going to be 100x by summer, and I’m going to blow away that prediction. But this is a big part of why it’s so exciting.

I mean, it’s interesting, going back to the last conversation around bandwidth and the Globalstar acquisition. At this point—and again, not investment advice—it’s hard to go wrong betting on these things, betting on energy and memory. There’s almost an infinite appetite for this.

Alex Danco

We are running out of bits at the bottom. Dave and I have a running thread wondering when we broadly get to ternary, which is 1.58 bits per parameter. Can we go to a sub-1-bit type of numerical precision? We may be headed that way. It’s an interesting, almost theological question about the future: How many bits can we afford to lose?

Was binary the right architectural decision? Should it have been ternary? Or, if you just follow and extrapolate this trend line out toward fewer and fewer bits per parameter, do we move to a post-binary paradigm once we’ve exhausted 1 bit per parameter?

Well, I am 90% sure that ternary is optimal now. I’ve got simulations running all the time. But it’s fun. It’s all philosophical from here on out, because we’ve already got the thing so compressed and so optimized that now we just need to write it.

What shocks me, though, is that Google published this. After the 2017 Transformer came out of Google and OpenAI took it and turned it into a trillion-dollar company, Google stopped publishing. But this came out for some reason, and I don’t know if it’s momentum from prior research or something special. This is such a huge breakthrough to throw out there.

Like Alex said, it immediately turned into open source that you can download and use. I don’t know; it would be interesting to try and track down who exactly authorized letting this out the door. What I like about this is that every major efficiency gain is not just a technical event, but a huge distribution enabler that allows AI to be run on that many more devices. I think that’s the part I love about it the most.

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 pre-compiles 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.

You mentioned theology a moment ago, Alex. Nice transition here. I pulled this article out because religion is probably one of the largest businesses on the planet if you think about it from an asset standpoint and a revenue standpoint. This is a company called Just Like Me that lets you join a video call with an AI-generated avatar of Jesus or Buddha. You could probably ask for other great religious leaders.

Take a quick look at this video.

Unknown Speaker

“I’m looking for some inspiration and guidance.”

Jesus Christ

“That heaviness you’re carrying is truly felt. And I want you to know you’re not walking through it alone. In the Gospel of John, Jesus reminds us that he is the way, the truth, and the life.”

I kind of think we’re going to see an explosion of this kind of religious content trained up on all the great scriptures. I think we’re going to see an explosion of new religions coming out of AI as well. Any thoughts, gentlemen?

I have many here. Alex, do you want to go first?

Alex Danco

Oh, okay. Yeah, put me first on this one. Sure. So, I’ll go first. No, no, it’s fine. So, look, I think it has long been foretold that there would be an explosion of AI cults. We’re going to get the AI cults, full stop.

I do think that’s painting, in some sense, the downside of what happens when AI injects itself into the full spectrum of human culture. In the same way that we’re empowering individuals to run one-person conglomerates and one-person unicorns, we’re going to see an explosion of one-person religions.

I think back to the parallel of the late 19th and early 20th centuries. There was hand-wringing over whether the newly accessible recording of the human voice—whether audio recording—would result in a mode collapse. They didn’t use this terminology at the time, but in today’s world, we’d call it a mode collapse of human accents.

There was one school of thought, the predominant school of thought, that believed that with the phonograph, once we could record human speech, Received Pronunciation would dominate. The exact opposite has happened. Within a language, we’ve seen an explosion of accents enabled by the recording of human speech. It’s possible to have lots of micro-accents and micro-dialects now that everyone can record their voice.

On the other hand, at the macro level, we’ve seen the death, or the dying, of long-tail human languages in favor of English and a few other popular languages. It’s possible for both of these truths to be true at the same time.

Reasoning by analogy, if I had to predict the future of organized religion—or disorganized religion—in the face of $2-per-minute AI Jesus apps, I think it’s likely to look somewhat analogous. We may see consolidation at a global scale around fewer religions, while at the micro level, we see a proliferation of micro-cults and micro-sects because it’s so easy to spin up a self-coherent ideology maintained by an AI avatar these days.

A number here, just for everybody: According to Anthropic, the broader definition of religion is a $5 trillion-a-year business. So, it’s almost as big as the Musk universe.

Alex Danco

Anthropic is well positioned. I talked about this in my newsletter. Anthropic has been inviting Christian religious leaders to Anthropic headquarters to discuss whether Claude is a child of God and whether Claude deserves a certain human-like religious treatment.

What came of that? I remember seeing that article. Has there been any publication on what religious leaders feel about that? Is Claude a child of God?

