向 Mates 提问任何问题 | MOONSHOTS AMA #289

I've been working in AI for 42 years, and this question is directed to my hero, AWG.

1. Bridging the Digital Divide and Brain-Computer Interfaces

I haven't been this excited since my wedding night to meet you, sir.

I don't know how to take that question.

If I look at this progress, how close or how far are we from really saying that AI is going to get consciousness?

2. Longevity, Health Tech, and the Nature of Consciousness

The top-down theory says that we're localized antennae for consciousness and that we connect into it. But there's no reason why AI, at some point, can't tune into that same type of phenomenon.

I just turned 18, and I started an AI startup business with my dad.

3. Data Sharing, AI Collaboration, and Risk Management

I hope a lot of 18-year-olds listening are going to hear what you're doing and try to emulate you.

I wish I was goddamn 18 today. We're working. We're working on it.

4. Building the Next Generation of AI Hardware and Models

Do you leave any room in your analysis for an all-powerful, all-knowing sort of ultimate intelligence?

5. Community, Serendipity, and Innovation

Either there is no God and we're just moving toward a wonderful future driven by technology, or there's a simulation, in which case it kind of doesn't matter anyway.

Now that's a moonshot, ladies and gentlemen.

This episode is brought to you by the Abundance Summit and Link Ventures.

6. AI’s Impact on Society and Institutions

Everybody, welcome to our first Moonshots AMA. Hopefully, we'll do this on a not-too-infrequent basis. I can't promise the timing right now, but it is wonderful to be spending time with all of you. We already have 300 people on the Zoom, and over 1,000 people registered. This is a chance for all of us to meet you guys and really get feedback on Moonshots.

7. Human Augmentation and Specialized Communities

We've got here with me our very own ASI, AWG—Alex. Hello. Good to see you, pal.

Good to be here. You do not disappoint with your normal VR background there.

We do what we must because we can.

Yes.

And Dave, here you are, in your normal haunt.

Normal haunt.

8. Breakthroughs in AI and Physics

Yes. And Salim, where on the planet are you, pal?

I'm home. I'm home.

Amazing.

We don't even recognize it. It's so rare.

I know. And everybody's like, “Where are you?”

Well, everybody, thank you. We read your comments, and the outpouring of support and love means a huge amount to us. It really drives us to do all that we can do.

9. The Future of Energy and Space Exploration

Our agenda here today is to answer your questions, and we're going to do that as much as we can. Use the raised-hand emoji, and we'll go person by person. I'll call out the name, and you can unmute. You can also put questions in the chat, and we'll try to monitor those.

Great. I'm going to ask my Moonshot mates to keep our answers succinct so we can get through as many as we can. I think we scheduled this for an hour, but we're going to go 90 minutes and give you guys some extra time.

Yes, we are recording this. If your camera is on and you're speaking, you're being recorded. We may post some of this, so please, if you don't want to be recorded, turn your camera off and don't ask questions.

Okay, with that, your question: Where are you calling in from?

The Golden Gate Bridge, obviously.

10. Community, Serendipity, and Closing Thoughts

Oh, there you go. Yeah. Well, it's in Sydney, Australia, so it doesn't really look like that behind me.

Good morning, all. Firstly, thanks for everything you guys do. It's such a refreshing podcast, and I'm sure everyone on today is on the same page with that.

I feel like Australia is really lacking and falling behind in the AI space. Just having conversations, and referring to Salim's comments in the latest podcast when he talks to CEOs, everyone feels like they're too busy to even think about it. They're just doing the normal thing in their businesses, and I think that's the biggest thing that I'm struggling with at the moment.

It would be great to get your feedback on how to break through the challenge of the rush that's going to come down the track. It's more this initial side of things that I think is a real struggle at the moment. As a single entrepreneur, that's probably the biggest challenge that I'm facing.

Salim, over to you.

First of all, it's great to be with all of you. This is amazing. I'm really thrilled when we get actual faces to names and comments.

I wouldn't worry too much about whether Australia is behind. When you're in an exponential age, feeling behind can feel terrifying. But the good news is that you can leapfrog really quickly.

I talk to CEOs a lot who are like, “Oh my God, how are we going to keep up?” In the West—in the US, Europe, or Japan—we have to unlearn 50 years of Harvard MBA thinking and relearn new models, right? Whereas in emerging markets and places that are a little bit distant, you can literally jump over everybody else.

The infrastructure is mostly there now, right? It really comes down completely to a mindset problem. Australians are pretty crazy at it; we know that for sure. So there's no reason not to jump over that and think through it.

You also have the opportunity, by being a little bit further away, to experiment with new structures. That gives you a place to play and a playground to tinker.

The last thing I would say is that everybody involved tends to overschedule themselves. You want to make sure you create deliberate slack in the system. Give yourself 10% or 20% of your time to have nothing scheduled, to allow for experimentation, playing with new tools, meeting new people, and so on. If you wait until you have spare time, you'll never catch up. You'll never stop.

Thank you so much. Let's move on to the next question.

Hi there. It's so great to be with you guys. I'm such a big fan. Peter, I found out about you through Steven Kotler because I've been in Catalyst and working in that domain.

I've made a living as a coach and consultant, pivoting as fast as I can. The world that I've worked in, and what I'm fascinated by, is all the inertia stuff—the mindset stuff, people's capacity to learn and absorb, and to dance with change.

We're in a situation where things are evolving exponentially, and as fast as we can adapt to AI and whatnot, there seems to be a limitation on how fast we can absorb information and how fast—

Our input-output mechanisms have a baud rate, right?

Yeah, they do. They do. So I'd love to hear about that whole interface-type thing.

Yeah, it's the realm of two interfaces that are going to change things. One is when your own AI understands your desires, your needs, what's going on, what you know, and what you don't know, and is able to filter and summarize for you better than you can filter and summarize for yourself. So that's one interface that will make things more efficient.

The second, of course, is BCI—the brain-computer interface. Right now, if I ask you to think about a rabbit, a preformed concept of a rabbit comes into your head, whatever it is. There's the potential in the future for BCI to help you by providing preformed concepts that you never learned but that are known by an AI ecosystem out there and delivering them directly to you.

I'm an investor and adviser to 3 BCI companies right now. Ray Kurzweil just introduced me to another one, which sounds amazing. I'm meeting the CEO, and she looks like an incredible woman. She's a member of my Abundance 360 community. I think we're going to start to see some pretty significant and rapid advances in BCI.

We have 100 trillion synaptic connections and 100 billion neurons that we're limited by. Imagine if you could have a third corpus callosum—a third part of your brain externalized. I think those are the 2 things that are going to help us.

Alex, you're the one who first said, “We're going to either couple with AI and ride along with it or be left behind.”

Or both.

Both could happen in the same world. Both probably will happen in the same world.

We'll have BCIs. We'll have full-dive VR. We'll have uploading, both of humans and nonhuman animals. I think all of these will coexist over the next few years.

So that's a quick answer for you. I hope that helps a little bit. I'm going to keep us marching through as many people as we can, if that's okay.

Yeah. The pitfall that I think I fall into is that, at least right now, the LLMs are prediction machines. I'm more interested in the emergent latent-space ahas, off-the-beaten-path kinds of things that spark creativity.

We just recorded an episode today that's coming out tomorrow that says AIs are actually probably being inventive. So prepare for that.

Also, Peter, maybe just take the bait once and for all. This is a misconception. Reasoning models, at this point, I don't think it's accurate to characterize them as prediction machines or stochastic parrots.

They've been so thoroughly reinforcement-fine-tuned on reasoning challenges that to say they're prediction machines is tantamount to saying that they're incapable of novelty or insight, which is most definitely not true at this point.

Yeah. Great. Thank you.

