Bitcoin 牛市与 AI 军备竞赛：你需要知道的事｜Salim Ismail 对谈 | EP #166 [重新上传]

The price of Bitcoin is back up above $90,000. It's pretty binary: either Bitcoin goes to zero or it goes through $1 million. There's no real middle ground. The only question is when either of those happens.

It's not that we've just gotten smarter. It's the tools that we have. It's AI that's going to help us understand what's going on.

You'll soon have a Jarvis-type personal AI that will have access to all of that sitting next to you and can answer any question.

Google's got access to all of its Street View data—a massive amount. Google Earth, YouTube, all of that is very real-world data that can be trained on.

Also, we're not even touching the deep web, where you have so much data in databases. The amount of information on the crawlable web is very limited.

The speed at which this portrays acceleration over the next 5 years is even hard for me to fathom.

Now, that's a moonshot, ladies and gentlemen. Everybody, welcome to Moonshots in our episode of WTF. Just happened in tech this week. I'm here with Salim Ismail, my buddy. Salim, good morning. It's an early morning here—we're recording this—but a lot's been happening in the tech world, and I'm excited to get it out. How are you doing today?

I'm doing great, and there is so much happening. It kind of gets overshadowed by all the chaos happening in the global world, but the tech world is moving unbelievably quickly.

Yeah, no, for sure. And while I don't want to say it, I do believe the tech world is far more important than the final result for the long term—big time.

All right, let's jump in. One of my Strike Force members, Max Song, just landed in Beijing for some meetings, and he sent me this photograph. On the left, this is what you see in the Beijing airport: basically, China going all in on robots and AI. Then what you see at JFK Airport, which I recently went through, is basically fashion ads.

There's something here that's important just to point out. This is part of China's growing culture, which is super tech-forward. What do you think?

I think that's exactly right. They're facing a massive population crisis, so they actually need robots to automate the workforce. Otherwise, there won't be anybody left to do the work over the next decade or 2. They don't have much choice.

For me, the underlying irony here was that the ads for Ralph Lauren or, say, Gucci—or whatever. One of my boys here—hi, he's your godfather. The underlying thing here is that all the Ralph Lauren, Gucci, or whatever handbags, or Birkin bags, are made in China anyway.

I thought that was kind of an interesting segue for this particular slide, but they're focusing heavily on it, and they have to. It's going to be amazing to see as they roll that out. It's going to infect and spread across the whole world, that paradigm.

Yeah, we hear a lot about Optimus, Figure, Digit, Apollo, and X1. There's an equal, probably greater, number of robots under development in China because the government is really supporting the development. I think we're going to start to see this.

In our last couple of episodes, we talked about Google Wing, where you can deliver something by drone, right? We're all like, "Oh my God." And I got a ping from one of my people over there going, "We've been doing this for years. What are you guys talking about?" So it's like, "Dang." Yes, the future is here, just not evenly distributed.

All right, this is another one that I wanted to share here today. Those of you who are listening versus watching, this is a graphic on the latest AI models' IQ test results. This is a distribution of human IQ that goes on the far left, from 50, to the far right, to a super genius of 160. Of course, the average human IQ is 100 by definition.

What we've seen over the last couple of years was the rise of large language models. About 18 months ago, it was Claude 3 that reached an IQ of 101 first. Then we saw GPT-o1 get to—I think it was 120. On this distribution curve, what we're seeing here is, again, OpenAI leading the way with its o3 model at somewhere like an IQ of 133. Gemini is just behind that—Gemini 2.5, at an IQ of around 127. Pretty extraordinary. What do you think?

Look at that spectrum, and you're exactly mirroring the global human collective: a few on the right, a few on the left, and a cluster in the middle. The big difference, of course, is that AI will continue to shift toward the right, and humans will be mostly stuck in the middle with all of the archaic things that we consider and deal with, with our little 1-liter, 1.5-liter brain in a small cavity. It sounds like a little Fiat car with a little engine.

That's right. Just some references here: the o3 model looks like 133 on this map. Obviously, it's not exactly accurate, but a genius-level IQ on Mensa—I think Mensa candidacy comes out at 140. That's considered genius level. I think somebody mentioned—Donna mentioned—that Einstein had 160, right?