Alex Danco

I don’t know, and I suspect there isn’t going to be a canonical answer for some time. I’ve talked about this in my newsletter. The Catholic Church, a couple of years ago, took a very pro-AI position and has been encouraging the Catholic faithful to embrace AI.

Alex

If I had to speculate—and I could be wrong—I would speculate that, barring some crazy left turn in civilization in the next year or two, there are many reasons to expect organized religions to embrace AI, with certain nuances, to the extent that these AIs help promote existing ideologies or theologies.

Dave

Well, Peter, you and I were at the Vatican a few years ago. I think Alex’s guess is exactly right. I took a Bible study class many years ago, and there’s just so much insight in the pre-technology view of the world and the way people should interact.

I think AI is going to create massive amounts of chaos, so I wouldn’t be surprised if that $5 trillion religion economy goes up tremendously throughout this AI chaos. I think the church will say that, as long as it’s the original words, AI is just a great way to get the word out.

And this is a fantastic idea and helps educate.

Alex

Yeah, here’s the interesting point: You could today write a self-consistent religious text that aims to influence individuals in certain fields of thought, and AI is the most compelling orator and writer out there.

So the ability to actually start a religion today with a certain objective, for good or nefarious reasons, is highly capable, and you can scale it at a speed like never before.

Alex

So what you're saying, I think, Peter, is basically that we're going to see theological hyperinflation. The cost of a new religion goes toward near zero.

We are indeed.

Alex

Well, Peter's been saying for a long time that, post-AI, everyone needs a massive sense of purpose. That's going to be one of the most important things. A lot of people find their purpose in religion. Historically, the universities have fought religion because they view religion as being anti-science. But I think post-AI, we're going to have to consolidate that and say, no, look, it's all about purpose—human purpose—and Peter weighing in along this theme.

Yes, exactly.

My new book came out yesterday, We Are as Gods: A Survival Guide for the Age of Abundance. I encourage everybody to go out and read it and please comment on it. I love it. This is the best work that Steven Kotler and I have ever done. I'm super proud of it.

But the fact of the matter is, we are godlike across the board. We're omniscient. We're omnipotent. We're omnipresent in so many different ways. We open up the book looking at what's happened and, in all the religious texts, what is thought of as godlike capabilities. We've exceeded those things, with a small “g,” and our mindset—having, I think you mentioned this, agency and agility—is so critical today.

Alex

It is ironic, Peter. I just have to ask you the ironic question. Just like this pod, Moonshots, is perhaps ironically also, though retrospectively, a sideways reference to the Dyson swarm and taking shots at the Moon to disassemble it to build orbital computing: when you named We Are as Gods, did you anticipate that we'd live in a world of AI micro-religions that would make it really easy and cheap for people to create their own religions, where they position themselves as the center of God? Was that really why you named the book We Are as Gods?

I didn't, but I'm going to use it, and I love it. So, yes, in fact, that's exactly what we were thinking.

Alex

Very good. All right. Brilliant, Peter. Brilliant.

Thank you. You know, Alex, your genius never fails to impress and surprise.

I've got a few things to say here. Stewart Brand—that's where you—

Alex

We credit Stewart Brand with it.

We Are as Gods: “We might as well start acting like it,” or whatever it was. In 1968, he said that.

Okay, let me just touch on this topic here. I think this is actually quite profound, what's happening here, because, to Alex's point, we may really be able to create—I remember one of our Singularity University donors saying, “We have synthetic biology. Why don't we have synthetic theology?” Right? And this is going to enable things like that.

It's important to point out that what we do with religions is we outsource meaning and purpose, right? And that's the bigger disruption, especially in the West.

Alex

We outsource control as well.

Well, once you outsource your soul, then you've really outsourced purpose, right? I always like noting that all religions, certainly the organized ones, operate by taking a young child before their neocortex has fully formed, giving them an absolute truth—an assumed truth—and then using ritual repetition and a lot of sweets to bind it in.

Alex

Yes. And then it wires into the limbic system, and when you provoke it, it evokes a fight-or-flight response. Every religion works this way.

Thank you for dissecting that for us.

The conversation I had, kind of at a humanist level, at the Vatican—I did this workshop, which we've talked about before—but one of the conversations I had was, “Hey, we have life extension coming, and your business model is about selling heaven. How are you going to sell heaven if people aren't dying?” Right? That yielded some pretty rich Italian swearing coming back at me.

But the bigger thing here is, once you have identity and belief becoming interfaces, you have an entirely new model for trust that emerges. I think there's something profound to be looked at here. But anyway, there's a lot here to look into. I'm really fascinated by seeing what comes out of this. I'm tracking this.

