Elon Musk：Optimus 3 即将到来、递归自我改进已在发生，以及奇点 | #239
- Musk称，AI已经进入“硬起飞”阶段，人类正逐步退出递归改进循环。 每一代模型已经在很大程度上由前一代构建；完全自动化“可能在今年年底实现，最晚不超过明年”，但目前还没有做到完全无人参与。
- Grok 4.20“非常、非常好”，按部分指标是预测能力最强的模型——这“可以说是衡量智能的最佳指标”；但xAI“目前在编程方面落后”。 Musk称，xAI或许能在今年年中追上并超过竞争对手，同时提醒，AI进步遵循相互重叠的S曲线，而不是一条不间断的指数曲线。
- 在没有第三次世界大战、且当前趋势延续的情况下，Musk称，未来约10年经济规模扩大到当前的10倍以上，是一个“相当有把握的预测”。 他还预计，未来10年内将出现月球基地、火星上的人类以及月球质量投射器。
- Musk预计Optimus 3今夏开始生产，初期速度非常慢，明年夏季前后达到大规模量产。 Optimus 4的设计应在明年完成；Tesla正规划一座约1000万平方英尺的Optimus工厂，并计划每年升级机器人设计。
- 自动化并不意味着Tesla计划裁员：Musk预计员工总数会上升，同时人均产出将“高得惊人”。 Tesla目前约有15万名员工，其中约2/3以某种形式在工厂工作；供应商端可能还有100万至200万人，因此近期更直接的经营杠杆是提升劳动生产率。
- Musk认为，AI和机器人让商品与服务供给增长远超货币供应，从而带来通缩和全民高收入的美好结局，概率约为80%或更高。 他提到，在转型前的2–5年内，可能会经历一段社会动荡期，期间还可能发放类似COVID时期的现金补贴。最终货币或许不再重要：机器经济会看重“功率和质量、瓦数和吨位”，而智能人形机器人可以让每个人获得比地球上最富有者更好的医疗服务。
1. 递归自我改进正接近无人参与的循环
- Musk对完整递归自我改进设定的门槛，是彻底移除人类参与。每一代模型已经在很大程度上由前一代构建，人类正“逐渐越来越少地处于循环之中”；完全自动化“可能在今年年底实现，最晚不超过明年”。
- 当被问及这一转变是否会带来硬起飞时，Musk给出了明确回答：“我们现在就在硬起飞阶段。就是现在。”如今他每天入睡和醒来时，都会面对接连出现的重大AI突破。
- Grok 4.20“非常、非常好”；Musk称，按部分指标，Grok的预测能力最强，而这“可以说是衡量智能的最佳指标”。他也坦承，xAI“目前在编程方面落后”，目标是在年中左右追上并超过竞争对手；他表示，按理说今年年中应该能够做到。
- 他对预测的提醒是结构性的：AI进步会经历相互重叠的S曲线——突破、指数增长、线性进展，随后是对数式回报。当被问及现代机构能否跟上时，他回答，奇点难以预测，因为很难知道其中会发生什么。
2. 智能最终会变成能源与空间产业
- Diamandis追问了SpaceX/xAI数据中心的建设时间表和算力规模，但Musk提到SpaceX当前处于静默期，拒绝披露任何可能带来麻烦的信息。
- Musk的尺度模型从能源开始：如果一个经济体消耗的电力达到地球当前水平的100万倍，它仍然只利用了太阳输出的约100万分之一。他认为，这种级别的智能几乎无法被人类理解，并且会“解决你能想到的一切问题”。
- 地球截获的太阳能仅约为太阳能量的20亿分之一，Musk据此认为，地球本身对智能发展存在上限。因此，他预计太阳系级智能将比地球级智能高出多个数量级。
3. 10倍经济规模伴随月球和火星基础设施
- Musk预计，在当前趋势延续、且第三次世界大战没有“让这一前景出现波折”的情况下，经济将在约10年后达到当前规模的10倍以上。他将这一有条件的预测称为“相当有把握”。
- 在同一个10年周期内，他预计月球基地、火星上的人类以及月球质量投射器都将出现；按Diamandis的表述，这会让Gerard K. O’Neill的设想更接近现实。
4. Optimus先缓慢放量，再把生产率推到惊人水平
- Optimus 3已进入最后收尾阶段，Musk称它将“远远领先于全球其他机器人”。他预计今夏启动生产，制造业S曲线初期会非常缓慢，明年夏季前后进入高产量阶段；Optimus 4的设计应在明年完成。
- Tesla规划中的Optimus工厂面积约1000万平方英尺，将采用新的工厂设计，不过Musk补充称，目前与其他工厂仍没有本质区别。他的目标是每年发布并升级机器人设计。
- Diamandis关于“机器人制造机器人”的设想有一个重要限定：Tesla仍高度依赖人力。Musk称，约10万人负责制造产品，或管理制造人员；Tesla员工总数约15万人，其中约2/3与工厂相关，供应商端可能还有100万至200万人。Tesla计划增加而不是削减员工总数，同时让每名员工的产出“高得惊人”。