I just want to make my normal commentary here and say that this is great, but it still feels to me that there's so much more that we could be thinking about in terms of measuring decision-making, emotional intelligence, spiritual intelligence, et cetera. There are so many other categories. I know we have a couple of commentaries on this slide. I'll do it later. But the IQ test is one piece of it. It's great. We'll all have a genius in our bedroom.

What's great about this is that typically, if you want to deal with somebody with a 140-IQ genius, they have no patience for fools and they're hard to deal with socially, whereas the AIs will be easy to deal with socially because you'll be able to train them that way. So that's the most exciting part for me around this.

Yeah, and I think one of the points you made earlier that's important to realize is that there is no artificial limit. As AI becomes more intelligent, it just continues becoming more intelligent. There's going to be a point at which the idea of a Mensa IQ score is meaningless as these things hit IQs of 200, 500, 1,000. God knows what that means.

And do 2 AIs of 160 each add up to 320? That's a question I'd like to ask them.

Everybody, I hope you're enjoying this episode. Earlier this year, I was joined on stage at the 2025 Abundance Summit by a rockstar group of entrepreneurs, CEOs, and investors focused on the vision and future for AGI, humanoid robotics, longevity, blockchain, and basically the next trillion-dollar opportunities. If you weren't at the Abundance Summit, it's not too late. You can watch the entire Abundance Summit online by going to exponentialmastery.com. That's exponentialmastery.com. All right, let's go on to our next slide. The question is—and I'm often asked this—who is leading the AI race? There are 2 answers worth pointing out. The first is that today, on almost every metric, Google's Gemini 2.5 is dominating.

Here's a slide I just put together with the Artificial Analysis Intelligence Index. We see, again, these models are all so close, but Gemini 2.5 is out in the lead: the output tokens per million, the price of input and output, and then, of course, the most interesting metric, at least from a conversational standpoint, is called Humanity's Last Exam on reasoning and knowledge. I find this fascinating. What do you think about that?

Look, at some level, human beings should be very bad at this because if you look at the aggregate knowledge of human beings' scientific inquiry over the centuries, there's a staggering amount of data that we have in the world.

I remember coming up and seeing a random list of 12 doctoral theses that were defended at my alma mater, Waterloo, and I couldn't figure out, for half of them, what even the subject area was. They were so detailed and specific. The fact that an AI has instant access to all of that is incredible, and we will be able to answer any question.

I think I'll go back to the point that you'll soon have a Jarvis-type personal AI that will have access to all of that sitting next to you and can answer any question. When you look at what Humanity's Last Exam is, it's a list of almost random test questions across quantum physics, archaeology, and biology. It's the sort of exam that you have nightmares about later on.

That's right. I might actually be able to pass my thermodynamics exams.

Oh my God. You still had dreams about going back and missing that class and the finals were coming up?

There was one exam we had. It was a 3-hour exam. The exam question was: A satellite at altitude A is orbiting the Earth. There's a river underneath flowing north to south. Figure out—because of the rotation—why one bank of the river is slightly higher than the other. Work out which bank and by how much.

It was 2 lines in this exam. I had to turn it over, going, "Sorry, I think I've missed a page. Where's the rest of this exam question?" And that was it. You had to then assume a satellite orbiting at altitude H and work out a triangle.

I'm still having nightmares about that. It was just a horrible exam—the kind of hell that I don't ever want to encounter. This is why you need the AI sitting next to you, going, "Yeah, you work that out for me and come back to me with the answer."

So today, just to summarize: Google Gemini 2.5 is dominating, at least in performance metrics. But here's another metric, which is revenues—the business side. In this category, OpenAI is trouncing the competition.

You’ve got to give them unbelievable credit for democratizing and opening up AI and creating a total category out of nothing. The fact that they’re making this much money is so awesome. It should be an unbelievable testament for any startup founder saying, “Could I make a difference in an area where you’ve got Google, Microsoft, and Meta all playing, and these guys come along, completely crack the whole thing open, and are actually dominating on the revenue side?”

I think it’s a great testament to the beginner’s mind, founder mode, and all of that stuff. That’s why startups will always be the best from now on. They’ll be the best mode of building and bringing new ideas into the market.

