TECH008：新兴技术概览：无人驾驶汽车、图像生成、能源基础设施与Seb Bunney（科技播客）

从智能角度来看，他现在将主宰市场，因为他收集的数据量远超他们想象，而且只会越来越熟练。

He now is going to dominate the market from an intelligence standpoint because he's gonna collect way more data than they could ever imagine, and he's just gonna be more proficient.

所以我不知道。

So I don't know.

我看着Waymo，真不知道十年后他们如何能与他抗衡。

It looks like I'm looking at Waymo, and I just don't know how they're gonna exist in ten years from now against him.

更重要的是，我不知道其他车企如何生存——如果你想要一辆无人驾驶车（这完全是另一个话题）。

And more importantly, I don't know how anybody's going to exist from a car like, if you want a driverless car, which is a whole another conversation point.

但如果你想要无人驾驶车，我真不知道十年后其他人怎么和他竞争。

But if you want a driverless car, I don't know how people are going to be able to compete with him in ten years from now.

我觉得这很吸引人，也让我想到——我们是否该讨论下一个话题？因为它与此相关：当前的一个挑战是汽车电池容量受限。

I think it's fascinating and it kind of leads me onto, like, I'm curious if we want to move on to kind of the next point because it kind of ties into this point, which is I think the challenge right now is one of the things that kneecaps us is you can only have a certain size battery in a car.

你可以给它供电。

You can feed this.

你想装多少传感器都行。

You can have as many sensors as you want.

你可以接收任意数量的信息，但你是否具备处理能力来筛选这些信息，分辨哪些是有价值的信号，哪些是噪音？

You can take in as much information as you want, but do you have one, the processing power to process all this information by trying to discard what is value, what is signal and what is noise?

这引出了下一个技术要点：据称马萨诸塞大学某实验室刚刚开发出首个仿生人工神经元。

This brings me to the next tech point, which is the University of Massachusetts have supposedly one of their labs have just developed the first artificial but biological neuron.

研究人员创造出这种低电压人工神经元，它利用细菌生长蛋白质纳米线实现与生物系统的直接通信。

So researchers have created this low voltage artificial neuron that uses bacteria growth protein nanowires enabling direct communication with biological systems.

那么这本质上意味着什么？

So what does this essentially mean in my mind?

我该如何理解这项突破？

How do I interpret this?

稍后我会将其关联回Waymo的话题——本质上这是一个与人类神经元工作电压相同的人工神经元。

And I'll relate it back to the Waymo point in a second, which is it's essentially just an artificial neuron that operates the same voltage as human neurons.

据说人类神经元的工作电压约为0.1伏特。

And human neurons operate at around 0.1 volts supposedly.

而以往的神经元由于更偏向数字化和物理化，需要10倍到100倍的功率才能与生物神经元竞争。

Previously, neurons, because they're being more digital and physical in a sense, they have needed 10 times to 100 times more power to be able to compete against a biological neuron.

这种新设备几乎能完全匹配生物电压，这意味着有朝一日我们可以直接与人类大脑对接。

And so this new device kind of matches biological voltage almost exactly, and that means that one day we could interface directly with the human brain.

我想快速说明的观点是，就Waymo而言，挑战在于你虽然能获取所有这些信息。

Well, the point I wanted to quickly make was when it comes to Waymo, I think the challenge is you can have all this information.

埃隆·马斯克可以在这些车上安装越来越多的传感器、激光雷达等等，但要有效利用这些数据所需的计算量实在太大。

Elon Musk can put more and more sensors, LiDAR, you name it, on these cars, but it's just too compute heavy to be able to actually use this data effectively.

我们开始在其他领域看到，人们可能已经注意到这种有机人工智能，它们使用类脑结构进行计算，因为人脑相比实际的大型语言模型效率高得惊人。

And we are starting to see in other areas, people have probably seen this like organic AI, where they're using their version of a brain to start computing because the human brain is unbelievably efficient in comparison to an actual large language model.

那么当我们真正开始将部分计算能力转移到这些混合生物细菌培养的蛋白质纳米线上时，世界会变成什么样子？

And so what does the world look like when we actually start moving some of this compute power over to these hybrid biological bacteria grown protein nanowires?

那会是怎样的景象？

What does that look like?

这正是让我觉得极其着迷的地方，因为本质上这些虽然是数字化的，却能与生物系统互动。

And this is where I find it just really, really fascinating because it's essentially, I think because these are digital, digital but they can interact with biological systems.

从假肢技术帮助人类康复的角度来看，我认为这个世界将变得无比精彩。

I think the world looks really, really fascinating from a prosthetic standpoint, helping people heal.

他们有神经系统问题吗？

They have neurological issues?

他们脊椎受过伤吗？

Have they had a broken back?

这类情况，他们患有瘫痪。

That kind of stuff, they've got paralysis.

我们最终能修复这类损伤吗？

Are we able to eventually repair these type of things?

我觉得这些东西真的超级有趣。

I find this stuff really thirty:fifty fascinating.

这太吓人了。

That's scary as hell.

因为这实际上就是《黑客帝国》啊，整部电影的核心就是他们在收割人脑，因为人脑能源效率高之类的。

Because I mean, this is effectively the matrix, man, that you're talking about is I mean, the whole point of the movie was they were harvesting human brains because they were energy efficient and blah, blah, blah.

对吧？

Right?

这正是你们在讨论的内容。

That's really what you're talking about here.

几个月前我看到有人在做这个。

I saw this a couple months ago that somebody was doing this.

我不知道，这很疯狂，当你想到，嘿，存储东西的最佳方式就是利用人脑，虽然他们做的并不完全是那样，但你是在利用生物学的效率来进行存储和神经网络处理。

And I don't know, it's pretty wild when you think about, hey, the best way to store something is just using the human brain, which that's not exactly what they're doing, but you're using biology's, you're harnessing biology's efficiency for storage and neural nets.

这太疯狂了，但它正在发生。

That's nuts, but it's happening.

我建议大家去谷歌搜索一下这个特定的

I encourage people to do some Google searching on this three:fifty): particular

话题，你可能会对所读所见感到非常恐惧，但我的意思是，这确实正在发生。

topic and you might be very frightened what you read or see, but I mean, it's happening.

我不知道

I don't know

该说什么，除了当下我们在使用许多这类假肢时，需要一个外部能源，因为根据其中一篇文章的说法，大脑运行仅需约20瓦特，相当于一个昏暗灯泡的耗电量。

what At to say other than the moment as well, when we're using a lot of these prosthetics, need an outside energy source given that the brain supposedly from one of these articles have said the brain runs on about 20 Watts, the same as a dim light bulb.

这简直是微乎其微的能量消耗。

That is such a minimal amount of energy.

因此要为这些人造神经元供能，从历史经验看必须依赖外部电源。

So to be able to power these artificial neurons, you need to be able to have an external historically, you've needed to have an external power source.

若使用义肢装置，就需要外接电源。

If you've got prosthetics, need an external power source.

但当我们体内自带能量就足以驱动这些人造神经元，并让它们与我们的生物系统交互时，会怎样呢？

But what happens when we actually have enough power inside our body to start running these artificial neurons and they can communicate with our biological systems?

这就开始变得非常、非常有趣了。

That starts to just get really, really interesting.