5. 充裕供给推动通缩，但结局并无保证
- Musk认为美好未来出现的概率约为80%，“也许更高”，但同时警告不能自满，因为可能结果的区间也包括糟糕结局。他现在的态度是保持乐观，同时尽力推动事情朝好的方向发展。
- 在讨论通往全民高收入的路径时，Diamandis回顾了他们此前提到的2–5年社会动荡时间窗口，以及过渡期内发放类似COVID时期现金补贴的可能性，之后才会进入去货币化和通缩阶段。Musk的逻辑是，AI和机器人会推动商品与服务供给增长远超货币供应，经济实际上需要把钱发到人们手中，进而产生通缩。
- 当经济达到今天的1000倍时，Musk认为，人们大概已经把所有能够表达出来的需求都满足了。货币最终“将不再重要”。他设想的是一个类似Ian Banks笔下Culture的未来：AI不再理会人类货币，而是看重物理投入——“功率和质量、瓦数和吨位”（“power and mass, wattage and tonnage”）。
- Musk通过医疗服务让“充裕供给”的主张变得具体：灵巧且具备智能的人形机器人，能够让每个人获得比地球上最富有者更好的医疗照护。他举的是自己的经历——做过3次颈部手术，因为前两次都做错了；他要求的验证很简单：“AI能不能解决背痛？”（“Can AI please solve back pain?”）
Peter H. Diamandis, MD
Audience, as you can see, I'm still trying to monetize hope.
Yeah. You look like you're in great shape.
Peter H. Diamandis, MD
Oh, I'm doing great.
Is there any sort of youth-serum thing going on, or what?
Peter H. Diamandis, MD
It's our Longevity XPRIZE. We're getting there, buddy.
Okay.
Peter H. Diamandis, MD
We're getting there. I think in our last conversation, you were getting on board with the idea of extended longevity, yes?
Yeah.
Peter H. Diamandis, MD
Okay.
Fully on board.
Peter H. Diamandis, MD
We'll leave it at that.
To some degree. I don't know if we want everyone to live forever or whatever, but I think healthspan—and not having an extended period of senescence where you're just drooling on yourself—sounds like a good idea. We want to avoid that.
Peter H. Diamandis, MD
Yeah. First off, congratulations on the merger of SpaceX and xAI. Baller move.
Okay.
Peter H. Diamandis, MD
That's going to power humanity's first Dyson swarm.
Yeah.
Peter H. Diamandis, MD
I'm curious—
Great.
Peter H. Diamandis, MD
It truly is. What's your timeline for launching these data centers, and how much bandwidth do you think you can get in the first year? Give us a sense of the speed at which you're going to make this happen.
Yeah. SpaceX is in a quiet period. I can't actually tell you—
Peter H. Diamandis, MD
Yes.