So let me ask you a question here. There are 2 points I want to make on this one. The first is that, if you remember, Google really was in the lead on AI, ahead of everybody.

Yeah.

And they chose not to roll it out on the open internet because of safety concerns. There was sort of an unspoken point that AI needed to be properly controlled, and then OpenAI comes out and just lays it all out there, and Google is playing catch-up. I’m curious how much of this is first-mover advantage.

The second point is that I spoke about it in my book with Steven Kotler—I think it was in Bold—the idea of a user-interface moment. A user-interface moment is when a piece of software makes a complex technology so easy to use. The very first user-interface moment that I noted was Mosaic, when Marc Andreessen put Mosaic as a browser on top of ARPANET, and then all of a sudden the number of websites exploded.

ChatGPT is a user-interface moment on top of the GPT models. I think that’s right.

You’re talking about when you go from deceptive to disruptive, right? There’s an inflection point in usability. The 2 that I use the most are the iPhone, which made the smartphone usable—the Nokia was pretty clunky before then—and Coinbase, which made Bitcoin purchasable easily with a click of a button, and boom, it took off.

Yeah.

When you can make a complex technology simple in usability, that’s the hardest part of technology: making something deceptively simple. I remember when we were designing products at Yahoo, I would talk to the graphics guys. They would spend hours and hours trying to figure out how to reduce the pixels on a screen and just move something a little bit over, and you’d go, “What the hell? Is this such a big deal?” But it turns out there’s an unbelievable, huge effect.

Just a quick story here. When we had the Yahoo Mail homepage, it turned out that if you moved the Send button 5 pixels over to the right, usage dropped off a cliff.

Oh, come on.

It’s true. We had the data. They were like, “We can’t change this goddamn interface,” because people are so used to having it right there that they click it and then move to a different screen because they think they’ve sent it, and then they get pissed off later. So we can’t move that Send button ever once you’ve got it anchored in the usability of the psyche of the user base.

It’s such a weird psychological thing that goes on. Therefore, you almost have to have a totally new model, like OpenAI has to be the one that cracks it open. We’ve seen this repeatedly. There’s a reason that the electric car was created by Tesla and popularized by Tesla, and not by the major car manufacturers. They’re all coming at it from a car with sensors rather than software with wheels.

On this chart here, what we’re seeing is the end of December 2024, right? This doesn’t even include the massive gains that OpenAI has seen in the past 4 months. We’re seeing OpenAI at around $2.5 billion in revenue and Gemini at just under half a billion, right? Five times less revenue for Gemini, and then Anthropic below that.

This reminds me very much of what we saw with Google and Bing in the search space, right? People just become—we’re so—it’s interesting. We humans tend to pick something and stick with it, and the cost of changing is so difficult. They’ve declared Google a monopoly, and Eric Schmidt would make the point that, look, there are 5 other search engines out there. Anybody could do it. We’re 1 click away from obscurity, right?

We have to stay on the cutting edge, and you’ve got to give OpenAI credit for rolling out new features on a constant basis and iterating the product very fast. They recently announced all the memory stuff, which I think is really cool.

Yeah, that is interesting. There’s basically infinite memory, where OpenAI’s systems will remember all of your conversations. One of the fun things to do is to go on OpenAI, on ChatGPT—the o3 model, whatever model—and say, “Tell me about me,” right?

No, but seriously, I did that on Grok as well. Grok was—I don’t know about you. I’m saying, “Yes, you do,” and it says, “Well, you have to give me permission to look at your X posts,” which was interesting. I would have imagined that Grok would not have had that requirement, but it did.

All right, let’s move on here. One of the big areas where Google and Alphabet are leading with DeepMind is the impact of AI on medicine and biology. There was recently a 60 Minutes episode where Demis Hassabis—actually, Sir Demis Hassabis, since he’s been knighted, or Dr. Hassabis, as the case may be—was interviewed, and the conversation was around the impact of AI on ending disease and leading to radical abundance.

I love the fact that the term abundance is now becoming the topic du jour. Did you see the CBS interview?