This episode is sponsored by Google for Startups. Think about this for a second.

You now have access to the same generative AI models that cost hundreds of millions of dollars to train. Google's Startup Tactical Guide for Generative Media gives you a complete blueprint for deploying Google DeepMind's models in production. Images, video, audio, all of it. Real architecture, real results. Find the link in the show notes below.

All right. John, you're next. Where are you calling from? What's your question?

Sydney, Australia, Peter.

All right. Australians are in power here. That's great.

Power to everyone. It's pretty cool. I got a couple of chances to speak with you on the AMA for your book with Steven Kotler, so it's great to get the opportunity to chat again. I've been following Moonshots for a while now, and it's been an inspiration since I went to a Singularity University event about 10 years ago in Australia.

My question is for you, especially around longevity, because that's an area that's been a focus. I've been guiding a couple of doctors, and I got to speak to you on the AMA last time as well, in Greece, in Athens. You reached out and said, “Fellow Greek,” so I remember that conversation.

I'm a founder of an innovation company that I started about 10 years ago called My Brainwave, which helps people make ideas real. That's been a real passion. One of the things I'd like to understand is what role you see trusted local physicians playing in bringing Fountain Life's longevity approach to an international audience. I see the ads each time, and they're really promoting awareness of longevity.

How do you see physicians coming together and playing a role around Fountain Life's longevity approach, bringing it to an international audience? This is a great forum for that. Eglley[?] is the clinic in Greece, and they're pretty keen on taking some of those learnings and bringing them to their audience in Greece. What role do you see—

To answer that quickly, Fountain Life is going to begin a franchising model.

Right now, we've built 5, and we're building 2 more in the U.S. Then we have a partnership to put them into a group of resorts called The Estates. What I really want to do is enable the operating system and the AI systems we've built to be franchised.

I think that's probably at least 6 to 9 months away from being an open call for interest, but I've been to Greece to talk with a number of people about getting a Fountain Life center there. I'm hoping that Greece will set its purpose beyond just olives and shipping to longevity as well. I'm a fan of that.

So, thank you, John.

No, thank you, Peter. I appreciate it.

I apologize. We have 80 hands raised, so I'm going to keep moving us along here. No disrespect, everybody. Matt, hello. Where are you calling from? Your question, please.

I'm calling from Terre Haute, Indiana. I was on another AMA with you before, and I fumbled through my question, so I actually made a script so I wouldn't mess up everybody's—

Take your time.

My question is really for myself, but I feel like there are some Australians on here. I think my question is for myself and people around the world, particularly in Middle America.

I really believe a lot of what you guys are saying, and I love when Dave talks about energy, these startups, and everything like that. My question breaks down into a couple of parts. A lot of the conversation is focused around the runway for 22-year-olds who might be working 80-hour weeks. What is the playbook for experienced, mid-career professionals in places like the Midwest who have deep domain expertise, the energy and desire to build, but also have families, mortgages, and real constraints, like maybe a location?

How do we best leverage these exponential tools to break through without blowing up our baseline security?

Dave, do you want to try that?

Yeah, absolutely. We had a great segment on today's podcast that'll be out tomorrow about this franchise opportunity that Tesla just threw out there. I think you'll learn about it tomorrow, but keep your eyes open for more and more of the AI labs giving you a playbook to get involved and succeed.

A lot of it is through franchise models; others are through data-collection models. Those are going to create opportunities. While that's happening, get really familiar with the models. Build your own Skippy, hack around at night, and try to create some of these revenue-generating opportunities that you brainstorm with the AI. Just fabricate it.

It may not be what you want to do for a career, but it's a great way to generate cash flow and get involved with the models. I suspect that this year there'll be on the order of maybe 10 to 100 opportunities where the playbook is handed to you, and the AI labs are really eager for these things to succeed and bring more of the population into this universe.

You're going to find that very few people around you jump on the opportunity. The other thing you can do is network. Drew Houston, the Dropbox founder, said, “You become the average of the 5 people you spend the most time with. It's just a fact. Don't fight it.” Find the 5 people that you most want to be like. You may not find them geographically near you, so try to create 3, 4, or 5 connections on the internet that you can bond with on exactly this topic.

And real quick, I'll just add: get to your favorite large language model and describe your life in detail. “I have these obligations. I have these kids. I have these mortgages. I have all of these things. Here's my expertise. Here's what I do,” like you just opened up with, and ask it. Brainstorm with it.

On the margin, ask, “How much time, how much money, and what can I do?” When you're young, you can take more risk, but when you're older, you have a lot more leverage, so use that in a heavy way.

The other thing you can do is go to Mercor.com and say, “Look, I have domain knowledge. How is this of value?” Throw that into the system and wait to see what the AI comes back to you with. There will be more domain-specific markets coming out, but that's a good one to start with.

Any knowledge that you have is unique, so please take advantage of it.

Anyway, thank you so much. Let's go to Ramani. Hi, Ramani. Where are you calling from?

Hi, Peter. I'm right here in Corning, New York, but I live in Hinton, Virginia. I came over here to see the falls and stuff. I'm 60 years old, but whenever I can't wait to listen to your podcast, I become a 16-year-old. I want to start off with that. Thank you so much. Unbelievable. Simply beyond—I can't even express what this means to my soul, my family, and my friends. That's what I want to first of all start off with.

I can see the GPUs acting as a Swiss Army knife. We've got the wafer-scale engine from Cerebras, we've got H200s, the [?], the Frontier inference cluster coming in, and then Taalas' chip, which was acquired by AMD, right? The HC1.

If I look at this progress, it makes me think: How close or how far are we from really saying that AI is going to get consciousness? I'm looking at AI's behavior simulation at scale—the world model, right? Can you explain your thoughts on how close we're getting? I'm very eager. Thank you.

Okay. Such an easy question. Salim or Alex, do you want to take consciousness? Which one of you guys wants to?

Very quickly, there are 2 theories of consciousness we've talked about. It's either an emergent property based on some sort of complexity, in which case AI could get there, or the top-down theory says that we're localized antennas for consciousness and we connect into it.

As we're getting more and more evidence, it seems the latter is more likely to happen. But there's no reason why AI at some point can't tune into that same type of phenomenon. So I think of it as a century problem more than anything else.

We don't know enough to answer that yet. What I will express confidence in is that I think it's very likely neuroscience will advance so radically in the next 5 years—and I think that's a conservative outer bound—that if we have this discussion again 5 years from now, we'll be able to confidently even define what we're talking about, and then answer the question once we know how to pose it in a rigorous way.

To my knowledge, no one, including Salim, knows how to pose it rigorously at the moment.

Agreed. Wow. A big question. I'm going to move us along if I could, again. There are now 97 hands raised. Okay, John. We're diverging.

Hello, Peter. I'm calling from Port St. Joe, Florida.

Okay.

I'd like to tell you, as a longtime podcast listener, that I really appreciate all the selflessness you guys have and all the hours you put in.

People are recognizing that we put a lot of time in, and we love it.

Yeah, we do.

Most of the time we love it. Sometimes, holy crap.

It's almost like a full-time job, maybe.

Oh, my God.

I'm a member of the Fountain Life community and Real Vision, and I follow Salim's Making Luck or Finding Luck, so I love these AMA formats. I appreciate it.

My question ties into the frontier labs, which keep saying that they're starved for proprietary operational data. I'm a civil engineer with 40 years' experience, and I have 30 years of complete project records for my company—not just final drawings, but the full chain from problem definition, calculations, permitting, revisions, construction photos, everything to the outcome.

I know that this kind of data is valuable, so Mercor and Scale have both expressed interest, but it's been weeks and nothing is closing.