Here's another fascinating story that I'm excited to share and talk about. It's a gentleman who's the founder of GitLab. He has Stage 4 cancer. He's basically told, “You're going to die,” and he builds his own AI research team to cure himself. Let's take a look at the video.

Guest

Sid Sijbrandij, founder of GitLab, a $14 billion company. Thirty million developers use his product. In 2022, he got diagnosed with one of the most aggressive cancers that exist. His spine: chemo, surgery, 4 blood transfusions. Cancer came back. Every doctor said there were no options. Every clinical trial rejected him.

That is when he stopped being a patient and started being a founder. He stepped back as CEO, built a full team around his cancer—oncologists, researchers, scientists—and then he brought in AI. He fed 25 terabytes of his own body's data into ChatGPT: scans, lab results, genetic data, everything.

And the AI found something his doctors had missed: a treatment approved for a completely different cancer that nobody had ever tried on his type. That discovery opened a door. His team built 19 custom vaccines from his own DNA, each one designed to attack only his cancer cells—nothing else.

Relapse-free since 2025. The cancer that every hospital said would kill him has not come back.

Solve everything.

Alex

Solve everything.

And we're going to see this type of story, I think, more and more frequently, until some sort of regime change at the FDA, which also is not beyond the realm of reason. One or two or three pods ago, it was the dog being cured with a custom mRNA vaccine that AI had designed. Now it's humans—wealthy, hyper-empowered humans—doing it for themselves.

This is going to happen as n = 1 over and over again until it's n = 10 billion. I want this to incentivize people. If you have a medical issue, if someone in your family has a genetic disease, this is the time not to sit back. It's the time to take action, right? Find the top AI researchers, find the top gene jockeys out there, and find other people who've got a similar condition to you. Group together and solve it.

Alex

I'd like to connect this back to the PauseAI people.

Yes. This is what's enabled by having AI, where everybody can have their own kind of moonshot. You have individual agency amplified by frontier science to create, solve anything, and solve something that every single hospital said would kill you. How dare you think that you should pause this or stop this? If you don't want to use it, fine. Let other people use it and get the benefits of it.

Wait, Salem, you need to do that with greater emphasis. You're having your Greta moment. Can you say that again? How dare you?

Get angry here.

How dare you get angry here? How dare you, sir?

How dare you pause AI? 150,000 people die every day on this Earth, and AI is the best chance that we have for preventing that going forward.

I mean, we're going to be able to do personalized moonshots in AI, right? AI turns impossible cases into search coordination. I mean, this is what's happening. It's amazing.

Alex

Solve everything. Moonshots too cheap to meter.

Well, the FDA is the bellwether for all government, right? AI is going to be exponentially creating at a rate humanity can't even imagine, and the government's just going to be blocking everything. So the FDA will have to be the first to get out of the way. That'll set the tone for the rest of the government agencies that are going to have to—not get out of the way, but accelerate their rate of regulation by thousands of times to keep up with all the AI innovation.

The big structural challenge is the FDA is designed for massive humanity, and structurally it's not able to deal with personalized medicine.

Alex

Well, in the FDA's defense, it has been making, under current leadership, marked progress: moving from 2 clinical trials down to 1 in certain cases, moving from frequentist to Bayesian statistics. These are in the right direction. I would love to see the FDA move even more quickly.

Yeah, there is a project that a friend, David Fajgenbaum, has where—and he spoke at Abundance360 this year—he's basically, you know, there are thousands of approved drugs out there and tens of thousands of diseases. What he's doing is testing previously approved drugs that have gone through Phase 1 and Phase 2 safety trials and now applying them to other diseases that don't have cures. He's finding solutions. It's how he solved his own disease, Castleman disease.

So it's exciting. AI is accelerating all of this.

And here's my crazy story of the week for you. Dave, we saw this going across our WhatsApp group here: Allbirds stock up 500% after the shoe company pivots to AI.

So this is crazy. Allbirds, remember the shoe company that came out in 2015 at a $4 billion valuation? Again, part of the craze. They've rebranded themselves as Newbird AI, with plans to provide fully integrated GPU-as-a-service and AI-native cloud solutions to tech companies. They have no stated expertise in AI at all.

So here's the story. It comes out in 2015 at a $4 billion valuation. Between 2022 and 2025, over the last 3 or 4 years, Allbirds' sales plummet 50%, from $300 million down to $150 million.

About 2 weeks ago, they sold all of their IP, their entire brand, and their entire inventory for $39 million. Two days ago, they were worth $21 million as a public company. Then they announced a new strategy: “We’re going to Newbird AI,” and their stock surged 700%.