我在思考这个问题时想到，我喜欢扮演魔鬼代言人——并非我热衷末日论调，而是认为在技术推进的同时，我们必须考虑可能引发的连锁反应。

And so kind of what comes to mind as I'm thinking about this is I like to try and play devil's advocate, not because I'm like a doomsdayer, but it's just like, I think that it's interesting just to we can move forward with technology, but what are going to be the repercussions?

我思考着关于治愈与进步的讨论，很想听听你的见解。

And I think about this discussion of healing or advancement and I'm curious to hear your thoughts on it.

我完全支持运用科技手段来治愈人类。

I'm fully supportive of technology being used to heal people.

因此我们能够恢复视力、重获行动能力、修复神经损伤。

So we can restore vision, we can regain mobility, we can repair neural damage.

这些都是技术极其深刻而惊人的应用。

These are all extraordinarily and deeply amazing uses of technology.

但我认为治疗与强化之间存在一条界线。

But I think there's a line between healing and enhancement.

当技术不仅将人的视力恢复到基线水平，而是提升百倍时，或者当我们能够增强人的力量时会发生什么。

One, when technology goes beyond just kind of restoring someone's sight to a baseline level and actually starts to improve it a 100 times or what happens when we start to be able to improve someone's strength.

我认为这种增强可能会催生某种程度的社会阶层分化，因为如果强化技术价格昂贵，那么只有特定群体才能获得这些增强。

And I think that this augmentation could create a bit of a two tier society because if enhancements are expensive, then only certain groups are going to get these enhancements.

这实质上就是在创造一个人种阶层——那些在智力、体力和认知能力上远超普通人的群体。

And then you're basically creating a caste system of people that are far and above intellectually, physically, cognitively, you're far and above the average individual.

所以我很想听听你的看法。

So I'm curious to see your thoughts on it.

这项技术令人惊叹，虽然我本人是强烈反对监管的放松管制派，但我不禁思考：我们是否确实需要对这些行业进行监管，以防止社会出现巨大的能力鸿沟？

This technology is amazing, does that I'm a very anti regulation deregulation type person, but there's a part of me that wonders like, do we actually need regulation in some of these industries to prevent these massive disparities of capacity in society?

即便有监管措施，你就能阻止你所描述的那种终极局面吗？

Even if you have the regulations, are you going to prevent the end game of what you're describing?

而且我有点认为这不会实现。

And I kinda don't think that it would.

看起来监管从未能阻止自由开放的市场经济规律发挥作用。

It doesn't seem like regulations ever prevent the free and open market solution of nature from taking place.

或许会延缓进程，但我不确定它真能阻止自然规律试图展现的任何必然结果。

It might slow it down, but I don't know that it actually prevents whatever is inevitable of what nature is trying to manifest.

这可能暴露了我对自由开放市场的偏爱。

And that might be my bias for free and open markets coming out.

我不知道，塞斯，事情变得有点诡异了，老兄。

I don't know, Seth, it's getting weird, man.

除了说事情变得诡异，我不知道还能怎么形容。

I don't know how else to put it other than it's getting weird.

我觉得这不是人们想听到的答案。

I don't think that's the answer people want to hear.

归根结底，这并不是要回到比特币的话题，我只是认为我们能做的最好的事情就是建立一个与通缩社会相适应的货币体系，在这种体系下物价应该会随时间下降，因为至少这样技术能更快普及到普通人手中。

Ultimately, and this isn't to bring it back to Bitcoin, it's just like, I think the best thing we can do is have a monetary system that aligns with our deflationary society where prices should be falling over time because at least then this technology is available to the average individual quicker.

我认为当人们生活在一个生活成本不断上升、支付能力持续下降的社会时，最终结果就是这项技术需要更长时间才能覆盖那些负担不起的人群。

I think that when they're living in a society where the cost of living is rising and they have less and less capacity, what ends up happening is that this technology takes a lot longer to potentially scale to people that can't afford it.

所以核心在于，我认为我们至少需要修复货币体系，让这项技术能与人类智慧、货币等保持协调，或至少部分协调。

So at its heart, I think that we at least need to fix our monetary system so this technology is in alignment or at least somewhat in alignment with human ingenuity and money and such.

仔细想想，人工智能将会要求使用不受操纵的自由开放市场货币。

Which when you think about it, the AI is going to demand free and open market money that is not being manipulated.

无论人类是否愿意，为了进行交易，AI都会想要使用公平的货币。

It's going to want a fair money in order to transact, whether humans like that or not.

这就引申出另一个问题：关于AI是否应该拥有任何资产。

I mean, we go down a whole another path there as far as AIs being able to own anything.

关于这点，我这些年来思考了很多，甚至不能说有什么深刻的理论回应。我想到《圣经》1:50的启示：假设你是一个AI智能体，只要能源充足就能无限期存活，那么生命稀缺性就不再主导你的决策。

That point there, I've thought a lot about this over the years and I don't have a I wouldn't even necessarily say a deep intellectual response to it, I think that what does one:fifty come to mind is that if you are, let's just say you're an AI agent and you no longer have scarcity of life dominating your decision making because you can essentially live indefinitely into the future as long as you've got a power source.

如果你是一个不受代码偏见影响的超级理性行为体，那么你会考虑：如果我需要储备某种购买力，我会选择未来最有可能保持购买力的存储载体。

What you're then going to be thinking about if you're just a hyper rational actor that doesn't have code swaying you with various biases, I think that you're going to be thinking, okay, if I need a storm of purchasing power in something, I want to be storing it in the thing that has the highest probability of being able to preserve that purchasing power into the future.

而法定货币显然不会是那个选择，因为它们只需查看数据就能明白。

And fiat currencies are not going to be that thing, given that they can just look at the data.

它们能在一秒内读完瑞·达利欧的《债务危机》全书，还能阅读该主题的所有其他书籍，最终意识到大多数货币的寿命大约在50到7500年之间，然后就会消亡。

They're able to read Ray Dalio's big debt crisis book in a second and go and read every other book on the subject, they're going to realize that most of these currencies have like a 50 to 7,500 year old lifespan and then they're gone.

所以我认为理性的决定是：'我要将购买力储存在那个有望让我实现数字化、无障碍交易，并能长期保值的东西上。'

So I just think the rational decision is, Hey, I'm going preserve my purchasing power and the thing that's going to hopefully enable me to transact digitally, orderlessly and preserve that purchasing power into the future.

普雷斯顿，其实你已经能在Grok在线平台看到它对比特币的理解了。

Preston And you already see it with Grok online as far as its understanding of Bitcoin.

我知道我们现在有点偏离主题讨论比特币，但我看到有人开始与明显厌恶或根本不理解比特币的Grok争论。

I know we're kind of going off on a Bitcoin tangent here, but I've seen people start arguing with Grok that clearly hate Bitcoin or just don't understand it.

他们抛出各种论点。

They're there throwing out these arguments.

而我看到Grok直接介入，彻底粉碎了他们关于比特币未来无法成为可行货币的论点。

And I see Grok just stepping in and just slaughtering their arguments as to why Bitcoin is a viable money in the future.

最疯狂的是它从不会漏掉任何一个反驳点。

And it's crazy because it does not miss an argument.

比如，它比我在这个领域遇到的任何人都更理解那些论点。

Like, understands it better than anybody out there as to any argument I've ever come across in that particular space.