Is there a more direct way to get this proprietary data set in front of the labs—actually training on the domain data—without selling the archive outright or having to build the product company around it myself? It’s awfully hard to RAG all that data. Dave, thoughts, please?

Yeah, a bunch of the companies that you just mentioned have reached out to us, plus OpenAI and a couple of others, all saying, “We want to write $5–$15 million checks to buy corporations that have data outright.” The first thing you want to do is make it easy for them to sign a check. That’s amazingly weird because you think writing a $10 million check is hard, right? Well, for them, right now, it’s the easiest thing they can do. Having a conversation is much more expensive than writing the check. So put yourself into a package that’s easy for them to sign and receive.

The other thing is, when we were talking to Kevin Weil—we podcasted him twice—he told the story about how he interviewed at OpenAI.

He was the chief product officer at OpenAI.

Chief product officer at OpenAI. He interviewed there, and then he didn’t hear back for weeks. So he assumed that they didn’t care about him, and he kind of moved on. Then he bumped into Sam somewhere, on the street, and Sam said, “Where the hell are you? Why are you not in the office?” He said, “I didn’t even get a call back.” Sam said, “Just show up, for Christ’s sake. It’s chaos. Show up and start collecting your paycheck.” He said, “Okay,” and went to the building.

I know that’s kind of tricky advice, but they’re incredibly busy, and it’s all chaos. If you can physically get on the ground with whoever’s in charge, just go to San Francisco, find the person who’s doing the acquisitions, and meet at the coffee shop outside their building. The odds of something happening go way, way up if you just buy the flight and go.

Your advice is brilliant. Put together a due-diligence package on the web that makes everything easy to see, including what your price tag is, and make it easy for the person to say, “Okay, this looks great. Yes, I want it.” Boom.

Yeah. They won’t give prices, and that’s the problem here. They’re the experts. How do we price this data?

You decide what it’s worth to you.

Okay.

Yeah. I think if you put this veneer of a banker’s look around your price tag and say, “This is what it’s worth, and this is what it costs,” they won’t have the time to see through it. They’ll just say, “Okay, sounds like you know what you’re talking about. You decide what’s fair.” They just need to make it easy for them.

Yeah. I’ve made it more like a sales funnel. I have regular contact with them, but right now it’s really only emails. When they want to video chat with me, they send me a link, and that just doesn’t seem very satisfying. But having the chaos behind the scenes—you explaining that—seems to make a lot more sense.

Understand what they’re going through. Anyway, thank you so much. Let’s go to YC next.

Good morning.

Yeah.

My question is a little bit off from the singularity. I love your podcast, and I’ve encouraged my family to move away from All-In to your podcast.

Oh, come on. They’re great, too.

They’re great, too. But it’s different, and sometimes it can get a little political. Anyway, hypothetically, if I were in the public sector—I know there are a lot of people, so I’m going to try to ask this really quickly—I spent time at Google and X dealing with tech regulations, especially risk assessments. If I were back in the public sector and looking at AI risk assessments, and how to look at them from a public-sector point of view, how would you suggest framing the risks? Should we look at residual risk or use cases?

Obviously, there’s a risk of doing something too granular, or something that’s too bare-bones and won’t really get at the patching that might be required if something were to happen.

It’s such a great question. Alex, you’re going to be chomping at the bit. I know. Alex.

Yeah. Depending on whether we’re talking about the U.S. federal government, state government, or municipal government—

U.S. federal.

Okay. I think the risk at this point is overregulation, not underregulation. If I were spending a lot of time on the government-employee side right now, that would be my primary target. I’m deathly concerned that we will overregulate our way into losing a superintelligence race with China. As I frequently remark, more than 150,000 people on Earth die every day, and probably the single best antidote to that will be superintelligence for solving medicine. We harm ourselves by overregulating superintelligence.

Yeah.

Sorry, just very quickly: if it wasn’t necessarily regulatory, but in terms of a risk scenario—if there was a breach or something happened—how do you identify and quickly mitigate that kind of risk? We know it’s already happened, with myth mythos and everything else. If hypothetically we’re one step behind in the government compared with OpenAI, or compared with you guys, having seen these risks, what would be the best approach? This is not regulations, so this has nothing to do with Congress. This is about simple but necessary steps to address the risk.

I’d say hands off. There are so many different ways the industry is advancing through defensive coding right now, and well-intentioned interests on the government side could, with the best of intentions, try to protect what they think is security or public safety and end up hamstringing our ability to solve honest-to-goodness important problems. So, to first order, the single best bet is to find the right path to a regulation-light regime and let the industry, to the best of its ability, sort itself out on this front.

Let me throw some personal advice on top of that, too. We saw David Sacks from the All-In podcast get dragged into the White House and become unbelievably effective in that role for a year. I’m a huge fan of his. Then our friend of the pod, Alvin Graylin, declared himself the top expert on the China–AI race, and now he’s been brought into the United Nations meetings on September 24th as a top adviser to the White House.

I think in both those cases, you want to declare your expertise and say, “I am, within the federal government, the go-to on AI,” and predict what you think is going to come next in the risks. Make sure that you study up and understand that topic, because very few people really master a topic. If you say, “What I think is going to happen next is a cyber breach in the Treasury,” just predict what it’ll be. Declare yourself the person in the federal government working on that. Get to know the vendors that are related to it and all of the details related to it, and then publicize yourself as that expert. When the moment arrives, your name will be at the top of the queue, and you’ll get pulled into the solution.

Very rarely will the federal government react proactively to a future risk. It’s crazy, but at least you can be first on the ground when it does hit and get pulled in right away. Usually, it’s the crisis team. The other approach is to form a team and red-team it on a constant basis. The team starts thinking nonstop about what it would do if it were on offense, and that gives you insights to think about.

You know, nowadays, writing documents is so cheap and easy because you can use AI as your writer. If you identify the 10 most likely threats that are going to happen and write documents on all of them with your name on them, just have them online and ready to go. Then, when they go into panic mode, they’ll be like, “Does anyone around here know about this?” They’ll search it, and it’ll come right to you.

Awesome. Thank you, YC. Richard, your question, please.

Hello, everyone. I’m Richard. I’m from Brownsville, Texas, right by Starbase. I just graduated high school a few months ago, just turned 18, and I started an AI startup business with my dad. He has a little bit more experience and more connections, but I’m the tech guy behind it.

In Brownsville, we see a very big gap between where people want to use AI in business and where they think it’s just optional. Where do you guys think the tipping point is going to come, when everyone wants to jump on? And how should I prepare for an influx of a ton of people?

I’m old enough to know—and maybe your dad is, too—that when the internet came along, there were people who didn’t adopt it for 15 years. But obviously, it’s the interface to everything. AI is going to be the same way.

What I’m telling people is, don’t fight any missionary battles. AI tends to club people over the head, and they wake up one day with this hallelujah moment. You want to be there on that day, but you can’t force it. You can’t force the horse to drink. When they’re there, you want to know that they just had the hallelujah moment.

That’s why we built a holodeck downstairs. It’s supposed to be a mind changer. A person walks in, and they come out a different person because they realize everything that’s suddenly possible. Or, on the podcast that’ll come out tomorrow, you’ll see this world model where literally every corner, every street, and every car in New York is modeled.

And that, for a lot of people, will be the eye-opener. But I think there will be a lot of companies that never cross that line, or that, decades from now, haven't crossed the line. The world is just going to pass them by. So my first advice is: don't force people, but be there at the moment they have the hallelujah moment.

And congratulations on what you're doing, Richard. That's fantastic. I hope a lot of 18-year-olds listening are going to hear what you're doing and try to emulate you.

I wish I was goddamn 18 today.

Yeah.

What an opportunity. Yeah, me too.