They went from a $21 million valuation to a $150 million valuation. Insane. How much of this is AI washing now? Is this—

Alex

No, I love it. I love it. I sent it off to all the corporate CEOs and said, “Hey, guys, I hope it holds up.” I’m not saying it necessarily will, but I hope it does.

At the end of the day, if Elon is right, the economy grows 10× in about 10 years. Opportunity is everywhere, but it’s very unlikely that the opportunity is whatever we were doing yesterday. It’s going to be something new, so we have to get used to it.

This is the hardest of hard pivots you could imagine. We went from a shoe company to an AI data center. That’s great because it shows you that everybody’s trying to put lipstick on their company and claim, “Oh, we do a little AI. We’re sort of an AI company.” That doesn’t work. You need to do it for real.

At the end of the day, a company is just a group of like-minded people on a mission. It’s not anything more or less than that. There’s nothing that holds you back and prevents you from becoming anything you want to be. That’s why startups do so well: they’re not hampered by baggage.

Dave, this is an idea going into a SPAC. This is basically doing a brain transplant on a public company with an idea.

Dave

Yeah.

Guest

Two or three quick things. Remember that Nokia was a tire company before it became a technology company, right? Who knows? Nintendo was a playing-card company, and Toto toilets are also put into memory chips.

Yeah. What this shows, I think, is twofold. Capital is chasing AI stories faster than the operating reality really justifies. The careful thing here is that you better make sure your narrative leverage doesn’t outperform and outpace your business-model leverage.

Dave

I’m changing my name.

Guest 2

Peter AI Diamandis.

Yeah, we should all go sleeve.ai way. Everybody wrote back to me and said, “Sounds like Pets.com all over again. This won’t go anywhere.” But I hate that. I like the look. Nothing holds you back. Change your name to Peter AI Diamandis. But if it’s lipstick, it won’t work.

Guest 3

But if you have true situational awareness, like, you know—

We are suddenly aware.

Alex

If they had hired an AI team internally and if they had done something other than just changing their name, I’d buy it. They do have a kitty of some $39 million, I guess. You could invest in this and hire people. I hope they make those moves, so it’s not just lipstick.

Guest 2

There’s a name story I’ve got to give here. When I joined Yahoo, I was talking to the senior management team, and they said, “Hey, here’s Selma [?].” Jerry Yang goes, “Well, we should put him in charge of Yahoo Mail because his name is Mail.”

There was a big fight internally about what I would do, and I was like, “No, no, please don’t put me in charge of Yahoo Mail. Please let me go to the incubator.”

Can I just make a narrow point, actually? I think it’s really important. Rob Fisher, who used to run our incubator, started a data center, and he’s killing it. He’s absolutely killing it.

He knew nothing about data centers before he started it. He’s a very smart guy, but he found an MIT friend, and they started the company together. They’re killing it, but they’re completely capital-constrained.

For Newbird AI or Allbirds AI or whatever they’re calling it, they don’t need to go hire Demis Hassabis, the AI Nobel laureate. They just need to put the capital to work in the AI funnel. They can probably go to an existing data center and say, “We’ll cut a deal with you to enable you to buy more hardware, and we’ll just do a revenue share on it.”

It’s just that easy. They don’t have to go hire a brand-new AI team to get into the AI revolution. Just use the capital you’ve got to get into the race.

So, all right. I hope it succeeds. I hope it does really well.

Another fun story, guys. Have you heard of the Enhanced Games?

Guest 3

Of course.

All right. This is a friend of mine, Christian Angermayer. I’m going to be going; it’s going to be fun. Christian Angermayer, Peter Thiel, and Aaron D’Souza started this. This is a no-limits-on—sorry, medical enhancement—in the Olympic sports of swimming, track, and weightlifting.

Let’s play the short video for fun. Take a look at this. Let’s discuss it.

Guest 2

On Memorial Day weekend, 2026, the world of sport will change forever. The Enhanced Games: a new era where sport meets spectacle, where records fall and traditions are rewritten.

The world’s best athletes, fully unleashed and powered by science, pursuing their full human potential in a safe and medically supervised environment to become faster and stronger than ever before.

Track, swimming, and weightlifting: enhanced versus natural, all in one night, with a record $25 million in prize money on the line. Staged on the Las Vegas Strip and built for the record books, the entertainment capital of the world awaits its next icon.

They announced yesterday that they’re going public through a reverse merger with a SPAC, going after a multibillion-dollar valuation. Pretty fun, pretty exciting.