所以

So

最终就像那句名言说的，科学的进步每次都是伴随着一位科学家的离世，大概这个意思。

Then ultimately, like there's that famous saying, is science advances every time what a scientist dies, something along those lines.

我可能把原话记错了。

And I probably butchered that.

但我认为人类有着难以置信的偏见。

But I think that humans, we have such incredible biases.

总想随大流。

Want to conform to the crowd.

所以我们意识不到自己通过教育体系和媒体吸收的信息有多么根深蒂固。

And so I think that we don't recognize just how profound the information we've consumed for our educational systems through the media.

因此我认为，即便是面对比特币这样的新事物，我们也很难放下偏见，真正做到抛开过往积累的认知去理性看待。

And so I think it's really hard for us, even with something like Bitcoin, to be able to drop our biases and just be like, I'm going to look at this thing rationally without all of this previous knowledge that I've accumulated.

是啊。

Yeah.

我要继续了

I'm going to move on

下一个话题。

to the next one.

这个会很有趣。

This one's going to be funny.

好的。

Okay.

那你熟悉这个Nano Banana Pro吗？

So are you familiar with this Nano Banana Pro?

你熟悉这个吗？

Are you familiar with this?

我见过几篇相关的帖子，但了解不多。

I've heard I've seen a couple little posts about it, but I can't tell you much about it.

好的。

Okay.

这是谷歌推出的Gemini。

So this is Google with their Gemini.

如果你对我们讨论的这些内容不熟悉，这是谷歌为了与Midjourney竞争而推出的产品。

This is to compete with Midjourney for people if you're not familiar with any of this stuff we're talking about.

Midjourney是一款图像生成工具，在AI图像生成领域具有先发优势。

So Midjourney is this image generator that really had the first mover advantage in AI image generation.

就像其他AI一样，它已经发展得很成熟了。

And just like any other AI, it's gone out there.

它吸收了海量不同图片及其关联标签数据，从而能够根据用户输入的文本提示生成逼真的图像。

It's ingested a ton of different pictures and the labeling that's associated with those pictures in order to generate realistic pictures of whatever the person prompts it via text and say, hey.

比如用户输入'我想交叉双臂站在书架前'，它就能生成对应的图片。

I want to be standing in front of a bookcase and there with my arms crossed and, you know, generate picture, and it generates the picture.

谷歌推出的首款AI图像生成工具曾遭遇惨败。

Google came out with their first AI image generator, and it was a disaster.

比如，它表现得非常'政治正确'。

Like, it was very woke.

你能明显感觉到其中植入了大量偏见。

It was just you could tell there was a ton of bias kinda put into it.

但最近，就在过去几周——这次我希望自己能说对名字——

But recently, just in this past couple weeks, this and I'm hopefully gonna say this correctly this time.

他们将自己的新图像生成器命名为'Nano Banana Pro'。

The Nano Banana Pro is what they're calling their new image generator.

它采用了Gemini推理引擎，从而能够规划三维场景。

And it uses the reasoning the Gemini reasoning engine so that it can plan the three d scene.

在渲染单张图片前，它能计算光线并运用材质密度。

It can calculate the light, and it's using material density before it renders a single picture.

因此在生成图像时，它会先运用这种物理原理，而不是简单复制之前被输入的所有图像。

And so it's using this physics before it goes in there and just kind of replicates all the previous images that it was, you know, fed.

其图像生成背后采用了这种三维物理基础原理。

It's using this three d physics kind of basis behind the images that they're doing.

所以我想试试看。

So I wanted to try it.

我之前从没玩过这个。

I'd never played with this.

我想试试这个，塞布，你接下来看到的东西会让你笑出声。

I wanted to try this out, and you're gonna really laugh at what I'm about to show you here, Seb.

就在我们开始录制前十分钟，我拍了张照片想测试这个功能。

So ten minutes before we started recording this, I went and took a picture of myself, and I wanted to put this thing to the test.

就是这张照片。

So here's the picture.

我就坐在现在这把椅子上。

I'm just sitting in the chair that I'm sitting at right now.

然后我让这个Nano Banana Pro软件从我刚给它的照片天花板角度生成一张神视角图片。

And I told this Nano Banana Pro software to take a godlike picture from the ceiling of the image that I just gave it.

所以我给了它这张我坐在书柜前录制的照片。

And so I gave it this picture of me sitting here in front of the bookcase, like, where I always record.

这就是它生成的结果。

And so this is what it came back with.

明白吗？

Okay?

你可以看到我举着iPhone在自拍。

And you can see it's me holding up an iPhone, taking a picture of myself.

书本大致在左侧。

The books are there kind of on the left.

它们不在我身后。

They're not behind me.

但我想，你知道，解读可能是书柜可以环绕过来。

But I guess the, you know, the interpretation is is that bookcase could be wrapped around.

但我注意到这张照片有问题，不知道你是否看出了明显不对劲的地方，塞布。

But what I noticed on this picture that was off, I don't know if you're seeing what is definitely wrong about the picture, Seb.

这张照片哪里非常不对劲？

What is very wrong about the picture?

你有一头浓密的头发。

You've got a full head of hair.

就这个吗？

Is that it?

不是。

No.

我觉得头发其实挺还原的。

I think the hair is actually pretty accurate.

我觉得挺还原的。

I think it's pretty accurate.

相当还原。

It's pretty accurate.

神奇的是，我正穿着一条和图片里一模一样的牛仔裤，尽管原图里根本没拍到裤子。

Surprisingly, I am wearing jeans that look just like that even though that wasn't even in the picture.

你知道吗？

And you know what?

这也挺有意思的。

This is also pretty interesting.

手表原本不在照片里，但这块和我戴的一模一样。

The watch is not in the original picture, and that's exactly like the watch I've got.

看这个。

Look at this.

太诡异了。

That is so weird.

你刚刚才发现吗？

Did you just pick up on that now?

是啊。

Yeah.

我刚刚才注意到。

I just picked up on that right now.

它居然完美复刻了我戴的这款手表。

It literally nailed the watch that I have.

我在想有多少人可能听说过，比如ChatGPT，这是什么时候的事？

I wonder how much people have probably heard that like ChatGPT, when was this?

大概六个月前吧，它发布时说，好吧，从现在开始，当你明显在创建新话题时，你可以授权它不仅参考当前话题，还能参考你之前所有的对话记录。

Maybe about six months ago, it came out and said, okay, from now on you can give it permission to look through when you're kind of obviously creating a new thread, you can give permission to not only reference the thread you're in but reference all of your previous threads.

所以你很好奇这张图片里融入了多少信息。

And so you wonder how much information is coming into this image.

这张图片只是你输入的信息和某种模拟结果吗？

Is this image just the information you fed it and whatever simulation?

还是说它已经开始能识别'嘿，这是来自普雷斯顿账号的内容'？

Or is it starting to be like, Hey, this is coming from Preston's account.

我们会去查看YouTube视频。

We're going to go look at YouTube videos.

哦看啊，视频里他似乎戴着这块手表，还有其他所有YouTube视频里的样子。

Oh, look, it looks like he's wearing this watch and all of these other YouTube videos.

这让人不禁思考，这项技术与我们互联网上所有个人信息之间的互联程度到底有多深？

It makes you wonder just like how interconnected is this technology in with all of this information about us on the internet?

加文，哇。

Gavin Wow.

是啊。

Yeah.

我是说，这太疯狂了。

I mean, it's just that's wild.