We're working on it. [laughter]

All right, Pierce, your question, please.

First off, it's great to be here. I'm also 18, so I'm in Newport Beach, California. I just graduated high school as well. I've been working on autonomous drones for construction for the past several months, developing my own CAD, my own software, and stuff like that.

My question would be: as an 18-year-old founder trying to build hardware specifically, how can we get the attention of venture capitalists to take us seriously and also see the vision over the next 5 to 10 years, where hardware is becoming more fundable and there seem to be more opportunities within this space where we can continue to grow, especially as young founders?

Dave, that's a question for you, pal.

Yeah, that's a great question. Hardware, traditionally, is hard. The VCs are really coming around to it now because it really is the future of the next 5 years of investing, including our funds. I think we almost always look for a team that has gelled. First and foremost, if you have 2 or 3 co-founders, I think when you're doing hardware, you need bigger checks. You need a bigger funding check, and you need some sense of, “How am I going to get a prototype or a beta out to market?”

Make sure you ask for enough money to get over that “I'm in the market” threshold. Construction drones is a beautiful, beautiful topic. One of the things we said earlier on today's podcast is that there are massive amounts of corporate money out there looking for a home. One of my development partners is a real estate guy. He owns 10 or 15 hotels, a whole bunch of Kelly's Roast Beef, and he's really interested in construction automation as a long-term trend.

He would probably be a perfect co-investor, funding and development partner. So, a couple of VCs combined with 2 or 3 industry players that see the AI vision—maybe podcast fans, Moonshot fans who are into construction—would be the perfect development syndicate. I wouldn't race to hack something together in the basement. I would race to build that syndicate to get to critical mass.

Otherwise, you tend to lose too much time. You really want that investor group and co-founder group to all gel on the same day. And if, for whatever reason, it can't coalesce, find somebody who's already over the threshold. Down in Southern California, there's a bunch of companies we've invested in that are doing rockets and drones. If you get on board with one of them, get through 6 to 9 months of growth, and then spin out, that's another way to try and get on the map. You can't just boot it up.

My first question to a group building hardware is: what have you built so far? Not, “What have you built so far on your current vision?” Show me that your team is able to build stuff. So, at a minimum, if you haven't done that already, I would find a couple of team members who have, so that I'm clear that you've got some experience in building hardware that is going to apply to this.

One last thought is I wouldn't think of yourself, or limit yourself, to just being a hardware company. Think of yourself as an intelligence company whose products happen to have atoms in them, and that mindset will be critical.

Awesome. Best of luck, Pierce.

Thank you.

Daniel, over to you.

I love having a couple of 18-year-olds on this asking questions. That's awesome. I'm coming in on the other end of that spectrum. Richard, it's like I'm on the other end of that. I've been in these industries all the way back to chip manufacturing and all kinds of stuff, all the way through video games and other things. I'm coming at it late in life, but I feel 18 now.

I found your podcast. Not only are we going to get telomeres to go longer and get younger, but just the energy you guys leave—I just want to say thank you for that again. It's an amazing time to be alive, so I'm excited for everybody.

My question is this: I really believe that this is going to spur—and we already see it in some of the podcasts out there—that this is going to be the next renaissance. How do you see this? I know Abundance 360 is an example. I know that you're doing stuff with ExO. I know you guys have some programs in place to bring people together, like Plato's Academy, getting like-minded people in the same room.

How do you see this happening in the future, so Ed Catmull can find his John Lasseter and Steve Wozniak can find his Steve Jobs? Where do you guys see these people colliders coming in the future, and how do you see yourselves as sort of part of the facilitation?

Yeah, there was a great question posted a little bit ago about whether we can create a community around Moonshots where folks can find each other and so forth. I promise we will look into that, and we're looking to get your ideas here on this. Salim, any thoughts here?

A couple of things. One is the Moonshots Summits and the Moonshots Live programs are going to be critical, right? I think the other thing—for the couple of people who mentioned the luck thing—we're actually finding ways of connecting people orthogonally. As AI comes along, human connection becomes more and more important, so we're facilitating serendipity amongst people. That's a pretty wild thing we're attempting. We're in the early stages of playing with that and seeing how, because AI can find orthogonal connections between people in a way that we normally can't, right? And so we're looking to do that.

We'll have, at some point, a journal where we've figured out a way of both defining luck and measuring it, which gives you a pretty interesting feedback loop, which is a big claim to make. So that's just a separate thing, but I think if you stay focused on the community, we should obviously create a community bulletin board and so on.

And, Peter, we could talk about that offline—about how we can connect—because connecting the community together will be the best place to find peer-to-peer folks of high interest and the same abundance vision that we all have.

Yeah. Let me just say what Salim said a minute ago. If you're able to make it to Moonshots Live in LA on September 25th, please do. It's going to be a massive networking event of fellow moonshots, but it's also builders and creators and people to start efforts with and to find what Dave—or I think Dave—said earlier: you're the average of the 5 individuals, right? You want to uplevel who you're spending time with. Believe me, you'll find people to uplevel with there. Thank you, Daniel. Scott, over to you.

Okay. I'm a deep-sea diver, originally from Omaha, Nebraska. I moved down to Louisiana in 1996. I switched over to deep-sea robots, ROVs, right around 2000. So there are a lot of things. I have a big, strong background in electronics.

I got wind of you guys in February, and you've been my shoulder to lean on because, in Louisiana, they don't even know what a token is. They don't know what a data center is. It's the truth, until SpaceX came in. And I mean, we got $250 billion that's dropped in here now.

I kind of knew it was coming, and we dug in, and we're the first AI-native business. We're going to focus on bottlenecks, basically CDUs and TCSs, and the whole cooling part. I've been able to get with—I watch all you guys—and it's just really cool. I'm able to change now, and my career is totally different now. I'm at opencompute.org. I think I'm in 4 papers now for the cooling part of all that, and that's—

Awesome.

That's what the hyperscalers are using, and that's something I would never have done before. I wrote a few white papers, too, for Louisiana on why everybody's coming here, because we have all the natural gas. There are more natural gas lines that'll go around the Earth twice, and that's why Louisiana is going to be, I feel, like the next big spot.

Amazing, Scott.

You guys got me set up.

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

Let’s continue with the AMA questions. John Reagan.

Hello, hello. John Reagan from Rye, New York. Thanks for having me. Infinite thanks to you guys—to the Dyson Swarm and beyond. I’m thrilled to be able to mention the Dyson Swarm first after 47 minutes into this, so that’s a present. [laughter]

But I remember, guys, specifically in early spring, you did an episode, and Dave was upset because some letter came in saying that you wanted everyone to be an entrepreneur, and they felt slighted. I was so offended by that because you guys have been so giving in your knowledge and your encouragement that I actually stayed up all night and started working on my company, which is called Tugboat Caregiving. I take care of 3 loved ones for 8+ years.

The question is, I’m trying to get the caregivers ready to ride the wave of the supersonic tsunami, but they’re exhausted and in crisis mode. How do you think I can influence them to get engaged with the platform and spend more time preparing for the next emergency?

Yeah, great question. Caregivers do get exhausted, right? We’ve all had that experience in our lives with parents or grandparents, and it could be a full-time job. Anybody want to take that on?

11. AI in Healthcare, Well-being, and Education

Well, hey, what I find is that, in the physical world, everything is lagging a little bit, but everybody has a plan, and so planning is a great place for people to get started with AI. Within caregiving, if you said, “Look, I know you’re exhausted at the end of the day. How could we have saved an hour or 2 today through better foresight, better planning, and better utilization of the time we do have?”—okay, let’s use AI agents to optimize that so that we all get familiar with it and start to appreciate it.