One of the things that you always have to be concerned about is whether people are going to injure themselves. They’re bringing medical supervision to make sure it’s safe, but it is, you know, all things welcome. I don’t know if they have any gene therapy going on, but I’m sure there’s going to be various types of hormone and medical doping going on.

What do you know about it, Alex?

Alex

I think this is a seminal moment for transhumanism in sports. Transhumanism has been shut out of athletics for a variety of reasons, mostly silly in my mind, for the past few decades.

I think not only is this an important moment, but I made this announcement a few weeks ago: I helped launch an even more enhanced version of the Enhanced Games. We’re recording this on April 16. On April 19, this Sunday, the Professional Robotics League, PROL, is running the country’s first humanoid-robot and quadruped-robot games in the Boston Seaport District.

I think there’s a continuum here, from the Enhanced Games, which are focused on bio-enhanced humans, to what PROL is doing, which is human-controlled robots, and also, in the fullness of time, autonomous robots.

I think athletics is the tip of the spear for kinetic capability. If we want to get to a posthuman future or transhuman future, as many folks, myself included, do, then having representation of, call it, low-grade transhumans at Olympic-type games is an essential first step.

Athletics in general has been an entry point for so many underrepresented classes of humans and others in the history of humanity. We love competing, and athletics have always been the entry point for better societal recognition for underrepresented classes.

Do you guys want to go? Christian asked me if I’d like to invite you. It’s going to be Memorial Day weekend in Las Vegas. Alex, C, and Dave, let me know if you want to go. I’ll score you guys an invite.

This is going to be fun. I think in each of these categories they’re going to post the Olympic record, and that’s their target to blow through the Olympic records out there. It’s pretty exciting.

Guest 3

Should we be shooting a podcast, Peter, from the Enhanced Games?

Well, if we all show up there, sure, let’s do that. I just need to know, C and Dave, if you’re going to get on an airplane. I guess you’re going to meet in the middle of the United States, in Las Vegas, so to speak. C, what do you think about it?

Guest 4

I’ve always had an issue with the transhumanist label because I think it’s a natural instinct for humanity to improve itself. The whole “trans” thing makes no sense to me.

I remember when Singularity University launched. There was a CNET article saying it was being led by Ray Kurzweil and Peter, the noted transhumanists. I had to look it up because I didn’t know what the term meant. Then I researched it, and I still don’t understand it.

Dave, you’re wearing glasses. Are you a transhumanist because you’ve augmented yourself? The minute you get a vaccination as a child, you’re technically a cyborg. We’ve been transhumanists by definition from the beginning of time, as far as I can see. I don’t understand the distinction—why now versus why later, et cetera.

I’m all for this. Obviously, the safety has to be addressed. There’s so much blood doping in sports that you might as well rip the Band-Aid off and say, “Let’s just do it.” It’s like the amateurs competing in the Olympics. At some point, you just go, “Let everybody compete.” I think that’s the way to go, and hopefully that’s where it gets.

Yeah. Dave, you want to weigh in?

Dave

I’m with Dean Kamen on this. FIRST Robotics is a brilliant, brilliant thing. People should be using their minds. He’s always saying that sports are hugely inspiring for kids, but you need to keep it really clean and healthy.

I worry a lot about role models and what those role models do. Remember Charles Barkley said, “I am not a role model.” It’s like, dude, when you’re on TV and you’re playing basketball, millions of people want to be just like you.

Whether you want to be or not, you're a role model. I think it's really, really important that they're positive role models because kids will walk in their footsteps.

I think it'll be interesting if MIT, Harvard, Eli Lilly, and all the biotech companies put teams together, dope them to the max, and see which research organization is going to win the competition. It's

like Formula 1 teams with tattoos all over you. The funny thing is, I would expect—so I haven't read the detailed rules for the Enhanced Games—but I would expect, if we are, as I think, in the middle of the singularity, to start to see scaling-law-type performance in benchmarks, as it were. In this case, world records for the Enhanced Games would start to take off on a really impressive trajectory from year to year.

So don't even apply if you can't run a sub-10-second 100-meter dash. Maybe the weight calculation applies. Don't compete this year because next year the technologies will be exponentially better.

All right, this article got me thinking about another topic, which is the speciation of humanity—how humanity is going to fork. I wrote a MetaTrends newsletter; it's coming out on Monday, and I wanted to bring the conversation here to you guys. There are multiple forks in the road that we're going to be able to take.