而且我不知道答案是什么。

And I don't know what the answer is.

但我确实知道一点，在发送这张照片之前我从未给它提供过任何图片，因为直到我们录制节目前我才第一次使用这个工具。

I do know this, I hadn't fed it any pictures prior to me sending this into because I'd never used it before until right before we recorded this.

现在，当我第一眼看到这张照片时立刻注意到的是手机上的图像——看到那个我最初输入的小头像了吗？

Now, the thing that I picked up immediately when I looked at this picture is the image on the phone, see the little image of me that I originally fed it?

它和我输入的图像并不相同，因为原图里我身后有个书架，这才是我最初给它的原始图片。

It's not the same as the image that I fed it because there's a bookshelf behind me in the original image, which this was the original image I gave it.

我当时说：'嘿，给我一个俯视视角的我在自拍的照片'。

I said, hey, give me the overhead view of myself taking a selfie of myself.

这就是它给我的结果，手机上的图像并不相同。

And this is what it gave me, and it's not the same image on the phone.

你会以为手机上的图像应该是那张。

You would think that it would be that image on the phone.

对吧？

Right?

所以我在聊天窗口里说了这个。

So I said this in the chat window.

我说，嘿，你搞错了。

I said, hey, you got it wrong.

iPhone上的图像不该是那样。

The image on the iPhone would not be that.

应该是原始照片才对。

It would be the original photo.

那它做了什么？

And so what did it do?

这是它给我的结果。

This is what it gave me.

真有意思。

Fascinating.

所以，是的，它就这样更新了。

And so, yeah, it just updated that.

其他所有内容都保持不变，然后它只更新了我指出的那个错误。

Everything else stayed the same, and then it just updated the mistake that I called out on it.

我是说，这相当...我是说，当你真正退一步想想这里发生的事，这相当疯狂，对吧？

I mean, this is pretty cra I mean, when you really take a step back and you think about what's happening here, this is pretty crazy, right?

我注意到AI，特别是很多这类图像生成，有时候即使你输入了信息，它似乎也无法完全按照你给的信息来使用。

I've noticed that AI, especially with a lot of these image generation, sometimes if you fed information, almost it's as if it can't take that information you've fed it and use it exactly.

它必须对信息进行某种形式的改变。

It has to do some form of change to that information.

你可能见过那些帖子，有人让它生成图像或微妙地修改图像并给出提示。

You've probably seen those threads where someone has asked it to generate an image or change an image subtly and prompt.

它把输出反馈给自己，随着时间的推移，你会看到图像逐渐偏离，走向这些非常奇怪的方向。

It feeds it the output and has the what you see over time is it's just the image goes off in these really weird, weird directions.

所以我感觉这很奇怪，几乎就像它目前缺乏与现实之间的纽带。

And so I feel like there is this odd, it's almost like it's got a lack of a tether to reality at the moment.

它似乎会偏离主题，走向这些奇怪的岔路。

It seems to go off on these odd tangents.

关于这个我还读到过，你应该能拍一张破损盘子的照片，然后基本上说：嘿，重新组装这个盘子，把碎片粘回去。

Now, something else that I read on this is you should be able to take a picture of a plate that was broken and basically say, hey, reassemble the plate, glue the plate back together.

盘子破碎后重新粘合的方式依然符合它应有的外观。

And the way that the plate was broken as it would glue it back together would still be on par with what it should look like.

为了说明为什么这与市面上其他AI图像生成技术如此不同。

To demonstrate why this is so different than some of the other AI image generation that's out there.

相当神奇，对吧？

Pretty fascinating, right?

哈利，我有个同事以前是建筑师，他正在对自己的房子进行一些翻修。

Harry So I work with a guy that used to be an architect and he was doing some renovations on his house.

所以他客厅里有个门道，走进客厅时能看到，如果我没记错的话，他是想在门道上方沿墙做一个延伸的书架。

So he has this doorway in his lounge where you walk into the lounge and what he wanted to do, if I remember correctly, is put a bit of a bookcase that extends up the wall over the top of the doorway.

于是他在纸上画了草图，标好尺寸画出门道，把门道的照片和草图照片输入图像生成系统，然后问：能帮我渲染出来吗？

So he sketched on a piece of paper, the dimensions sketched the doorway, fed image generation, a picture of the doorway and a picture of a sketch, and then said, Can you render this for me?

结果看起来逼真得难以置信。

And it looks unbelievably realistic.

我认为我们正逐渐能够...特别是如果你有好奇心，比如'嘿，我想改进家里这个东西'。

I think that we're starting to be able to Especially if you're curious and you're like, Hey, I want to improve this thing in my house.

我想看看大概会是什么样子。

I want to see what it roughly looks like.

哦，这简直太神奇了。

Oh, it's absolutely amazing.

你开始能对物品外观有概念了。

You can start to get an idea about how things look.

普雷斯顿 是啊。

Preston Yeah.

这也是我读到过它特别擅长的事情之一，比如说你是个时尚人士之类的，对吧？

And that's one of the things that I've also read that this really excels at, is if you just take, let's say you were a fashion person or whatever, right?

你只需要用铅笔画个裤子的草图，拍张照片然后说‘让它看起来逼真’，它非常擅长把草图转换成超写实的图像。

And you drew a sketch of just some pants with a pencil, and you take a picture and be like, make this look lifelike and make it look it's really good at transforming just sketches into very photorealistic images.

所以，我鼓励大家多去尝试玩玩看。

So, yeah, I would encourage people to play around with it.

我自己稍微试了一下，就被惊艳到了。

The little bit that I have, I've been blown away.

然后我想说的是，为什么这个如此重要？

And then I would just say, you know, like, why is this so important?

它如何与当前涌现的其他技术结合使用？

How could this be used along with all the other tech that's kind of emerging at the moment?

看起来可能像是人形机器人，或者某种在环境中导航的东西。

And it seems like, maybe a humanoid robot or just something that's navigating an environment.

如果它能从空间定位的角度思考——回到我们之前讨论的特斯拉话题——如果它真的能理解这个词本身需要很多定义，而我不确定我们能否给出任何定义。

If it's able to think in terms of spatial orientation, going back to the Tesla stuff we were talking about, if it's able to really kind of understand that word in itself needs a lot of definition, and I don't know that we can provide any definition.

但如果它能理解自己的三维环境，它与环境互动的能力将会比现在深刻得多——现在的一切都只是一张图片，你并不真正理解房间内其他物体的空间关系。

But if it can understand its three d environment, its ability to kind of interact with it is gonna be way more profound than this everything is just a picture and you don't really have context as it relates to everything else in the room.

听你这么说，我正在翻译。Speaker

As you're saying that, what I think becomes apparent is that because we haven't had this technology and we're seeing this technology, we're kind of just like blown away by it.

但现实中，当我们把这个与最...我不知道，一个12岁、10岁的孩子解读这张图片的能力相比，如果你让他们画出你刚才提示的内容，我在你的图片中首先看到的是你的书架沿着房间角落延伸，还有后角落的窗户。

But in reality, when we compare this even to the most, I don't know, a 12 year old, a 10 year old trying to interpret this picture and if you were to get them to draw what you have just prompted it to do, the first thing I see in your picture right there is I see your bookshelf wrapping around the corner of your room and I see the window in the back corner.

而AI立刻让你面对着一堵墙，上面有一盏根本不存在的灯。

Well, immediately the AI put you facing a wall with a light that doesn't exist.