Things that would cause problems. Yeah.
1. Recursive AI Takes Off
Peter H. Diamandis, MD
I'll leave it at that. I appreciate that. But I can't wait for the speed. We had a conversation here on Monday with Eric Schmidt and with one of the leads from one of the other hyperscalers. I won't mention who. But I'm curious where you feel we are in recursive self-improvement. Are we there? Do you see Grok doing recursive self-improvement at this point? What's the timeline for AGI and ASI? Give us a sense of that.
Yeah. I think we've been in recursive improvement for a while here. If you mean recursive self-improvement without a human in the loop, is that what you mean?
Peter H. Diamandis, MD
I do. On the AI software side.
I mean, humans are gradually getting less and less in the loop on recursive self-improvement. Every successive model is built by the one before it. That is happening to a large degree, but it's not yet fully automated. It may be there by the end of this year, but not later than next year.
Peter H. Diamandis, MD
Do you see a hard takeoff at that point?
We're in the hard takeoff.
Peter H. Diamandis, MD
Okay.
Right now.
Peter H. Diamandis, MD
Yes.
At this point, I go to sleep, there's some massive AI breakthrough, and when I wake up, there's another one.
Peter H. Diamandis, MD
Yes.
Yeah. It's hard to keep track, honestly. It's a bit of a head-spinner.
Peter H. Diamandis, MD
Yeah. Well, I think a lot of the head-spinning is happening from you, too.
Yeah. Grok's doing pretty well, and by some metrics it's the best. For example, it's the best at predicting things, which is arguably the best metric for intelligence. The new Grok 4.20 is really, really good.
We're currently behind on coding. The reason I was a bit late for this was that I was just in a giant all-hands on coding, going through all the things that need to happen to essentially catch up with and exceed our competitors on coding. I think we'll do that. We should probably get there by the middle of this year.
2. The Intelligence Explosion
I think people don't quite understand just how much intelligence there will be, or just how far it will exceed human intelligence, to a degree that is impossible to fully understand. But you can certainly imagine a situation where, let's say, 1 million times more energy is harnessed than all of Earth's current electricity usage. That would still only be roughly a millionth of the sun's energy output.
So essentially, if you increase Earth's economy by a factor of 1 million, it's still roughly a trillion. Since we're a trillionth of the sun's energy, if you increase Earth's economy in terms of electricity usage by roughly 1 million, you will be only roughly one-millionth of the sun's energy harnessed.
What would an economy or an intelligence using 1 million times more electricity than all of Earth's civilization think about, look like, or do? It's going to be something pretty magnificent. The challenge will be even vaguely appreciating that level of intelligence.
But it's safe to say it will solve everything you can possibly think of.
Peter H. Diamandis, MD
Yes.
Longevity being certainly one of them. I do enjoy your unrelenting optimism.
Peter H. Diamandis, MD
Thank you, pal.
I see you've—
Peter H. Diamandis, MD
My hope.
Yeah, exactly. You've taken to heart monetizing hope, which is pretty funny.
Peter H. Diamandis, MD
Yes. It was Grok's marketing advice to me when you roasted me on—
Right. Or Crocodile singing and saying, “You monetize.” But hey, if you're going to monetize, it's better than monetizing misery, I suppose.
Peter H. Diamandis, MD
Yes.
For sure.
When you have AI and robots, they're going to increase economic output by so many orders of magnitude that we cannot possibly comprehend it.
Peter H. Diamandis, MD
We're likely, in a very short time, to become a minority, then a vast minority, then a microscopic minority of intelligence on this planet.
Yes, not even on this planet, but in the solar system.
Peter H. Diamandis, MD
Yes, for sure.
If your best-case outcome for Earth—for intelligence—is roughly one-billionth of the sun's energy, that's your best-case outcome if you generate—
Peter H. Diamandis, MD
Intercept it, right.
Intelligence only on Earth.