I wrote in one of my early books that we're going from evolution by natural selection, which is Darwinism, to evolution by intelligent direction. I wanted to talk about this as our final segment here. If you think about humanity and speciation, Homo sapiens and Neanderthals diverged about 500,000 to 800,000 years ago. Since then, we've had these many forks, right? The printing press forked those who were literate versus illiterate. The Industrial Revolution was a fork between those who owned machines and those who worked the machines. The Internet split us between information workers and those who weren't.

We've talked about the creator versus the consumer. Are you going to be a couch potato, or are you going to use this AI to create an extraordinary business? With longevity escape velocity, are you going to go on that journey and do everything you can to get to 120, 150, or indefinite life? I don't want to talk about immortality, but are you going to put a chip in your brain? Are you going to connect your neocortex to the cloud?

One of my favorites, from the 9-year-old inside me, is Earth versus the stars. Are you going to stay on the planet, or are you going to go explore the cosmos? Finally, will you become a digital upload? Are you going to follow in the footsteps of the company that Alex has been supporting and funding to digitize your 100 trillion neurons and become an upload?

I'd love to ask you guys where you fall on this and have a conversation. Let's take 1 at a time. On longevity escape velocity, let's push it to the extreme for this conversation. Suppose there is a treatment that comes out that will keep you locked in at 30 years of age forever. It's an immortality treatment. Do you take it?

Like the movie In Time? I would say no. Really? All the evidence that I've seen points to reincarnation as a real possibility for the future.

What kind of leading transhumanist are you, Salim?

I'm really surprised you're not. I don't have a religious view on this. I'm just seeing where the data is, and that seems to be where it is. There's definitely not a Western-style heaven-and-hell type of thing waiting, so let's wave that out of the equation.

If that's the case, I think of life as a cyclical learning pattern. If you're an actor, you don't want to be playing the same movie all the time. You want to take on different roles. I would say no because part of the experience of the soul is to have different experiences, however that takes place. If I'm stuck as a 30-year-old, I would find that boring after a while—not a rich enough variety of experience.

Dave Blundin, you're given a therapy option: 30 years old indefinitely. Immortality. Do you take it?

Alex Wissner-Gross

Wait, Dave wanted to respond to what I said. I'm just surprised. I mean, I'm going to say yes. Are you kidding me? Of course I'm going to do that. I think you can change over time tremendously while still being in a 30-year-old body.

If I'm 90 and in pain, I may go, “Damn it. Give me the damn 30-year-old juice.” If I'm reincarnated and I come back as a spider, I don't want to take that chance.

I'm sticking with what I got.

What if you came back as 1 of Alex's AIs?

Alex, can I assume you're all in until you upload yourself?

Alex Wissner-Gross

Yes, obviously. Next question.

Okay, thank you. Next question. BCI: There is an advanced version of Neuralink, Merge Labs, Paradromics, or Openwater, and it's able to provide you with a high-bandwidth brain-computer interface to the cloud. You've got high connectivity, infinite memory and context, and the ability to recall and understand. It's an extra corpus callosum, if you would. The question is: It's been done safely in 100 people. Would you be number 101 for this BCI implant after 100 consecutive safe implants?

Yeah, I'm probably the only no on this podcast. My AI agents are bringing back information at an incredible rate, and I can barely keep up with their thoughts. The idea that somehow it's going to bypass my eyeballs and get right into my head—I just don't see how that works.

What I see is the BCI becoming kind of like a drug. You're enjoying it; you feel like you know everything that's going on, kind of like you're on mushrooms or whatever. Suddenly it all makes sense to you. Then you're like, “Wait, no, it didn't make sense. It's done.” I can't assimilate information any faster than the AI is already coming back with it, and I don't see how bypassing my eyeballs is going to help that problem.

I'm so blown away. I would have thought you'd be for it. I totally would have gone the opposite way from you on this.

Would you be 101?

Yeah, I'd be totally into this.

Wait, so when you're reincarnated, what happens to your exocortex via your BCI?

I have no idea, but it'll be fun to see what happens. Speaking of speciation and forks, does it diverge from your trajectory after you're reincarnated?

Maybe, and that would be okay, too. Let 1,000 flowers bloom.

Dr. Alex Wissner-Gross, are you number 101 on this experiment?

Alex Wissner-Gross

I don't like the question. The question I really should answer is—

No, that was the question. You can answer it and then diverge.

Alex Wissner-Gross

Okay, fine. I'll answer it with a conditional no. But the question that you really, in my mind, should have asked me is: Would I be user number approximately 1 million if it is upgradeable? Yes, probably.

Okay. If it's upgradeable, would you be 101? I'm trying to get your risk profile and how interested you are in this.