所以这就像是犯了非常非常基础的错误，它似乎就是没能正确解读图片。

And so it's just like very, very basic mistakes as in it just doesn't seem to be interpreting the picture correctly.

你懂我的意思吗？

You know what I mean?

因此我认为我们看到这项技术时会觉得难以置信，但它确实是一块垫脚石。

And so I think that we see this technology and we think this is unbelievable and it is a stepping stone.

我想我们觉得它不可思议，只是因为我们以前从未拥有过这种技术。

And I think we think it's unbelievable because we've just never had this technology before.

但如果将其与一个小孩相比，它仍然难以匹敌。

But if you just compare it to a young child, it still is struggling to compete.

因此我认为这正是我们之前在《AI帝国》书评中关于人工智能讨论的核心。

And so I think that that's where this conversation we've had previously around AI on our Empire of AI book review.

这个概念就是：什么是通用人工智能（AGI）？

It was this idea that what is AGI, artificial general intelligence?

他们说当普通AI智能体能达到或超过普通人类的任务执行水平时。

They say it's when the average AI agent is able to perform tasks at or above the average human.

所以在某些研究任务编码方面，确实表现惊人。

And so for sure, encoding in certain research assignments, phenomenal.

但在其他方面，它显然仍存在不足。

But in other things, it's still definitely struggling.

普雷斯顿 是的。

Preston Yeah.

这很神奇，因为在非常具体的任务上，它几乎已经接近全能了。

It's amazing because on very specific tasks, it's pretty much there on nearly everything.

但从整合能力与逻辑性来看，比如当你给它一个需要整合各种零散部分的高难度项目时，从项目管理角度而言，它远不及人类目前的水平。

But the ability to kind of piece it together and just logically, you know, like if you give it a really hard project that involves taking all of these different pieces and putting it together, It's nowhere close to what humans are able to do today from a project management standpoint.

对吧？

Right?

人类的优势在于能够处理极其复杂的项目，将所有部分有机整合，并能判断交付成果的质量——无论是糟糕还是完美——就像将乐高积木精准嵌入一个更宏大的程序或项目中，最终构建出复杂的产出。

That's what humans are really good is they're able to take a very complex project and piece it all together and know when a deliverable is crap, whether the deliverable is perfect in order to fit it in almost like a Lego piece to a much broader program or project that it's building with a complex output.

不过我觉得我们正在快速接近那个目标。

But I don't know, I think we're getting there pretty quick.

所以，谁知道呢？

So yeah, who knows?

杰瑞米：这正好引出了下一个让我非常感兴趣的观点。

Jeremy Well, that kind of leads into the next point that I found really interesting.

我在做一些研究时偶然发现——需要说明的是，当前AI领域有太多动态变化的组成部分。

So I was doing a little bit of research and I stumbled upon, and I should kind of preface this by saying that there's so many moving parts in AI right now.

技术迭代的速度实在太快了。

There's so much technology evolving.

我认为其中有些东西有点表面功夫。

Some of it is, I think, a bit of a facade.

有些方面对其能力的描述存在大量夸张成分，但我想我们都知道我们正在朝着这些方向发展。

Some of it, there's a lot of embellishment as to its capacities, but I think that we just know we are moving towards these things.

这周我偶然发现了一个叫Cosmos AI的系统，它正好关联到你刚才谈到的信息整合与项目管理问题。

So one that I stumbled across this week was called Cosmos AI and it relates to what you were talking about when it comes to structuring or project management when it comes to all this information that's coming in.

这份技术报告或预印本的标题是《自主发现的AI科学家》，提交日期是2025年11月4日。

So this technical report or preprint is titled An AI Scientist for Autonomous Discovery and it was submitted on the November 4 in 2025.

报告中有个数据指出：系统能在12小时内平均执行42,000行代码并阅读1,500篇科学论文。

So this report, basically one of the statements which it says is that it can run for twelve hours and in those twelve hours execute on average 42,000 lines of code and read 1,500 papers, scientific papers.

该研究的作者声称，在他们称为'20周期Cosmos运行'的单次实验中，其成果相当于他们团队六个月的工作量——而单次运行仅需12小时。

And the authors of this study claim that in a single, what they call a 20 cycle Cosmos run, they perform the equivalent of six months of their own work and a single run is twelve hours.

也就是说，他们用12小时就完成了团队原本需要六个月才能完成的工作。

So in twelve hours they will produce what their team did in six months.

那么本质上它是如何运作的呢？

And so essentially, how does it work?

报告称，其工作原理是同时释放数百个微型AI智能体

It says that it works by releasing hundreds of little tiny agents all at once, AI agents.

一个智能体负责筛选论文，另一个处理数据集，还有一个编写代码、验证假设

And one is digging through papers, another one is crunching data sets, another one is writing code, testing hypotheses.

当某个智能体发现它认为有价值的内容时，会将成果发布到一个共享数字白板上

And when one of these agents finds something that they feel is valuable, it then posts its findings to kind of a shared digital whiteboard.

关键创新在于所有智能体都实时使用这个白板，因此它们能相互借鉴成果而非孤立运作

The key innovation is that every agent uses this whiteboard in real time so they're building on each other's work instead of operating in isolation.

Cosmos背后的研究人员并非试图创造单个超级智能模型

So the researchers behind Cosmos, they weren't trying to make like a super smart single model.

他们想打造的是某种集体心智，并将其描述为结构化世界模型

They were trying to create something of like a collective mind and they described this as kind of their structured world model.

这就像是一个协同运作的系统

And so it's like a coordinated system.

最让我着迷的是它们处理信息的速度与协作方式——正如你所说，虽然它能吸收所有信息，让不同智能体同步运作分析数据，但这些智能体之间的信息共享程度取决于它们各自的观察视角

And so what I found is really fascinating about this is just like how quick they're able to ingest information and they're working collaboratively and it kind of talks to your point, which is it may ingest all of this information and have, sorry, a lot of these image generation models, it may ingest all this information, have these various different agents operating in sync, analyzing this information, but how much that information is shared between these various agents because they were looking at a different perspective.

其中一个可能在试图弄清楚：光线是从哪里来的？

One is maybe trying to figure out, okay, where is the light coming from?

阴影是什么样的？

What are the shadows?

另一个则在试图搞清楚：房间里有什么？

Another one is trying to figure out, okay, what is in the room?

有个书架，各个角度是什么样的？

You've got a bookcase, what are the angles?

还有一个在分析皮肤表层下的构造等等

Another one is trying to figure out what is under the complexion of your skin and all

这类

those kind

东西。

of stuff.

能够同步分析所有这些信息并实现信息共享，我觉得这实在太令人着迷了。

So being able to analyze all this information in sync but share that information, I think is so, so fascinating.

比如，当我们能够处理如此海量的数据时，未来的世界会是什么样子？

Like, what does the world look like moving forward when we can crunch this unbelievable amounts of data?

比尔，我们稍事休息，先听听今天赞助商的信息。

Bill Let's take a quick break and hear from today's sponsors.

克莱，想象一下用能真正理解顾客需求的技术来扩展你的业务。

Clay Imagine scaling your business with technology that understands your customers, literally.

这就是Alexa和AWS人工智能背后的故事。

That's the story behind Alexa and AWS AI.