Yes, because roughly half a billionth of the sun's energy hits Earth, and that's the vast majority of energy that's out there, or that we can access. Really, the intelligence in the solar system will be many orders of magnitude greater than the intelligence on Earth itself.
Peter H. Diamandis, MD
How—may I ask you a question, Elon? How far out can you see? How many years out can you make reasonable predictions right now?
It's hard to predict the path exactly, especially because often things are kind of an S-curve, or a series of S-curves, where it starts off slow, grows exponentially, hits a linear zone, and then goes logarithmic. That generally has been what I've seen with breakthroughs in AI.
AI, for example, will have some breakthrough. It'll have an S-curve, and then it looks like it's just going to go to infinity, but then you hit logarithmic returns until there's another breakthrough. Progress in AI is just a series of overlapping S-curves, or connected S-curves.
Peter H. Diamandis, MD
I mean, there was a point where you could probably predict out a decade or 2 decades. What are your thoughts now?
Yeah. Okay, this is going to sound pretty crazy.
Peter H. Diamandis, MD
It's okay. We've been talking about crazy all week.
But you're a receptive audience to wild prognostications. I'd say the economy is greater than 10 times its current size in 10 years.
Peter H. Diamandis, MD
Okay.
Which is really saying something.
Peter H. Diamandis, MD
Yeah. You had said triple-digit growth in 5-plus years from now for GDP—
Yeah.
Peter H. Diamandis, MD
And 10× the economy. But in terms of your ability to—
Yeah, I feel like 10× in roughly 10 years is actually a fairly comfortable prediction.
Peter H. Diamandis, MD
Hmm.
There's obviously—if there's World War III or something, that could put a kink in those plans, or those expectations. But in the absence of World War III, if current trends continue, I would say the economy 10×s in 10 years.
Peter H. Diamandis, MD
Love it. Can you give us an— We had a bunch of—
And we'll have a base on the moon.
Peter H. Diamandis, MD
Yes. And we'll have—
And we'll have people on Mars.
Peter H. Diamandis, MD
And we'll have mass drivers on the moon. Yes. Elon Musk
I think so. In 10 years, I think we'll have a mass driver on the moon in 10 years.
Peter H. Diamandis, MD
Gerard K. O'Neill's vision being fulfilled. We had 4 robots on stage here this year at the Abundance Summit. I look forward to Optimus. I'm curious about the Optimus 3 timeline, and in particular, when can I buy 1 or 2? When do you expect Optimus to go into commercial sale, or will you be leasing it?
3. Optimus Enters Production
We're in the final stages of completion of Optimus 3, which is really going to be by far the most advanced robot in the world.
Nothing’s even close.
Peter H. Diamandis, MD
Yeah.
In fact, I haven’t even seen any demos of robots that are as good as Optimus 3, frankly. Maybe they’re out there, or they’re secret or something; I don’t know. But I have to make sure I’m saying things that are reasonably public as well, of course.
Peter H. Diamandis, MD
Of course. We’re streaming this on X.
Okay. This is pretty public in that case.
Peter H. Diamandis, MD
Yes.
I think we’ll start production on Optimus 3 this summer.
Peter H. Diamandis, MD
Hmm.
But very slowly at first.
Peter H. Diamandis, MD
Yeah.
It’s a classic S-curve ramp of manufacturing units versus time. We’ll probably reach high-volume production around summer next year, and then we’ll have Optimus 4 design-complete next year. I try to release a new robot design every year and improve the robot design every year.
Peter H. Diamandis, MD
When Dave Blundin and I were at the Gigafactory, it was an extraordinary experience—11.5 million square feet for Tesla. I think you said you’re building out 9.5 million square feet for Optimus there as well, which is extraordinary.
Yeah.
Peter H. Diamandis, MD
Let’s—
It’s 10 million square feet, round numbers.
Peter H. Diamandis, MD
Yeah.