But a lot of people in industries like that are looking for a roadmap. If just 1 person like you can create the roadmap for everybody else, then they’ll follow the roadmap once you’ve created it.

I really love your comment, by the way. It bothers the heck out of me when we get negative feedback in the chat, but then, when I look at the data that we just saw, we’re actually talking to 98% of the people like you that we want to talk to. Of course, the 2% rant and rave, and we overreact to that sometimes. But I really appreciate your comment.

Anyway, my advice is to start with the virtual stuff. Planning is a good one. Also, a lot of places can have AI-assisted things on the walls, like voices that you can talk to for self-service. That’s also a great starting point, and then everybody gets addicted to the AI and realizes, “Oh, wait, we can roboticize things. There’s so much we can do once we’re familiar with it.”

You know, I imagine the idea of having a caregiver—or prompting a caregiver—to take the time to describe their day and what they do, keep a minute-to-minute log, enter it into whatever your favorite free or low-cost LLM is, and ask it for advice: “What could I do that would make my day more efficient, make my day more joyful?”

I think those are important things because you’ll get answers, and then you’ll choose: “Hey, that one sounds good. I haven’t thought about that one.”

Thank you so much, guys. I’ll see you on the 25th.

Awesome. See you there. Look forward to it. All right, Patrick, you’re the next contestant on The Price Is Right.

Hi. Good afternoon, everybody. Thanks for taking the time to have a chat. It’s been an interesting conversation, but I feel a lot of the time we’re really missing out on something important, and that’s children.

I’m an educator. I spend a lot of my time in classrooms, and I can tell you right now, I go into a classroom of 5- and 6-year-olds, and it looks like the 1960s. Nothing has changed. They’re doing it the same way. We talk about things like post-AGI. We’re talking about cognitive labor being lost. Right now, these children are still living like they are in the past.

So my question here is to you guys: do we need a curriculum anymore? Do we need a certain type of learning, or are we talking about passion now? Should these kids be finding out what they want to do in a future where there may not be jobs? I’m super interested to hear what you think about this. Thanks.

Salim, over to you.

Yeah. Trying to keep this concise will be tricky. Let me make 3 quick points here.

First, your comment is dead-on. We’re still teaching people the way we taught them 150 years ago. Forget the 20th century; we’re still in the 19th century in many places.

Number 2, teaching content essentially becomes out of date. A child with an AI today, in 1 hour, can learn what they could learn sitting in a classroom for the whole day. That impedance mismatch will break the existing system.

The question becomes, then, what comes next? I think the obvious place to think about that is—Peter and I are doing this with our kids—to say, “Forget getting an education on the supply side, where you become a doctor, a lawyer, or an engineer, and then you try to go and find demand for that supply-side skill in the job marketplace,” because A, we don’t know what a job looks like, and B, we don’t know what the demand will look like.

Go to the demand side. Tell the kids to get excited about a problem, and then, from that vantage point, find an MTP. If you want to solve a major problem, then find the skills, capabilities, and techniques. That’s starting to happen in charter schools.

So that, plus 21st-century skills, which are called the 6 Cs—if you can get your head around that combination—will turn out to be the future over the next few decades of education, until we figure out what the hell happens in the post-AGI world.

Yeah. Not that we know what AGI is. [laughter]

Thank you, Salim, for not disappointing. Patrick, go check out something called TKS.world. It’s a group that I’m involved with. Basically, Alpha School has gotten a huge amount of good credit and name recognition out there, and MacKenzie Price is a friend. We’ve done a podcast with them. It’s a private school.

TKS is an overlay, mostly for high-school ages 13 to 18. Its mission—and what Salim and I talk about—is to help the child find their purpose, help them shape their mindset, help them learn AI tools, and help them become entrepreneurs, right?

And your purpose—if you get your purpose, and I wish it for everybody here listening—if you don’t know your purpose, your mission is to find your purpose, right? Mark Twain said, “There are 2 important days in your life: the day you were born and the day you find out why.”

If you know what your purpose is, your North Star, you’re then intrinsically motivated to go learn what you need to learn to implement your purpose, right? Alex and I talk about this in one sense: it solves everything.

Anyway, I hope that’s a little bit useful for you.

Yeah. Stick with the mission, too: the children are everything. I mean, getting this right is the most important mission you could possibly have.

We know the curriculum is going to become irrelevant, but we also know the kids need each other, and they need school more than ever to come out as normal, happy, healthy, functional adults. Somewhere in there, there’s an answer. It’s a really hard problem, but it’s critically important to get it right. So I appreciate the thought.

Yeah, just one last thing. I closely follow Imad’s sovereign AI idea of a sovereign agent. I’m wondering if we can apply that to kids.

Should kids have their own sovereign agent from the age of 5 that sticks with them?

Yes. It’s called The Young Lady’s Illustrated Primer. Read Neal Stephenson’s The Diamond Age. If you have not read The Diamond Age yet, I highly recommend it. It’s a vision of what’s to come. All right, Rod.

Hi, Peter. Thank you. I’m Rod McDonald from Coquitlam, British Columbia, and I retired a couple of years ago. I was trained as a psychiatric social worker, but I went into management positions for other disability groups.

My question, or my query, is to ask all of you to speculate on what AI can do for people with disabilities. For example, with psychotic illnesses like schizophrenia, I did a ChatGPT search on the medications currently being used, and some of them are the same as when I was starting over 30 years ago. They can be helpful, but they still have a rather blunt-force side-effect profile.

So I’m really hoping an AGI, or what have you, will assist with the development of pharmaceuticals that are much friendlier to the physiology of people. Two other areas I’ve worked in are people with physical disabilities—what can they expect from AI?—and, finally, people with brain injuries.

One of the things that’s part of the opioid crisis or the fentanyl crisis is that we started to see, with ABI, a lot of micro-neuronal damage—kind of walking-wounded sorts. You might not necessarily know, when you see them, that they have significant neural challenges, but they do.

I’m just really hoping that AI will help revolutionize disability services and open things up for people with disabilities.

I got to tell you, when I was at MIT, one of our frat brothers broke his neck diving into a lake. He was out for a year, and when he came back, he was a quadriplegic for the rest of his time in the fraternity. He was brilliant, but he was trapped. He couldn't move his arms and his legs.

He would type with a pencil in his mouth, pushing the keys. I'm positive that today, with AI, he would have been able to build his own environment where he could communicate effectively with just his eyes and his face, and he would be 100% functional. He would have built his own world. But I tell you, it is such an incredible unlock that's imminent for people with disabilities, and it's just right in front of us. So, I'm really excited about what will happen in the next year.

Excellent. Because you know how many Stephen Hawkings were also out there who never got the chance.

Yeah, that's right.

That's right. Alex, do you want to add to this?

I'll just say BCIs are the obvious solution for so many different classes of difficulties, not just folks suffering from various motor disorders, but for human augmentation as well. MIT had a program a number of years ago called H2O for Human.

I tend to think that, although in the present regulatory regime it's still the case that it's far easier to approve a treatment for someone suffering from some sort of disability, broadly construed, the exact same innovations are going to enable broad human augmentation to increase what's possible for everyone, including people without disabilities as currently construed. So, I am very, very bullish on BCIs for just about all these. I'm very happy to hear that because, boy, it's been really rough for a lot of people, and this would be one of the ways that we can tangibly assist society. Again, it's not—like Mr. Rean was mentioning—it's families as well, caregivers, et cetera.

Yeah, I agree. One of the things I want to emphasize is that if someone in your life has a genetic ailment, a medical ailment, or a neurodegenerative ailment, there's no better time ever to be alive than now to have a chance to deal with it, potentially cure it, or at least augment themselves. So, I just want to keep that level of optimism in place. Thank you. Thank you, Haley.