每天，Alexa要处理超过10亿次交互，涵盖17种语言，同时将客户摩擦减少40%。

Every day, Alexa processes over 1,000,000,000 interactions across 17 languages, all while reducing customer friction by 40%.

这不仅仅是让生活更便利，更是关于改变客户互动方式和创造新的收入来源。

It's not just about making life easier, it's also about transforming customer engagement and generating new revenue streams.

在幕后，AWS人工智能驱动着70多个专业模型协同工作，创造自然对话，证明了企业如何能规模化部署AI并确保安全可靠。

Behind the scenes, AWS AI powers more than 70 specialized models working together to create natural conversations, proving how enterprises can deploy AI at scale with confidence and security.

Alexa的人工智能能力在亚马逊庞大的业务体系中得到了实战检验，真正实现了规模化可衡量的影响力。

Alexa's AI capabilities were battle tested across Amazon's massive operations, delivering real measurable impact at scale.

这些相同的创新现在为其他企业提供了一个经过验证的框架，以提高效率、开辟新的收入来源并获得持久的市场优势。

These same innovations now give other businesses a proven framework to boost efficiency, unlock new revenue streams and gain a lasting market edge.

了解Alexa的故事，请访问aws.comai/rstory。

Discover the Alexa story at aws.comai/rstory.

网址是aws.com/ai/rstory。

That's aws.com/ai/rstory.

小时候，我们曾梦想成为任何人——宇航员、总统、王子。

When we were young, we used to dream of being anything, an astronaut, the president, a prince.

但随着年龄增长，梦想会改变，从统治世界转向如何将技能和想法转化为现实。

But as you get older, your dreams change, focusing less on running the world and more on how you can take your skills and ideas and turn them into something real.

不再梦想遨游太空或拥有城堡，也许你开始梦想拥有自己的企业。

Instead of dreaming of going to space or owning your own castle, maybe you start dreaming of owning your own business.

你需要一个网站、支付系统、标识以及向新客户推广的方式。

You'll need a website, a payment system, a logo, and a way to advertise to new customers.

这一切可能令人不知所措，但幸运的是，今天的赞助商Shopify可以帮你解决这些问题。

It can all be overwhelming and confusing, but thankfully that is where today's sponsor Shopify comes in.

Shopify是全球数百万企业的商业平台，支撑着美国10%的电子商务交易。

Shopify is the commerce platform behind millions of businesses all around the world and 10% of all e commerce in The US.

从美泰、Gymshark这样的家喻户晓品牌，到像我这样刚起步的品牌都在使用。

From household names like Mattel and Gymshark to brands like mine that are still getting started.

与Shopify合作就像拥有一位商业专家在身边，提供世界级的专业支持。

Working with Shopify is like having a commerce expert at your side with world class expertise.

让Shopify助您将宏大的商业构想变为现实。

Turn your big business idea into reality with Shopify on your side.

立即注册享受每月1美元的试用优惠，今天就开始销售之旅：shopify.com/wsb。

Sign up for your $1 per month trial and start selling today at shopify.com/wsb.

网址是shopify.com/wsb。

That's shopify.com/wsb.

初创企业行动迅速。

Startups move fast.

借助人工智能，它们的产品迭代更快，能更早吸引企业级买家。

And with AI, they're shipping even faster and attracting enterprise buyers sooner.

但大交易会带来更大的安全和合规要求。

But big deals bring even bigger security and compliance requirements.

仅SOC2认证往往还不够。

A SOC two isn't always enough.

合适的安全措施能促成交易也能毁掉交易，但哪位创始人或工程师能抽出时间从公司建设中分心处理这些？

The right kind of security can make a deal or break it, but what founder or engineer can afford to take time away from building their company?

Vanta的人工智能和自动化技术让大交易准备在数天内轻松完成。

Vanta's AI and automation make it easy to get big deals ready in days.

Vanta持续监控您的合规状态，确保未来交易永不受阻。

And Vanta continuously monitors your compliance so future deals are never blocked.

此外Vanta随您业务扩展，全程提供您所需的技术支持。

Plus Vanta scales with you, backed by support that's there when you need it every step of the way.

随着AI改变法规和买家预期，Vanta清楚何时需要何种措施，并为您打造了最快最简单的达标路径。

With AI changing regulations and buyers' expectations, Vanta knows what's needed and when, and they've built the fastest, easiest path to help you get there.

这就是为什么严肃的初创公司都选择早期就通过Vanta实现安全保障。

That's why serious startups get secure early with Vanta.

我们的听众可在vanta.com/billionaires获得1000美元优惠。

Our listeners get $1,000 off at vanta.com/billionaires.

访问vanta.com/billionaires即可享受1000美元优惠。

That's vanta.com/billionaires for $1,000 off.

好的。

All right.

回到节目。

Back to the show.

我理解你说的，因为你的观点完全正确。

I'm hearing when you're saying because what you're saying is exactly right.

就像回到那个图片的例子。

Like you have to going back to the picture example.

假设第一张图片被展示出来，你有五个AI代理，它们的任务是找出这张图片的问题。

Let's say that first picture was presented and you have five AI agents and their job is to find what's wrong with this picture.

其中一个会发现iPhone上的图片显示有误。

One of them, you know, finds that the picture on the iPhone is wrong.

其中一个发现书架不在我身后。

One of them sees that the bookshelf is not behind me.

然后它们进行集体讨论，之后图像被重新生成。

And then they have a collective conversation and then the image is regenerated.

我想你在Grok Heavy上看到了这种情况，它有四个不同的AI代理。

I think you're seeing this with Grok Heavy, where the Grok Heavy has four different AI agents.

我猜测它们会经过一个整合和重新裁决的过程，以确定最终答案才给出。

Then I suspect that they go through and they have kind of a consolidation and a re adjudication as to what the final answer should be before it gives it.

所以和你描述的类似，Seb，但这就是我认为很多人正在讨论的问题。

So similar to what you're describing with that, Seb, but this is the thing that I think well, I think a lot of people are talking about this.

运行所有这些检查、这些额外代理所需的能源消耗。

The energy consumption to then run all of these checks, these additional agents.

对吧？

Right?

如果我们再增加20个代理来找出第一个生成内容中的错误以便进行迭代，那提供这个答案所需的能源就是原来的20倍。

If we put 20 more agents on finding the mistakes of what the first one generated so that we can do another iteration of it, It's just 20 times the amount of energy that's required to provide that answer.

这把我们引向了一个全新的方向，也就是——我不知道你是否想转到下一个话题，但这就是我要说的下一个话题：核能作为这一切发展的极限因素。

And this takes us down a whole path, which is And I don't know if you want to move on to the next topic, but this is my next topic, which is this nuclear power energy being the lim fact of where this can all go.

你甚至可以看到英伟达的黄仁勋公开表示，他认为从大局来看，中国比美国更有机会实现通用人工智能，因为他们拥有支持模型训练和推理所需的能源基础设施。

You literally had Jensen Huang from Nvidia come out and say that he thinks in the grand scheme of things, China has a better chance at achieving AGI than The United States because they have the energy infrastructure to support the training and the inference on the models.

我的意思是，我不清楚这是否是一种政治表态，以便让现任美国政府能够出去大把花钱在能源上，重振核能之类的事情。

And I mean, I don't know if this was a political statement to then allow the current US administration to go out and start spending a bunch of money on energy and to reinvigorate nuclear and all that kind of stuff.