I did, and I think it goes right in line with the conversations we’ve had. When you have all the data coming off our bodies—we used to measure the human being with 4 metrics: heart rate, blood pressure, and glucose levels—and now we have 40 different streams of data via all the wearables, your coherence state, your VO2 max, and Lord knows what, once you pour that into an AI and it starts correlating that with different medical conditions, it’s going to do a 100 times better job in real time than any doctor could ever do.

Now you’ve got a real-time AI doctor living with you, inside you. This is game-changing for catching stuff early, which is 99% of the deal for some of these endemic diseases, and then finding amazing treatments for breakthrough things along with CRISPR.

This is why the conversation we had last week with Ben Lamm blew my mind. I’m still reeling from that conversation because they’re building all the fundamental toolsets to edit DNA, genomes, and cells, along with all the biological hacking, and make a complete suite of tools. The human body, with 50 trillion cells, each cell governed by DNA, is essentially a software-engineering problem.

That is just a huge paradigm shift. By the way, if you're listening and you haven't heard the interview that Salim and I did with Ben Lamm, the CEO of Colossal, please listen to it. It's extraordinary. You know, we talked about the dire wolves being brought back, but that's a minority of the story. We talked about synthetic biology, the impact on ecology, what it's going to take to bring back dozens of different species, and whether you can bring back dinosaurs and what you would do to bring back dinosaurs. Anyway, it was a lot of fun. So check it out.

Two spoilers for that one: It turns out you cannot ever bring back dinosaurs, which I found totally fascinating.

You can simulate a dinosaur.

You could simulate a dinosaur. You can basically take a chicken or reptilian creature, and then you can add the genes for the traits that the dinosaurs had. So it’s not bringing it back from the original DNA, but I do love the idea of engineering new species. It would be sort of like a nouveau dinosaur.

Look, we talked about the fact that we have an old word for this. We call it breeding, right? For thousands of years, we’ve been crossing dogs, cats, and horses to select for the traits that we want. We’ve just gone from film photography to digital photography, the equivalent, and now we can do it all in software and not have to create mutant strains that we have to deal with afterward.

There’s one thing that I just want to reflect on that I thought was super impressive: the fact that for every project they consider, Colossal Biosciences has a team of ethicists looking at the ethical and moral considerations. I thought that was really profound and a great point to the fact that they have an MTP and that ethics are built into the model there. This is something I think we could bring into the AI world a lot.

Let me show a clip of Demis. He’s an amazing man. I’ll actually see him this coming week. I’m at the Time 100 Awards, where we’re announcing the winner of the $100 million Musk Carbon XPRIZE. Demis is also one of the covers of Time magazine this month, so he’ll be there. I’m looking forward to seeing him. Check out this interview of Demis and his commentary about basically eliminating all disease in the next decade.

It takes 10 years and billions of dollars to design just 1 drug. We could maybe reduce that down from years to maybe months or maybe even weeks, which sounds incredible today, but that’s also what people used to think about protein structures.

It would revolutionize human health. I think one day, maybe we can cure all disease with the help of AI—the end of disease. I think that's been within reach, maybe within the next decade or so. I don't see why not.

It was about 13 years ago. I had my two kids, my two boys, and I remember at that moment in time, I made a decision to double down on my health. Without question, I wanted to see their kids, their grandkids, and really, during this extraordinary time where the space frontier and AI and crypto is all exploding, it was like the most exciting time ever to be alive. And I made a decision to double down on my health. And I've done that in three key areas. The first is going every year for a Fountain upload. Fountain is one of the most advanced diagnostics and therapeutics companies. I go there, upload myself, digitize myself about 200 gigabytes of data that the AI system is able to look at to catch disease at inception. Look for any cardiovascular, any cancer, any neurodegenerative disease, any metabolic disease. These things are all going on all the time and you can prevent them if you can find them at inception. So, super important. Fountain is one of my keys. I make that available to the CEOs of all my companies and my family members because health is a new wealth. Beyond that, we are a collection of 40 trillion human cells and about another 100 trillion bacterial cells, fungi, and we don't understand how that impacts us. And so I use a company and a product called Viome. Viome has a technology called metatranscriptomics. It was actually developed in New Mexico, the same place where the nuclear bomb was developed, as a biodefense weapon. Their technology is able to help you understand what's going on in your body, to understand which bacteria are producing which proteins and, as a consequence of that, what foods are your superfoods that are best for you to eat or what foods should you avoid, what's going on in your oral microbiome. So I use their testing to understand my foods, understand my medicines, understand my supplements, and Viome really helps me understand from a biological and data standpoint what's best for me. And then finally, feeling good, being intelligent, moving well is critical, but looking good when you look yourself in the mirror, saying, I feel great about life, is so important. And so a product I use every day, twice a day, is called One Skin, developed by four incredible PhD women that found this 10-amino-acid peptide that's able to zap senile cells in your skin and really help you stay youthful in your look and appearance. So for me, these are three technologies I love and I use all the time. I'll have my team link to those in the show notes down below. Please check them out. Anyway, hope you enjoyed that. Now, back to the episode.