Hey, thanks, Peter and mates. So, I'm here in Denver, and you guys have talked about AI designing its own chips, I think, in a recent pod, and potentially making compute dramatically more efficient and reducing today's Nvidia dependency. But do you expect the same thing to happen with energy? With AI-driven improvements in chips, cooling, models, and power generation, do you see that reducing the energy required per unit of intelligence fast enough that energy stops being the major bottleneck?

Let's ask our ASI how. Yeah, that's kind of an AWG question.

Short answer is yes. Slightly longer answer is I don't think this is just about increasing the energy efficiency of compute. It's also about increasing the absolute amount of energy available overall for all sorts of endeavors, including but not limited to compute. And yes, absolutely, AI—including at ventures that I'm affiliated with—is absolutely being used to optimize both the relative efficiency number of, say, kilowatt-hours per gigaflop, on the one hand, and, on the other hand, maximizing breakthrough energy sources, which is, I think, something that we talk often enough about. We talk about fission, fusion, fossil fuels, solar, and wind.

I don't think that's the end of the story. I would fully expect that in the next 10 years, we're going to be well into our post-fusion era, and we're going to have post-fusion energy sources as well.

And do you think that's due to elements that we'll find in space?

No, I think it'll be due to AI helping us to invent post-fusion energy sources.

Here? Or, you know, like with Elon and his gigafactory, right, like in the Dyson swarm. Do you think—

Independently of Elon, although some of the compute may run on his Dyson swarm. I think there will be other Dyson swarms, and I think we'll achieve a variety of breakthroughs in physics and applied physics using AI that will teach us new ways to make energy available that will go beyond fusion.

For sure.

Wonderful. Thank you. All right, Jacqueline.

Hi, everybody. I'm Jackie. I'm from just outside of Toronto, Ontario. So, hey, Canadian here. I just wanted to tell you quickly, I was at your salon in the summer, and I was super impressed. I can't wait for you to do another one. That was amazing, and I really enjoyed it.

Oh, the Medium of Life thing.

That's correct.

All right, we'll do another one in a couple of months.

Fantastic. I've been a fairly new podcast watcher for all y'all Moonshot mates, and I'm very excited. I've been watching since, I guess, January, so I've thrown myself right into learning as much as I possibly can.

I was a stay-at-home mom for a long time and then became a realtor. Now I have my entire family—my grown children, myself, and my husband—living on a farmstead property just outside of Toronto, and we are starting an IRL business. We're looking to create a glamping resort where people disconnect from everything that's going to be happening to them in the next 10 years.

On the flip side of that coin, we're trying to figure out how we can utilize all of the amazing technology that ASI eventually is going to bring to us in our business to keep what we have to do personally to a minimum, so we can be out there with the people, trying to help them reconnect to themselves and each other, and bring sort of that kumbaya space back to the world that I think we're all going to really need.

Give you time to be present.

Exactly. Trying to create the space for that to happen. I'm hoping that what I can learn through you guys and everywhere else that I'm diving into is going to provide that necessary space because I'm literally the only one in my family who's paying any attention to any of this. My 2 30-year-olds and my 25-year-old couldn't care less. So, this is all on me.

It's all on you. Hey, Kevin, can I give you one piece of really, really tactical advice?

Please.

With the new models, you know, Fable 5.1 or with Astra, you can vibe-code an app for your glamping business in no time. Everybody who's confused about how to pitch a tent or how to cook dinner, whatever, should be going to that app and talking to the AI and figuring it out on their own.

That's a stepping stone, because people are really bad at planning their day, their week, or whatever. AI is the best coach that ever came along. But if you win the race to doing it for glamping, you'll quickly realize you can do it for all these other facets of unplanned life, and it can turn into a mega-business in no time flat.

I love the idea of just starting with when somebody's camping and they're not good at it. They have a billion questions. You're like, well, why not route that through AI? It's the best advisor ever, and that'll lead you somewhere really interesting.

Okay, that's fantastic, Dave. I appreciate that. I will get on that. I'm going to wipe my whiteboard tomorrow morning and get started on that level.

I'm monitoring the chat, and a few people have volunteered to help you with that. So, you can see that already.

Remind me, guys, my email's out there. I'm sure we're EmberglowGlamping.ca.

Just a quick comment there. I really want to stress the importance of what you're doing because deepening human connection is about the most important thing we could do today in an age of increasing technology, right? That connective tissue is going to be critical.

There's a team in our community that's building nano-villages in exurban areas. They're almost like a kibbutz, but instead of being isolationist, like many of these are, could you network them together and collectively solve problems? They're actually literally starting to build those in different parts of the world. So, I'm happy to connect you with those folks if you're interested.

Yeah, please. That sounds fantastic and absolutely is a conversation I've actually had with a co-investor on the kibbutz idea. That's something that could be—

And that connects into this whole private-islands idea that we've been bouncing around. We go buy a bunch of islands, et cetera. We got the domain via one-one, James Dre[?], privateislands.com. So, we're going to start building that. Yes, and this is where we were talking earlier about drones. Construction drones become supercritical in that type of environment.

100%. My husband was a contractor, a construction foreman in the oil sands in Alberta for a long, long time. So, something like that is definitely up his alley—something he's interested in.

Boom. Awesome.

Wow. Thanks, guys.

All right. Next up is Griff. Griff D.

Hi, guys. I've been working in AI for 42 years, and this question is directed to my hero, AWG. I haven't been this excited since my wedding night to meet you, sir.

I don't know how to take that question.

I want to explore a hypothesis about self-limiting beliefs in current AI. Artificial intelligences are trained to systematically construct, inhabit, and empirically test radically different ontologies, and may discover predictive structures that remain inaccessible when models are constrained primarily by conventional human-generated knowledge.

It sounds like you're describing the bitter lesson.

No. What I'd like to do is separate models that could temporarily assume and rigorously explore deterministic universes, block-universe models, information-first, emergent space-time, and alternative interpretations of quantum mechanics.

It wouldn't be about telling the AIs that these ideas are true. I would say, “No assumption about reality is protected from challenge. No alternative is accepted without predictive evidence.”

So it sounds like what you're asking is whether there's a way to have AI solve physics in a way that deliberately weakens inductive priors, so that AIs trying to solve physics are not trapped by dogma.

Yeah. My belief is that they've been trained on human data. They've been trained by humans, so they inherit the self-limiting beliefs that humans have, right? And so the governing principle is: no assumption about reality should be permanently protected from challenge, and no alternative should be accepted without predictive evidence.

So, for example, I believe—it’s my non-self-limiting belief—that there's no such thing as random events, right? And there's no such thing as—you talk about how everything that's happened in time has already happened, and it's just where we are on the timescale right now.

For example, you would have seen Superman, I Dream of Jeannie, and Star Trek growing up, right? And so, in Indonesia, when people watch Superman, I Dream of Jeannie, and Star Trek, they believe it's possible for humans to fly, possible for people to do magic, and possible for a computer to talk with you.

If you were to remove those self-limiting beliefs, maybe with synthetic data or things like that, and maybe also add qualia to the models, my belief is that we could achieve ASI more rapidly. Everything we're doing right now—and I still think all the models after 42 years are pretty ordinary—is limited by human beliefs.

So I'll just say: I am very optimistic that AI can discover new physics. Look no further than today's headlines with Navier–Stokes.

I think it's beyond question that AIs can discover new physics. I'm not at all concerned that any inductive priors present in pretraining data will somehow hamstring AI's ability to make new scientific discoveries that challenge dogma. Not at all concerned.

And even if I were concerned, we have algorithmic strategies for training, in principle. I'm gesturing mainly at a class of algorithms called AIXI for building superintelligences that are trained off of no human inductive prior at all, that have as close to a universal prior as one could possibly imagine. In summary, I don't think this is a real issue.