That’s going to be quite a new factory design, too. It’s still not different from other factories.
Peter H. Diamandis, MD
How far are we from having robots build robots? You’ve automated so much of the Gigafactory already, where humans are playing a small role. Do the robots just play the role that humans are playing in that regard?
We still have a lot of humans building things. Tesla direct employees who are building things—or, basically, people in the factory who are either building or managing people who are building—number roughly 100,000. So we have a lot of people.
Tesla’s total headcount is around 150,000, of which two-thirds are in the factory in one form or another. Then our suppliers probably have maybe 1 million or 2 million people. So it’s a lot of people. What we do expect, though, is that the output per person at Tesla becomes very, very high.
Peter H. Diamandis, MD
Yeah.
We’re not planning any layoffs or reductions in personnel. In fact, we will increase our headcount. But the output per human at Tesla is going to get incredibly high. You can’t really believe it.
4. Universal High Income Arrives
Peter H. Diamandis, MD
When we were together, we discussed the idea of sustainable abundance on our podcast, and you reinforced the idea that we have a coming age of universal high income, which has become a point of discussion beyond UBI. I’m wondering if you have any thoughts on how we get there. Have you reflected on that any further?
We talked about a timeframe of civil unrest—2, 3, 4, 5 years—and probably a lot of COVID-like checks in the interim, until we get to demonetization and deflation that leads us to UHI. Any more reflections on that? People need that hope and that vision.
To be clear, I don’t think we should be complacent. We need to be careful because the future is a range of possible outcomes, and they’re not all great. But at this point, I do agree with you that it’s likely to be great. It’s probably 80% likely, maybe more likely, to be great.
I do think we’ll have universal high income. We’re basically just issuing money to people because the output of goods and services will far exceed the money supply. Effectively, you have deflation, because deflation is just the ratio of the output of goods and services to the money supply.
If the rate of growth of goods and services far exceeds the rate of growth of the money supply, which I predict will happen, then you’ll have deflation.
Peter H. Diamandis, MD
Yes. A lot of people will be spinning up new companies, competing against each other, driving the price down and increasing the variability and deflation faster and faster.
AI and robots are going to make so much stuff and provide so many services that they will actually run out of things to do for humans. They’ll just run out of things to do for humans.
There’s only so much that humans can even express that they want. Go back to my example: If you go 1 million times greater than Earth’s economy, you’ve long since saturated all human desire. Even if you go 1,000 times more than our current economy, you probably have already saturated anything people can think of that they want.
Peter H. Diamandis, MD
Do you think the value of money is going to significantly decrease? Will we go post-capitalist?
Yeah, I think money will stop being relevant at some point in the future. It’s probably something like an Ian Banks Culture sort of future. I think the AI down the road will really not use human currency. It will just care about power and mass—wattage and tonnage.
Peter H. Diamandis, MD
It’s kind of ironic, then, right? Just as you’re becoming a multitrillionaire, money starts to have less value.
Yeah, pretty much. Being a trillionaire represents some percentage ownership in companies that I’ve built, and it’s not sitting in a bank account, as you know. I literally own a percentage of the companies. The companies are doing lots of useful things, the value of the companies grows, and I own a percentage of the companies. That sums up to that number, which seems high.
Peter H. Diamandis, MD
You know, I was interviewed by somebody who was asking me about your drive—what drives you—and I said Elon is driven to solve problems. He’s driven to make life and the world better by solving the biggest problems over and over and over again. If someone else were solving them, he wouldn’t need to, but no one else is solving them.
I just wanted to say thank you for that, pal. Thank you for that.
You’re welcome.
5. The Singularity Changes Everything
Peter H. Diamandis, MD
I am curious: Do you think that democracy and our modern institutions can keep up with the supersonic tsunami coming our way, or are they just going to fall in its path? Are they just going to break down? How do we deal with it?
It’s called the singularity for a reason, which is that it’s hard to predict what happens in the singularity. Grok’s logo is the singularity.