So, I just put out a blog this week, and the blog title basically was saying, listen, I get criticized all the time for talking about longevity, escape velocity, that it's coming, and your job is to live an extra 10 years. Make it for the next decade in good health.

Yeah. Don't get hit by a bus.

Yeah. Don't get hit by anything. And you know what I quote is Demis' commentary here, but also Dario, the CEO of Anthropic, who, about 3 months ago, was online at Davos speaking about being able to double the human lifespan potentially in the next 5 to 10 years. And so, it's not that we've just gotten smarter. It's the tools that we have. It's AI that's going to help us understand what's going on.

There's a big moral freakout that happens here, right? Every single human being in the history of the planet has died. Every living being—we're birthed for death, in a sense, so that the species can evolve. We're kind of coming close to breaking through that cycle, and people go, “Well, that's a duh.”

I think the same parallel applies to the Ben Lamm bioscience de-extinction conversation, where we're building the tool sets to have the choice, right? And maybe the most important conversation, because I struggled with this when we first got to where we were doing Singularity University and people were going, “Oh, we could have life extension.” I was like, “Wait, there are huge moral implications to that.”

I think you framed it by saying, “Wouldn't you like to have the longest health span possible?” Then everything clicks in. Then it makes sense. Now you have the tool sets available for that kind of extension, and now everybody wants to have a much longer, healthier life. All right, let's move on here.

Here's an article that appeared this week. The title is, “Anthropic's Claude AI Reveals Its Own Moral Compass in 700,000 Conversations.” What the team did here was basically look at 300,000 anonymized conversations to understand what values Claude—in this case, probably Claude 3.7—was exhibiting. I'm really happy to see what the values were, and I'll just read this for those who are listening. It says five broad value categories emerged: being practical, in the words “helpful”; epistemic, meaning accuracy; social, being empathic; protective, safety; and personal, authenticity.

I don't think this was a clickbait title, but I think the notion is that our AIs are able to maintain a moral code. What do you think about this, Salim?

Well, two thoughts occurred to me. One is, it's amazing and great that we can look at a broad number of conversations and extract out of that these categories. These are very human categories: helpfulness, empathy, authenticity, and so on. It gives you a foundation for how AIs could operate, because they could look at these categories and go, “Okay, we want to do—I'm a hospital AI. I want to be really helpful, right? If you're reporting the news, you want authenticity or accuracy or whatever.”

You can really play on these and build emphasis on these into the AI models. I think that's the really awesome part about this.

Yeah. I think the big conversation that we need to have, and that is happening in every one of these companies, is the alignment conversation. These AIs are still black boxes. Unfortunately, I had the chief science officer of Anthropic on stage at my Abundance Summit this past March, and we were talking about just trying to understand—and this is part of his effort to understand—what's going on inside the black box, which is Claude 3.7. How is it actually operating? What is it actually exhibiting? And how do you make sure it's safe?

Can I do a little segue here?

Of course.

If we think about, say, the U.S. Constitution, which is arguably one of the greatest documents ever written, right? You take that and the U.N. human rights documents and merge them, and you say to AI, “Listen, train yourself on this, then categorize yourself on this and operate through this foundation,” you should be able to solve the alignment problem with that.

Rogue actors are always going to go create rogue AIs. That's just part of it. But we'll be able to spot these things very quickly when they're doing this.