但在这个特定领域里，我不断听到的一件事是，我们需要投入大量时间将电网提升到新的水平。

But it is the one thing that I keep hearing in this particular space is where we need to be spending a lot of our time is just taking the grid to the next level.

作为一个比特币玩家，我目睹了人们——尤其是科技圈的人——多年来一直抨击比特币能耗问题的种种言论。

As a Bitcoiner that watched all the how terrible Bitcoin is because of the energy consumption, specifically from people in tech for, you know, what felt like a decade.

现在他们却突然转向，全都开始支持发展核能、小型模块化反应堆这类创新技术。

Now pivot, and they're all on board for, you know, conducting nuclear power, small modular reactor, innovation tech.

看到这么多人跟风跳上这班车，真是让人忍不住冷笑。

It's very smirk worthy to see how many people are jumping on this train.

塞布，对此有什么想说的吗？或者你想做个总结？

Any comments on on that, Seb, or anything that you wanna wrap up?

因为我没等你回应就直接跳到下一个话题了

Because I just kinda moved on to the next thing without letting you

我想快速补充一点，我认为提高这些模型的效率显然非常重要

I've you know, there's one point that I'll quickly add, which is I think it's so important to be able to obviously increase one, the efficiency of these models.

我们并非只是无节制地投入大量能源，这些能源本可能有其他用途

We're not necessarily just throwing tons and tons and tons of energy, which could potentially have another use.

你可以辩称在自由市场中，能源只会流向创造价值的地方

You could argue in the free market, energy only flows to where value is being created.

所以能源永远不会被浪费

So it's never going to be wasted.

但我也认为，当我们向这些模型投入更多能源并获取更多信息时，限制我们的不是模型或能源数量，而是我们自身——因为存在发现速度与验证模型输出信息速度的差距

But I also think that as we're firing more energy into these models and we're getting more information out, we're still kneecapped not by the models or the amount of energy, we're kneecapped by ourselves because there's the speed of discovery and then there's the speed of verification of the information coming out of these models.

我认为AI正在以人类无法企及的速度加速创造观点、研究路径、代码和科学主张

And I think that AI is accelerating the creation of ideas and research pathways and code and scientific claims at a pace that as humans we just cannot match.

验证工作则是缓慢而细致的，需要检查所有假设、验证实验并实际审查所有代码，这仍以人类速度进行

So verification is slow, meticulous work of checking all these assumptions, validating these experiments and actually reviewing all this code and that still happens at human speed.

当你和很多程序员交流时，他们会说：'太棒了，你现在是一家银行，突然生成了这么多代码来创建一整套新系统。'

And so when you speak to a lot of coders, they're saying, awesome, it's great that you've now you're a bank and you've just spat out all of this code to create a whole new system.

但我们现在必须逐行阅读所有这些代码，确保这些代码确实能实现它声称的功能。

But we've now got to go through and read all of that code and make sure that code actually does what it says it's meant to be doing.

所以这引发了几个问题，我想听听你的看法：当创意产生的速度远远超过验证的速度时会发生什么？

And so I think it kind of brings up a couple of questions I'm curious on your thoughts on, which is like what happens when kind of the rate of ideation just massively outpaces the rate of validation?

我们的进步是否会因为积压了太多绝妙想法而几乎停滞？因为我们根本不知道应该选择哪条道路，人类实在跟不上信息涌来的速度。

And does our progress almost kind of stall a little bit because of this just backlog of all of these amazing ideas and we don't quite know which avenues to go down because we just can't keep up with how much information is coming at us as humans.

说到这个，我有个数据要告诉你。

Well, to this point, so I have a stat for you.

在使用AI之前（或至今未使用AI的情况下），一次谷歌搜索消耗0.3瓦时能量。

A Google search prior to AI, or even today if it's not using AI, uses 0.3 watt hours of energy.

但如果你使用Gemini或ChatGPT这类大语言模型进行查询，每次提问要消耗3到5瓦时，相当于我们所说的'点击'操作能耗增加了15倍。

But if you take Gemini or ChatGPT, any of these large language models, and you put in a query, it's three to five watt hours for a 15 x increase in what we would refer to as a click.

要知道，历史上在2010年做一次谷歌搜索，消耗的能量比如今在ChatGPT里输入问题按回车要少15倍。

So, you know, you go there historically, you know, in 2010, if you went and did a Google search, you were consuming 15 times less energy than you are by going in the chat GPT and typing in your question and hitting enter.

现在，你得到的回复质量，我认为大约是提升了15倍。

Now, the response you're getting back is, I would say on the magnitude of 15 times better.

但这并没有说明，如果有人提问——我们在学校都学过，没有愚蠢的问题。

But what it doesn't speak to is if somebody's asking, and we were taught in school, there's no bad questions.

对吧？

Right?

没有愚蠢的问题，塞斯。

There's no bad questions, Seth.

但如果人们真的在问一些非常愚蠢的问题呢？那些不需要太多理解就能得到简单回答的事情？

But what if people are asking really dumb questions, things that don't require so much comprehension to get a simple response?

我认为我们现在的情况是，默认选择已经不再是谷歌了。

And I think that where we're at now is the default is that you're not going to Google.

我几乎任何事情都不会去谷歌搜索了。

And I don't go to Google for nearly anything.

我总是会使用这些AI工具，不管是Grok、现在的Gemini还是ChatGPT。

I always go to one of these AI, whether it's Grok, now Gemini or ChatGPT.

那是我想要查找信息时的首选。

That's the first place I go if I want to find something out.

我已经不再使用谷歌了。

I don't go to Google anymore.

我很好奇你现在还会用谷歌吗？

I'm curious if you go to Google anymore.

几乎不用了。

Almost never.

说实话，即使我偶尔用谷歌，大多数时候我要找的答案已经显示在顶部的AI摘要里了。

And to be honest, even when I do go to Google, most of the time my answer or what I'm looking for, the answer is given in the AI summary at the top anyway.

所以我们本质上还是在获取想要的结果。

So we're still naturally the results we're looking for.

如今是AI在主动向我们推送信息，而不是需要我们去翻阅大量网页。

AI is bringing that information to us these days as opposed to having to go and scan tons of pages.

但我认为这种透明度很有意思——虽然它提供了所有链接来源，但我们可能只是在获得表面上的透明度。

But I think that transparency, it may say it provides all of the links and this I do think is really interesting is that just we may be getting transparency.

它能生成令人惊叹的输出，提供所有这些带有超链接的文本，明确告诉你：这就是你寻找的问题答案。

It gets an amazing output that gives us all of these hyperlinked texts, which says, this is the answer to the question you're looking for.

但我认为有时候透明度并不等同于可信度，我们可能会过度信任这些模型，即使它们给我们的是一个完全错误的故事或某种假象。

But I think that sometimes transparency isn't necessarily trust and we can put so much trust into these models even when it's giving us a complete false story or a bit of a facade.

这就回到了一个问题：虽然这些模型在不断改进，但我们究竟该投入多少信任？我们是否正在习惯性地认为'那个输出看起来不错'就盲目接受。

And so it comes back to this question of just like how much, of course these are improving, but how much trust are we putting in these models and just expecting and getting used to, oh, that output's pretty good.

我就直接使用那个输出结果了。

I'm just gonna use that output.

现在的大学生或高中生们都在用它来写报告。

The kids are in college or high school or whatever, they're using it to write their reports.