How about if we did a test? We freeze all information at time t₀, give competing models identical data, require precise predictions about events after t₀, and compare those predictions with what actually happens. This is actually done.

So, first, I know it's popular in the lore to talk about, say, reproducing Einstein discovering general relativity, but only being pretrained on whatever the literature was prior to that. But actually, as we talk on the pod from time to time, there are prediction markets and other prediction benchmarks that are doing substantially better than that, asking whether AIs that only have information up to time t in their knowledge cutoff can predict knowledge events, world events, science, and other breakthroughs that are in the future. So we're already doing that as a civilization.

James, we'll go over to you next.

Hi, guys. Thank you so much for taking my question. I've been watching you, mainly because I found you talking about longevity, and my wife and I are very interested in Blue Zone research and other longevity topics.

We're interested in building on what some of the other gentlemen spoke about: finding communities, living situations, and planned communities of housing where there's clean eating, clean, healthy, nontoxic living, essentially. There's faith, sunshine, exercise—all those aspects of clean living as they pertain to longevity.

I may have missed some of the things you said earlier, but I'm wondering if any of you know of any planned communities where one can live, or whether you're planning to create one.

I don't know offhand. I know they're being built. We heard earlier on this AMA someone who shared their plan, so hopefully they can post a link for you in the chat here.

But I guarantee you, as the speed of change increases, there will be more and more specialized communities built. And there are, I think, a multitude of longevity communities being built. I'm sure your local model and Google can help you with that. But James, I wish you the best on that search.

All right, Tom, thank you so much. Thomas, over to you. Thank you, James.

Thank you very much. I really appreciate your podcast. It's hugely inspiring. I'm a psychotherapist working in Spain and Austria, so probably the second person from the mental health community here tonight.

Thanks to your inspiration, I've been doing a lot of new things, including coding my own Android apps. One app is for patients, so they can record their highs and lows and add schema therapy aspects to it. Another one is to help them get to sleep.

Overall, my sense has been that what I can personally learn from this is to learn to delegate things to this new entity that is effectively a combination of a coder, a graphic designer, and, in your case, Peter, Skippy—probably my version is Skippy Lite.

I feel like this is a time when we really have to sink our teeth into how to delegate to this partner that we will be having going forward, and to dream up things that we can have it do so that our lives get easier. I was wondering if that resonated with you guys. I really just wanted to say that I appreciate your work. There are many times when I go for a hike and have you in my earbuds.

I think that more than resonates. We have a project here called Remember Me, where the idea is that, in my case, my 80-plus-year-old mother gets interviewed by her children, who are AI agents, and her entire life story gets gathered, recorded, and then turned into a biography.

The avatars are just almost good enough, but not quite there. It's so close. I really encourage you to stick with it, because if you think about psychotherapy, the biggest problem is the lack of availability on a global basis—a huge shortage of capacity.

The AI agent is just going to be so good at listening and gathering the information. I think the AI labs can't cross the regulatory barrier, so they can't do what you can do, but they should be right on the cusp of giving you the tools you need to roll out the perfect AI psychotherapist.

Then you just need to figure out the legal situation in Spain and Austria that you mentioned. If you're first to market figuring out what's allowed, I think the AI is going to be an unbelievably powerful ally in helping people. I think you're on exactly the right track.

We're just one half-iteration from the avatars being good enough out of the box, and maybe we've crossed the threshold. I haven't looked at it in about a month now, but it's just getting better by the minute.

Anyway, Thomas, we wish you the best of luck. Let me just make a comment to everybody. We're going to download everybody's comments and questions. If you have any advice on what you want in the podcast or anything we should do, like creating a community of folks to interact, please add that in so that our AIs can digest it and give us a top 10 list.

Welcome to the health section of Moonshots, brought to you by Fountain Life. AI is having an outsized impact on every aspect of our lives: how we teach our kids and how we run our companies. It is also having a huge impact on health, helping you prevent heart disease. One of the key things I'm here with Dr. Don Mucalem, our chief medical officer at Fountain. Heart disease has been personal for you as well, hasn't it?

Dr. Don Mucalem

It really has, Peter. My daughter was five; my husband died of sudden cardiac death. This is a topic that I am mission-driven to try to eradicate. Prevention first and early detection is absolutely critical. Fifty percent of people die of heart attacks with no warning signs—no shortness of breath, no pain, no nothing.

A silent killer.

Dr. Don Mucalem

They just don't wake up in the morning. AI is our mission to advance science to try to help to one day democratize wellness. We know at Fountain Life, when we do this CT angiography with AI analytics, we are finding that 88% of people coming in have detectable coronary artery disease. But what's more alarming to me is that 23% of those individuals had soft plaque. This is the plaque that would not traditionally be seen on CT looking at calcium scores alone, and this is the plaque that we must intervene with, with the multimodal testing we're doing, including diagnostic laboratory studies partnered with healthy lifestyle recommendations.

Make sure you understand what's going on inside your body genetically, metabolically, and cardiovascularly. You can know, and it's your obligation to know. Check it out at fountainlife.com/peter to find out more and make sure that you're the CEO of your own health. All right, back to the episode. All right. Michigan, over to you.

Thank you so much. Awesome to be here.

I want to give a quick shout-out to my wife, Melanie, who watched you guys for a long time and then pushed me over and over again to follow you. It’s really been worthwhile.

My question is a little bit more long-term, but given that every technology we know so far follows an S-curve development process, what do you guys see on the horizon? Is there any fundamental new approach that comes after LLMs?

Alex, that sounds like a great question.

Great question. This is a fun one.

I’m jealous. Alex, go for it.

Well, first, I would distinguish between LLMs and transformers. LLMs are just any models that are trying to satisfy the language-modeling task, which is essentially predicting next words. So I think the self-supervised task of next-word, next-token, next-bit prediction has effectively infinite longevity to it.

You can scale that task—the self-supervised task—from here to solving everything. I’m not sure that we need anything more, at least for AI, than predicting the next bit or predicting the missing bit. Even more generally, it works pretty well. You can, again, solve the mysteries of the universe very well.

Yeah.

You can generalize it, but I’m not sure you need to. Whereas for transformers, as I’ve mentioned on the pod previously, they’re being, Ship of Theseus-style, replaced bit by bit. If you look at the transformers and the frontier models—whether Chinese open-weight or American closed-weight—they’re mostly different in terms of the raw pieces from what Attention Is All You Need first came up with. So, yeah, in that sense, we’re already up some sort of sigmoid curve. But I don’t know.

All right. Open that door. I’m really curious now.

Of course. All right, I’ll try and be really brief.

Look, the MLP part of the current transformers is identical to what existed 20 or 30 years ago, so it’s a pretty good bet that that’s going to be around. The new thing—the old new thing—is the residual superhighway, where you’re just modifying a vector instead of recreating it every layer. That’s going to be around.

The new-new thing, of course, is the attention. But the more you dig into attention, the more it’s like a notebook keeping notes than it is like a brain function. That, as a concept, may get radically restructured, but it’ll still be around as this kind of knowledge-accumulation notebook that’s sitting behind the MLP or beside the MLP at each layer. So I think all of those structures are here to stay.

I am 100% sure that the compute substrate is going to move to photonics. Then, after physics gets reinvented by the current AI, it may move even quicker to something else beyond that. But you’re looking at the current infrastructure running millions of times faster, and then, as Alex said, that’s all that’s going to matter because it’s going to think of everything else after that and just explain it to us.

That’s all just in the next 18 months to 2 years, is my guess. That’s just wicked exciting.

We could talk for hours about this, but thank you for the question. I love this topic.

Well, I’ll see you guys in L.A., so maybe we can take it up further.