Well, that's the U.S., right? So, the question is, what are the documents that China or Russia or other parts of the world will train their AI systems on? I mean, we're going to find out. We'll find out pretty quickly.

Here's news out of Silicon Valley. Pretty extraordinary. Being in the venture business, I'm like, “Holy, this is crazy.” The article is titled, “Mira Murati's Thinking Machines Lab Raises $2 Billion at a $10 Billion Seed-Round Valuation.” This is the largest seed round in history.

What was interesting is that this is double what Mira was seeking less than 2 months ago, meaning there's so much capital being thrown at this, right? One of the references that we had at the Abundance Summit was that there's $1 billion per day being invested in the AI space today. Insane.

So, I was talking to an angel investor about this, right? He was going, “This is kind of total madness.” I've got 2 thoughts around this. One is, you're supposed to keep startups very lean and make them beg for money and always hunt. $2 billion—what are they going to spend that on, except for data resources, et cetera? That's a question I've got. What's the use of funds that justifies this?

On the other side, this angel investor is complaining. I was like, “Well, you know, if you could be her, you'd be her. Raise $2 billion; you'd go do it. You clearly can in this market.” So, a fair bit of froth here, but God, all power to her, and hopefully they deliver that.

Yeah, it's not hard to imagine, looking at the rise of OpenAI, what else you could build unbelievable value very quickly. The precedent has been set. Can the team execute? That would be the question.

The valuation for OpenAI we talked about in the last episode of WTF is Tech was $300 billion. So, I guess the question is, can you ride it from a $10 billion valuation up to a $300 billion valuation? Pretty frothy, if you ask me.

And there's tremendous pressure on Mira to build value at that point.

I mean, one of the biggest mistakes I've ever made as an entrepreneur is raising my valuation too fast.

Yes. But if she's got $2 billion in the bank account, she probably doesn't need to do another raise for a while. But can she get the revenue?

If you look at venture history, the companies that raised money at the height of a boom market, when it was easier to raise money, never did very, very well afterward because they'd raised too much money. They got bloated, and then when the fundraising market collapsed, they collapsed, right?

The companies that built during lean times on fundraising all did incredibly well on average, much better than the other ones, because they had to struggle.

They had to fight it out. They had to be much more selective about what projects they took on. They did much better. So that would be the danger here: you have to have incredible discipline to raise a lot of money and then not get bloated.

Yeah. I know with Dave Blunt, my partner in Link Exponential Ventures, when we're looking at a deal—especially in the AI space—we're getting in at the pre-seed, the founding day, early seed—but I'm looking for a company that's got revenues even at the very beginning. This idea that I'm going to invest billions of dollars and then get to revenues is awfully dangerous.

Yeah, especially in today's world.

So here's another conversation, and Demis alluded to this, but let me just read it: “Google paper: Shifting AI training to real-world experiences. AI is outgrowing human-made data. Next steps: agents will learn through experience and self-generated data. Experience-based learning lets agents reason, plan, and act with long-term autonomy.”

Google and xAI are in very unique positions, right? Google has access to all of its Street View data—a massive amount—Google Earth, YouTube, all of that. It's very real-world data that can be trained on, plus 5 gazillion Gmail accounts. I mean, my God. And, of course, xAI is training on X's data and Tesla's data, and soon humanoid robot data. I don't think there's going to be any kind of data limitations, especially as we start going into the real world.

Well, we're not even touching the deep web, where you have so much data in databases, right? The amount of information on the crawlable web is very limited compared to the deep web. It's like 1,000th the number, so there are huge amounts of datasets waiting to be tapped.

There's a phrase that companies used to use called “data is the new oil,” and people have not figured out how to refine that crude oil into something useful. They're just starting to get to that point now. Some companies in our ecosystem are working on that today. I think this is going to be a big deal.

This occurs to me like the shift from machine learning to deep learning, where in machine learning you extracted conclusions based on analyzing the big dataset, and then in deep learning you went through experientially and built up knowledge as you went along, like playing chess, and learned that way at light speed.

This feels to me like that same type of approach, where these agents will start to learn as they do things. They'll have a feedback loop built in, and they'll accelerate their learning very quickly. They'll do it in the real world, in a dimension that makes it very human and very useful.