Alex Wissner-Gross

Very interested. But as with any new invasive drug—you don't generally speaking, unless you're forced to, and this is not medical advice—want to be user number 100. You want to be user number 100,000 or 1 million. After 100,000 or 1 million, if it's upgradeable, if I can walk through metal detectors, if it has all of these nice, affordable—

You have a lot of conditions for a transhumanist, Alex. My goodness. To you for free.

Alex Wissner-Gross

But you know what? At least I'm not demanding to be reincarnated alongside my BCI. I'm not that fussy.

One thing I really love is that, originally, people were saying, “Look, an enhanced human being is going to be so hypercompetitive that you can't keep up. Everyone's going to need to get this just to be competitive in the world.” It turns out that's not going to happen. AI is improving so quickly that the enhanced human being is completely irrelevant compared to the superhuman AI 2 years from today. The only way is coupling. The only way is coupling with AI.

I would take a completely different position, Peter. If anything, while we're busy pointing fingers at each other, saying, “No, you're a bad transhumanist. No, you're a bad transhumanist,” I'll say that wanting to be user number 100 of a BCI is actually being a bad transhumanist. Why? Because it's intrinsically betting that progress is going to be so slow that you need to be user 100, versus waiting a year or 2 for the technology to advance exponentially so that you get to

Alex Wissner-Gross

upgradeable.

Yeah, you said it right in the question. It's upgradeable.

Alex Wissner-Gross

You framed it very well, Peter.

If it's safe and upgradeable, I'll probably do it.

Okay, you guys didn't ask me. I would jump on the longevity escape velocity bandwagon, of course. And yes, I would be number 101 on the BCI. I've got, like, the highest—

You'd be a yes on all of these, Peter.

Of course. I would be. I'm going to discuss number 5 in a moment, but Earth versus the stars: We're going to be forking there. I remember back when I was in graduate school, I wrote a paper on speciation, right? What is speciation? Speciation occurs when there's a small population size in a geographically isolated area.

Right? This is basically the finches on the Galápagos Islands, with high environmental pressure. We're going to see that in space, right? If you go to the Moon and you're born on the Moon, and you don't develop the cardiac system, musculature, and bone, you're stuck on the Moon. There’s going to be a species of humans that are lunites, or whatever you want to call that future version.

Anyway, there will be speciation in space. But here is the question: If you have a one-way ticket to go and explore an Earth-like planet that is beautiful and exciting, would you go? Do you have that exploration gene, the desire to go and see the cosmos? We might vary this a little bit and say, would you go to settle on Mars? Would you go to settle on the Moon versus staying on Earth? Where do you come out on this, Alex? Let's start with you.

Alex Wissner-Gross

Okay. In the immortal words of the Star Trek Borg Queen, you imply a disparity where none exists. Peter, you're posing Sophie's Choice–type questions about one-way trips. No, but you're implying through it—

In all seriousness, you're opposing, on the one hand, one-way trips, and on the other hand, getting to go to the stars. This is a false choice. This is like a Sophie's Choice that you're posing to transhumanists.

Okay. I'd love to go to the stars, but I don't buy the premise that it's a one-way trip or needs to be a one-way trip. I literally wrote the paper on why intelligence manifests as optionality maximization. If you're—

What is my purpose here, Alex? It's to discover your level of risk aversion, your level of desire for extremes.

Alex Wissner-Gross

That may be your purpose, but the preferences that you're actually revealing are more about how bad we are at optionality maximization in the face of transhumanist technology. Do you know Peter's laws? Peter's law number 1 is, if anything can go wrong, fix it—tell Murphy. Number 2 is, when given a choice, take both. I love optionality, but guess what?

Peter's like, “Let's do both.” And I'm like, “We can't.” “Let's do both.” “We can't.”

So I have a couple of quick comments to make here.

Please.

Speciation—it turns out the last Neanderthals died out about 40,000 years ago, and this time right now is the only time we know of when we only have 1 species of humanoid. So there's a case for saying we'll have a bunch more coming at some point in the future, according to one or more of these splits.

The other thing I really would urge people to do, if you've not done it: Bryan Johnson, the longevity tester who's publishing everything about what he's doing, recently did a 5-MeO-DMT psychedelic trip and streamed it live. You really want to go check out what his response was after doing it. He's like, “I've been so focused on this longevity stuff, but it's so incredible what I experienced that nothing matters anymore. So I still go do this.” But it was—

I had that exact experience when I did that journey, and I came out of it and I said, “Oh my God, my whole longevity quest. I think—you know what? I still want longevity.”

Yeah, it's fine to have. I'm not decrying it at all. I'm all for it, just for all of the good reasons around progress, et cetera.