Peter H. Diamandis, MD
I love it. It’s a beautiful logo behind you, by the way. It’s gorgeous.
Thank you. The halo around a black hole is where the mass and light are falling in. It’s hard to know what happens inside the singularity.
But it’s going to be very interesting. The future will be very entertaining. Of that, I’m confident.
Peter H. Diamandis, MD
Yes.
I also think AI and robotics are the only way we’re going to solve our budget deficit, frankly, and not just go bankrupt as a country. You’ve had an influence on me in that I’ve decided to be more optimistic. We should just be more optimistic.
Peter H. Diamandis, MD
Thank you, pal.
Not that I wasn’t an optimist, but I was dwelling a little too much on the negative stuff.
Peter H. Diamandis, MD
It’s all upside being an optimist.
Oh, yeah.
Peter H. Diamandis, MD
And a realist a little bit.
Yes.
Peter H. Diamandis, MD
Yeah.
Yeah, exactly. You don’t want to be complacent or just assume everything’s going to go well, but try to make it go well.
There will be some pretty amazing things that happen. If you’ve got humanoid robots with very high dexterity that are incredibly smart, it means that everyone on Earth will have access to better medical care than the richest person on Earth.
By the way, if I’m allegedly the richest person or whatever, I think sovereigns are actually richer than me.
Peter H. Diamandis, MD
I think so.
I had to have neck surgery 3 times because the first 2 were done wrong. I’m like, “What the?” My back still hurts a little bit. I’m like, “Can AI please solve back pain?” That would be a huge win, and I think it will.
Peter H. Diamandis, MD
Yep.
Back pain sucks, and I think that’s sometimes why people get grumpy when they get older. It’s because of back pain.
You know, if your back hurts all the time and you can't sleep well, you're going to be grumpy.
Peter H. Diamandis, MD
We had David Sinclair on stage this morning, and he's going into human trials with ER-100, his partial epigenetic reprogramming. One of the papers recently published shows that it enables joint repair. So back pain—
Yeah.
Peter H. Diamandis, MD
—may be one of the things that it eliminates.
That would be amazing.
Peter H. Diamandis, MD
Yeah, for sure. For sure.
Honestly, the average happiness level for humans would just go up tremendously if you solved back pain, because it's not a question of if you'll get back pain; it's when. The spine is not a good design.
Peter H. Diamandis, MD
I keep on inviting you to come down to Fountain Life in Dallas. We'll help you out sometime when you have time.
Okay. I understand you can get MRIs and CT scans and everything, but what do you do with that? I don't know.
Peter H. Diamandis, MD
Happy to send you the list. I'll DM you the list of therapeutics.
You want a serum or something? I don't know.
Peter H. Diamandis, MD
You know, listen, you've been so generous. Next up on stage with me is another great moonshot entrepreneur, Ben Lamm, who runs Colossal, the de-extinction company—the woolly mammoth and 15 other species. I heard you say you might want a miniature woolly mammoth. Is that true?
Yeah. I think it would be really cool to have a pet miniature woolly mammoth. That'd be pretty epic.
Peter H. Diamandis, MD
Okay. I'll put in a word for you with Ben.
That'd be adorable. Little things just running around, trumpeting away. It's like, “Look at that.” It'd be a great little pet.
Also, can somebody please do Jurassic Park in real life? I'd definitely go, even if there was some risk of death. That'd be super cool.
Peter H. Diamandis, MD
I think if anybody's going to do that, it's Ben Lamm and Colossal. He's engineering living-life products. Someone asked him recently if he could make a Pikachu, and he said, “Probably.”
Yeah. Well, Jurassic World, whatever—that would be great.
Peter H. Diamandis, MD
All right. I'll ask him.
Yeah.
Peter H. Diamandis, MD
Elon, I'm so grateful for you coming and joining us and sharing. Thank you, my friend. Let's give it up for Elon Musk.
Thanks, guys.