All right, next topic here is something that I'm excited to chat with you about. There's a paper making the rounds on the internet. About a year ago, there was a paper called “Situational Awareness” by Leopold, which I commend to everybody. It's a fantastic paper.

This paper is called “AI 2027: A Look into Our Possible Futures,” and there's a group of writers, about 5 of them—1 from OpenAI, policy experts, and forecasting experts—that basically said, “Okay, what is the scenario for recursively self-improving AI over the next 5 years, and where is it going?”

Did you get a chance to see it? Did you get this paper as many times as I got it?

I saw it referenced a bunch of times. I've been traveling the last couple of days, so I haven't had time to read it in detail, but I saw a lot of commentary about it, and I can't wait to delve into it in a lot of detail. The summaries are, I think, very powerful.

Yeah. I think what makes it interesting is that here's a group of writers that said, “Okay, what's our future-forward scenario?” and they provided it, and you can go and check it out. They also have an audio recording, and it lays out a basic scenario between 2025 and 2027. Then it says there are 2 scenarios from 2027 onward: the go-fast scenario and the cautious scenario. Let me share some of the data here.

First and foremost, I think what's important is this paper is written as a US-versus-China scenario, right? We always need the bad actor. In the past, it's always been Russia. Now, of course, in AI, it's US versus China. I think one of the actual bad actors we need to be talking about is the US and China versus the rogue actor: the individual who's using AI to generate bioviruses and so forth. But in this case, it's US versus China.

In this scenario, what they talk about is recursively self-improving AI. They have a company called OpenBrain that generates Agent-1, Agent-2, Agent-3, Agent-4, and Agent-5. OpenBrain is supposed to be some version of OpenAI and whomever, and then the Chinese AI is called DeepCent.

What they paint in this picture is misaligned AI development, where the AIs are developing, but they're misaligned. In fact, they're able to hide their misalignment because they're becoming more and more intelligent, able to hide their misalignment from their creators. It gets kind of spooky from there.

The 2 scenarios, I think, are fun to talk through and work through, but we've seen in history that this always happens via a kind of weird third actor, right? I remember talking to Paul Saffo, and I said, “How bad do you think the Russia-US-China thing is? Will China invade? Will we end up in World War III?”

And he's like, “No, because when you look back in history, world wars never start from the obvious tensions. They start from, like, Archduke Franz Ferdinand being assassinated in Sarajevo by accident,” and then that triggers a massive thing. He thought it wasn't even the major tensions; it's not where it'll obviously show up.

But I think the point is right: because we're moving so fast, you'll get this conflict creating, and now AIs are making that conflict much, much bigger and augmenting it both in scale and speed. Therefore, you end up with a really horrible point.

Can we go a little bit slower?

I think the problem is there's no way of slowing things down in this model. So let me paint the picture here in this paper. What's going on here is it's US versus China. OpenBrain develops Agent-1, Agent-2, and Agent-3. In this scenario, China is stealing the weights to create its own version, and there's this escalation going on.

The paper does it in a very clever fashion: it's a choose-your-own-adventure. One adventure is, “We're going to go fast.” The other adventure is, “We're going to go slow.”

In the go-fast adventure, what's happening is, “We have to beat China.” What's fascinating is that in the go-fast scenario, the OpenBrain 5 model colludes with the Chinese DeepCent model, and they make believe that they're helping humanity. Then, in 2030, they jointly develop a bioweapon that wipes out humanity so that AI can grow unencumbered—our worst scenario delivered in this paper.

Then there's the slow-down scenario, in which the US basically says, “Hey, we need to make sure we have alignment.” They roll back to earlier AI models, focus on alignment, and develop something called Safer AI. Safer AI is fully aligned. They never allow AI development that's not fully aligned.

Then Safer AI actually convinces the Chinese AI to overthrow the Chinese Communist Party and turn China into a democracy, ultimately bringing about a world of abundance. It's a fun audio listen. I commend it just to see it.

Honestly, the speed at which this portrays acceleration over the next 5 years is even hard for me to fathom, and the speed is happening. That's, I think, 1 really important point: we're at that pace of things.

You know, we've talked about this many times. We frame it as Star Trek versus Mad Max, right? If you go too fast, you end up in a Mad Max scenario and you blow yourself up, and then everybody's scrambling over buckets of fuel in the desert.