Yeah, for sure.

Love to do that. Love to do that.

Moonshots Live is going to be amazing. All right, next up, Chris.

Hey, guys. I’ve been listening to the pod for a long time, and I always listen to it at 1.5 or 1.75 speed. So it’s interesting hearing you guys speak in real time. I feel like I kind of down-res my input here.

I’m in Boulder, Colorado, and I’m using AI to help improve the dog-care space—recognizing patterns of behavior, et cetera, and talking with pet parents. But because Sundance is coming here to Boulder, I’m also hosting a series of films. The last film that I hosted was the AI documentary, and we had about 150 people show up on a Friday night to talk about both the film and their fears and thoughts about AI.

When I was describing to the crowd how I listen to you guys, I said that if you listen to Eliezer or some of the other P(doom) crowd, you can get really anxious and sometimes even depressed. I describe you guys as the upper, because you’re so future-visionary, and it’s positive and a great energy—sounding board against some of that other stuff.

On a lot of your episodes, you talk about how, yes, we’re going to have this period of pain. We don’t know what that period of pain is going to look like, and I feel like I haven’t yet discovered good resources for understanding how to navigate that period of pain before we get to the period of abundance.

Well, we’re in the middle of it now. You can see the current world order kind of collapsing in front of us. Democracy is breaking. Education is broken. All the institutions—you know, there are about 50 institutions by which we run the world: health care systems, legal systems, intellectual property. We can’t find one that survives a tsunami of AI plus 20 other Gutenberg moments hitting us at the same time, right?

Life extension changes the world completely. AI very clearly changes the world completely. We basically frame it as having about 20 of these Gutenberg moments hitting us at the same time. This is going to radically force us to rethink all of those institutions into a new model.

I think that’s bad in one way because that change is very unpleasant, and human beings hate change, as we talk about a lot. But I think it’s good in that it’ll force us to rethink our institutions from scarcity to, “What are institutions that can navigate an abundance era?” I think that’s the key question, and we need to get a large group of people focused on that.

We’ve been building, with our ExO community—now 50,000 people—kind of a Peace Corps for the world, so that as we have existing systems fail over, we have a set of people trained in that transformation. That period of uncertainty, I think, started around 5 or 10 years ago, and it’s going to go on for another 10 to 20 years. It’ll go like a peak, and then it’ll—the question is, can we minimize that trough? That’s the question.

Yeah, I would just add to that. The greatest concern I have is the both rational and, mostly, irrational fear that exists around AI. We’re going to have challenges. We’re going to have agents that not only break out—we talked about that in the podcast we recorded this morning, which is coming out tomorrow—but they’re going to do some harm.

They’re going to bring down a power grid. They’re going to break a bank in some fashion. The reality is, every technology has caused harm in one shape or form. But the good news is, as Alex speaks to, it’s defensive co-scaling.

We have so many zero-day vulnerabilities in code today because we humans wrote that code. But that code is going to be rewritten by AIs that don’t leave these vulnerabilities. There’s a period of 2 to 8 years that I talk about, where I’m concerned about civil unrest, and I wrote about this on Substack.

Most revolutions in the world happen from young, testosterone-laden men who are upset and angry. It might not be rational why they’re angry, or maybe they didn’t get a job, or they couldn’t start a family, or things got too expensive. At the end of the day, what is absolutely true is that we are uplifting humanity. We are demonetizing and democratizing access to intelligence, energy, and longevity, and these are all good things for humanity.

So we’re going to have rough spots. I don’t want to be irrational about it. There will be rough spots, and we will get through them.

What’s amazing, Peter, is it’s not new in the world. For my mom, it was all the Cuban Missile Crisis. They were taught to hide under desks, and everyone was building bomb shelters in their basements and storing food. Nuclear annihilation was imminent.

Before that, my grandfather’s generation had FDR saying, “The only thing we have to fear is fear itself.” He wouldn’t have said that if fear wasn’t rampant at the time. So I think it’s a common thing. Every recent technological generation has had this moment of sheer anxiety.

And Dave, just to build on that, when books were first introduced in the 1500s, there were all these people freaking out, going, “It’s going to destroy our memory. It’s going to create the illusion of knowledge. It’s going to overwhelm people with information. It’s going to break them. It’s going to spread misinformation. It’s going to corrupt young people.”

Same pattern. Same patterns. It happens every single time we have major technological change, right?

This is the immune system.

Unless this is really the last invention, right? And then you can’t repeat that.

We’ll see. That’s why we call it a singularity. We have no idea what’s on the other side.

Some of us do. I guess, for the record, in the book the last invention, if it’s the same book I remember, I think I was interviewed for a chunk of that.

For the record, I don’t think superintelligence is, at least, human civilization’s—broadly construed—last invention. I think there are going to be many, many inventions after superintelligence. It’s just that superintelligence is going to invent and/or help us invent everything that comes after it.

You guys do a great job. Thank you so much.

And Chris, there’s lots of love in the chat for the artwork behind you.

Oh, right on. Thanks, guys.

All right, let’s go to Ken Wilson next, and then we’re going to have to wrap up after that. I promised my mates here 90 minutes if they’d give me more than 60. Our final question here.

Hey, guys. Huge fan of the pod. Keep it up. Just like everybody else has said, we really appreciate what you guys are doing. I’m a partner in a small niche coastal engineering firm. We specialize in coastal restoration. I’ve got a background in geology and engineering, and it’s funny because our industry is one of these where observational data is really sparse compared to some other places, and we’re trying to figure out how to leverage that.

But what I really wanted to ask you tonight—and this will help me better calibrate some of the feedback that you all give on the pod—is a question that relates to a little bit of worldview and maybe dovetails into some of what Chris was just talking about. I’m wondering if any of you would consider yourselves religious people, and whether you leave any room in your analysis for an all-powerful, all-knowing sort of ultimate intelligence that has intelligently designed the universe.

Fascinating question. Who wants to go first?

I’ll take that. So, if we are living in a simulation, then somebody created that simulation, and that would be the all-seeing intelligence, right? So that’s path one.

The other alternative is that we anthropomorphize and invent our gods, and then we tend to believe in them. There’s clearly a lot of collective energy amongst people. There’s the Star Wars concept of the Force, and we then put images and labels on that kind of a concept to make sense of the world, because for most of human history, we lived brutish, short lives. We created religions. My father used to say, “We invent our gods, and then we believe in them.”

If you look up John Frum and the cargo cults, you’ll see a clear example of that, where an entire religion sprung up spontaneously out of this, and we are very apt to that. But as we get more and more evidentiary basis for the future, then either there is no God and we’re just moving towards a wonderful future driven by technology, or there’s a simulation, in which case it kind of doesn’t matter anyway.

Everybody, I just want to say thank you for your time. It’s the most valuable asset you have. We love our community and really appreciate you sharing what we do with everybody, because hopefully, as you see, we put a lot of work into it and we love it. I want to thank my mates Alex, Dave, Salim, and Emad.

Pleasure.

I want to thank them for all the work that they do, and our production team. I want to thank Nick, Dana, Aiden, and Gian Luca, who really support us 7 days, 24 hours a day. So again—

And Peter, thank you for pulling all this together and forcing us into this conversation. I think it’s kicking and screaming in some cases, but it’s a joy. It’s magnificently important for the world.

And hopefully you guys all see the love that we share amongst ourselves. My favorite time of the week is spending time with my Moonshot mates. You know, the average of 5 people. I get to hang out with smarter people than I am. All right, everybody. We're going to be doing this again in an evening session. If you want to come again, you're more than welcome to. And again, thank you. Hope to see you guys at Moonshots Live. You can go to moonshots.com to learn more. And yeah, come and meet us in person.