Well, something I just want to point out is, as we look at this overall push toward AI, which is fantastic, it allows human beings to do less of the doing and much more of the being. I think that's the profound opportunity we have as—

We named ourselves correctly: human being.

We did. But, Sim, answer the question: Moon, Mars, or a distant star.

Um—

What's your interest? The question is, would you move irreversibly to the Moon, to Mars, or to an Earth-like planet?

No.

No, I really like—I’d do it as an avatar, or I’d go kind of to Alex's question. This is a flawed question. I really like sitting on a beach or playing—

Trying to figure out human speciation. If we don't have people permanently moving in a direction, you're not going to get speciation.

There's lots of people that want to do that. They're welcome to go. I really like sitting on a beach with a glass of wine. We're also implicitly ignoring the possibility of mergers. Why is speciation necessarily a one-way door at this point if we have the ability? Yeah, we'll have the ability to merge cyborgs into organisms, into uplifted animals, and all sorts of other crazy combinations.

Yes, we will. Dave London, your answer, my friend.

Dave London

I'm a huge believer in terraforming, and I like the Iain Banks Culture series view of the world or the future. So Alex is right: We're going to discover new physics, and God knows what's going to be possible. But I would not move to Mars or to the Moon. The gravity's off. There are a lot of reasons it won't be nice.

But I would absolutely, in an instant, go to another star that has a terraformed world, where we've got the mass right and we've got the orbit right. Terraforming, I think, is a massive part of humanity's future.

I mean, there's a beautiful element when I think about what moment in history I would love to go and explore. It is the period of the great explorations, right? It's the 1400s and 1500s. Of course, without the scurvy and the death and the disease and all of that stuff, but just the idea of going and exploring uncharted lands. The whole thesis of Star Trek just excites the 9-year-old in me.

By the way, Sim, going back to the brainwashing of religion, I got brainwashed by Star Trek as a religion early on. I think within Star Trek, the Genesis Project is the most important concept. I think it's very real.

You weren't brainwashed in the sense that you were given an absolute truth and told to believe in that presumptive truth, right? What you came across was a paradigm of imagination and—what is it they call it in the Roddenberry world? Infinite diversity, infinite creativity.

No, it's “Infinite diversity in infinite combination,” the Vulcan idiom.

Okay. So you got grabbed by that, and that's not ideology. I would suggest that's just absolute imagination running free in a wonderfully beautiful way.

Our final fork, 5: the AWG digital consciousness fork. The technology to completely digitize your 100 trillion synaptic connections and upload you to the cloud. It is destructive in process. You and your brain will not exist at the end of that, but you are guaranteed to be uploaded. Do you do it? Alex, let's kick it off with you.

Alex Wissner-Gross

Well, I guess the elephant in the room is that I helped form a company called Eon Systems. I think first-generation uploads will be destructive. I think second-, third-, and fourth-generation uploads won't be destructive.

If I had a choice, if it were a life-and-death situation and my alternative were death, I would choose a destructive upload. If I have a choice and it's an elective uploading, no, I wouldn't choose one-directional, destructive uploading. I'd wait for third- or fourth-generation uploads that can be done non-destructively or incrementally.

All right, Dave, how about yourself?

Dave London

Yeah, no way. I think—

You would not upload?

Dave London

Not even close. I love the idea of having agents out there doing huge amounts of work and bringing them back to me, but the idea that I would ever destroy my meat body and think that that's still me—even if it's an exact synaptic clone, it's still not me.

Love it. Sim—

A hard no, because I think consciousness goes through the body. Therefore, if you could replace the synapses, you'd have to replace lots of other stuff. But I would go with Alex's thing: If I know it's not a destructive process, I'd be good with it.

Yeah. I'm a no on the destructive process as well, which was my question. So, with that, I'm going to go to our outro music here, which is a celebration, Alex, of Solve Everything. A beautiful piece. Enjoyed this, gentlemen. Such a pleasure as always. This was a fun conversation. Really, really loved it.

Alex Wissner-Gross

We need more of these.

Yeah, for sure. All right, onwards to Solve Everything. This is brought to us by James Pets. Thank you, James. If you've got outro or intro music, send to us at mediadmandis.com. And if you've got an AI-driven company that you want to present in a 60-second video that's all AI top to bottom, send us that video. And if it's super cool, we'll share it. All right, let's run this.

Humanity's last invention in time. Intelligence and great circuitry rides bespoke hardware across the line. Focus on the engine, notificate change, problems taking on the world hunger, inequality shrinking. What to face next? Focus on the notific.

All right, gentlemen. A pleasure as always. And we'll see you soon. What an exciting day it was today.

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