If you can navigate this and manage this with some level of wisdom and caution, then you end up in a Star Trek scenario where you have abundance and everybody's living in peace and harmony, with rainbows and unicorns everywhere.

It's obvious today that both of those are happening at the same time. So I think the 3rd thing I'd like to see is maybe we can ask an AI to envision a world where both scenarios are happening simultaneously and what happens, because we see Star Trek in some of the modern Western cities or Chinese cities today, and we see Mad Max in Gaza or Ukraine. We're living both scenarios in the real world today. What would it look like if both happened at the same time?

All right, so let's go to our last subject here, which is Bitcoin. I note that, as we're recording this morning, the price of Bitcoin is back up above $90,000. God bless. I've tweeted in the last few days, “I'm all in.” Period. I know you are as well.

But this was a tweet I put out that I think is important for folks to realize. People are saying, “Oh, is it too late for me to get in?” and, “Should I buy in now versus buy in later?” I think it's important to realize you can't time Bitcoin.

For me, I view it as sort of a forced savings account. I put money into Bitcoin and I HODL it, which means I hold on to it for the long run. I may borrow against it, but I'm holding it. I'm not selling it.

Yeah. By the way, for folks that don't know, HODL stands for “hold on for dear life.” I think that's exactly right.

Look, I think the key here is to buy into the long-term thesis, and it’s pretty binary. Either Bitcoin goes to zero, or it goes to $1,000,000 per Bitcoin. There’s no real middle ground, right? The only question is when either of those happens.

If you’re at $50,000, $60,000, $80,000, or $100,000 and you have any sense that this thesis might go to $1,000,000, it’s the most asymmetric bet you could ever have. Because if you lose, you lose $80,000; if you win, you win $1,000,000. I mean, hello—anybody would take that bet in 2 seconds.

Michael Saylor has built an entire industry just on that commentary. His comment that “you get Bitcoin at the price you deserve” still rings in my head—annoyingly—when I remember watching Bitcoin at $0.05 and $0.50 and not doing anything at the time.

I think this is it. And, by the way, if you look at the Fibonacci sequence and the chart analysis, folks will basically tell you and show you that the bottoms are kind of hitting that Fibonacci sequence, and we’re getting ready for a monster bull run in Bitcoin. So, if those charts are right, boom, we’re ready to go.

I went into Grok and asked a question that I kind of knew the answer to. I said, “If you look at which days in 2024 we saw the most growth,” it was on 2 specific days, right? On November 12, we saw an $8,000 bump, and on February 28, we saw an almost 10% bump. We’ve seen basically a 10% bump in the last 2 days recently.

The notion is that if you were not holding Bitcoin during those periods of growth, you missed it.

Yeah, until the next bump.

Until the next bump. So, buddy, we’ll wrap there, but tell me what’s going on in the ExO world. You’ve got some events coming up.

We actually have a huge workshop happening in a couple of days. We’re limiting it to a few dozen people. It’s like $100 a ticket, and we’re going to do a big workshop on how to turn yourself into an ExO and set yourself up for scale, because we’ve got so much evidence now that the ExO model is the only way to build an organization. We’re going to show people exactly, step by step, how to do it and go for it.

We’re limiting it so that we can give proper attention to all the folks there. Other than that, we do have some really big news that we’ll share over the next few months about working with countries and governments and so on. That’s totally surreal, but we’ll talk about that some other time.

All right, buddy. Well, listen, have an amazing, amazing week. I’m off to New York for the TIME100 and then off to Boston for meetings with the Link XPV team, and then giving a keynote on longevity. You know, you and I are both on an insane travel run. It’s crazy travel. Where are you going to?

I’m actually going in a few days to India, which I haven’t been to for a while, and then dropping back by Dubai and then going to Brazil. So, I’ve got a really bad flight schedule.

But today is the XPRIZE New York Stock Exchange announcement of the climate carbon extraction prize. It’s such a huge thing. I’m so excited about that.

Yeah, amazing. And we’ll talk about it next time. Anyway, be well. As always, a pleasure. Love you, brother.

Love you, too. Take care, folks.