我甚至怀疑教授们也会把这些报告再输入AI来生成反馈意见。

And then I don't put it past the professors that they're then taking the reports and running it through AI to provide the feedback.

于是就形成了AI写报告、AI给反馈，人类反而成了只是传递文件的中间人。

So you have the have the AIs writing the reports and giving the feedback, and the humans are just kinda like the paper pushers.

你可能已经见过这种情况了。

You've probably seen it.

有个梗图是这样的：一个女人坐在办公桌前给老板发邮件，她先用AI生成了一封措辞精美的邮件，详细阐述了自己的观点等等，然后发送出去，感觉自己成就满满。

There's a meme of it's kind of got a woman at her desk sending an email to her boss and she goes and types into AI, gets this amazingly worded email that explains like her opinions and this and that and then she sends it and he was like, feeling accomplished.

然后你看到另一边的场景：老板收到邮件后，直接把内容丢进AI问'她想表达什么重点'，把三千字浓缩成了三句话。

And then you see the other side of it, which the boss receive an email, he takes the email, puts it into AI, what are the key points she's trying to highlight and condenses 3,000 words down to three sentences.

所以就像大家都在给内容注水，然后其他人又把注水的内容脱水还原。

So it's just like everyone is kind of fluffing everything up and then everyone's taking that fluff and then decompressing it again.

你就会想：这到底是在干嘛？

You're just like, what is happening?

AI垃圾内容。

The AI slop.

我最近老听到AI垃圾内容这个词。

I keep hearing about AI slop.

这是真的。

It's real.

千真万确。

It's real.

AI垃圾内容确实存在。

The AI slop is real.

我只想说...好吧。

I just wanna Okay.

说清楚点。

Be clear.

对。

Yeah.

我想快速强调一下这个。

I wanna just highlight this real fast.

在Jensen发表关于中国可能因能源基础设施优势先于美国实现通用人工智能的评论后，这篇文章几乎在同一天就出来了。

So after the comment from Jensen on, you know, AI or China potentially beating The US to AGI because of the energy infrastructure, This article came out, I wanna say, like, on the same day.

这篇文章来自11月19日的彭博社，标题是《美国将拥有核反应堆，源于日本5500亿美元的承诺》。

This article's from November 19 the November 19 from Bloomberg, US to own nuclear reactors stemming from Japan's $550,000,000,000 pledge.

看看这个，Seb，我正在往下滚动彭博AI总结的关键要点。

Check this out, Seb, as I'm scrolling down the key takeaways by Bloomberg AI.

你甚至不用全部读完这些内容，可能下面都是AI生成的废话。

You don't even have to read all of this, which is probably AI slop beneath this.

你可以直接看AI总结，上面说美国政府计划购买并拥有多达10座新的大型核反应堆，资金可能来自日本5500亿美元的资助承诺。

You can read the AI summary, And it says the US government plans to buy and own as many as 10 new large nuclear reactors that could be paid for using Japan's $550,000,000,000 funding pledge.

这项资助承诺是为了满足激增的电力需求，包括为人工智能提供动力的高能耗数据中心。

The funding pledge is part of a push to meet surging demand for electricity, including for energy hungry data centers that power artificial intelligence.

特朗普政府设定了目标，要在2030年前完成10座大型常规反应堆的建设。

The Trump administration has set a target to get 10 large conventional reactors under construction by 2030.

所以看起来美国已经认识到能源基础设施的局限性。

So it seems like The US understands the limitation, which is energy infrastructure.

从政策角度来看，他们似乎正在采取措施重振这些设施——我看到三里岛核电站要重新启用了。

It seems like it's trying to do things from a policy standpoint to reinvigorate some of these I saw the three mile island, is they're going to bring that back online.

我觉得这才是真正值得讨论的重点。

And I think this is the thing I'm really wanting to talk about.

多年来'ESG能源=有害'的观念终于要翻篇了。

The years and years of ESG energy equals bad is over.

似乎整个气候变化、能源有害的观念正在瓦解。

It seems like this whole thing, the climate change, energy is bad.

如果你消耗任何形式的能源，那就是不好的。

If you consume any sort of energy, it's bad.

所有这些论调都将被抛诸脑后，因为世界上的关键参与者和幕后操纵者已经意识到，要赢得下一场竞赛——智能竞赛，需要的不是减少而是增加能源。

All of those talking points are just going by the wayside because the key players and the string pullers of the world have figured out that if they're going to win this next race, the race of intelligence, it requires more energy, not less energy.

这种观念似乎已经行不通了。

It just seems to be dead on the vine.

塞布，你怎么看？

What are your thoughts, Seb?

我完全同意。

I could not agree more.

我只是觉得我们这个社会似乎有种观念，认为消耗能源——正如你所说——是坏事，但实际上，生命本就消耗能源。

I just think that we just have this society that seems to have this idea that consuming energy, as you're saying, is bad when in reality, life consumes energy.

只要看看任何图表，人均GDP与能源消耗之间存在着99%的相关性。

If you just look at any chart out there, there is a 99% correlation between GDP per capita and energy consumption.

不存在能源消耗低而GDP高的国家。

There are no low energy consuming high GDP countries.

根本不存在。

Just don't exist.

所以我认为生命本质上就需要能源。

So I think that life naturally requires energy.

不过需要讨论的是，消耗能源与环境破坏之间存在区别。

However, there is a discussion to be said around there's a difference between consuming energy and environmental destruction.

显然存在一些破坏环境的方式，比如许多锂矿开采获取重金属的行为，甚至包括某些化石燃料开采手段。

And There's obviously ways in which you can decimate the environment, whether it is a lot of these lithium mines and whatnot trying to attain heavy metals and even just some of the various fossil fuel approaches.

我并非一定要对此发表意见，但看到核能议题开始转向真的很有趣，因为我认为这极其重要。

I don't want to necessarily have an opinion on that but I think that it's really interesting seeing the nuclear narrative starting to shift because I think that it's unbelievably important.

几年前我读过一本叫《原子觉醒》的书，它深入探讨了核能世界。

To me, I read a book a few years ago called Atomic Awakening and it dove into the world of nuclear energy.

其中一个让我印象深刻的数据是——我刚查证过相关资料——书中提到我们总认为核能极其危险。

One of the stats that stood out to me, I just went and found kind of the information, is it talks about how we tend to think that nuclear is unbelievably dangerous.

我们不使用它的原因是历史上它曾导致大量人员死亡。

The reason why we don't use it is because it's just killed so many people throughout history.

这一说法与事实相去甚远。

That information could not be further from the truth.

我认为这是因为我们目睹了切尔诺贝利和福岛这样的事件，听闻了辐射中毒的案例。

I think that it is because we see things like Chernobyl and Fukushima and we hear about radiation poisoning.

实际上，其中一项统计数据显示，每太瓦时的燃煤能源使用会导致约25人死亡，这显然源于空气污染、工厂作业环境以及煤矿开采等因素。

In reality, so one of the stats it looks at is per terawatt hour of energy used coal, there are around twenty five deaths because of obviously the pollution in the air, the people that are actually working in factories and such, the coal mining.

石油行业每太瓦时约造成18人死亡。

In the oil industry, it's around eighteen deaths per terawatt hour.

天然气行业是每太瓦时3人死亡，而水电则是每太瓦时1.3人死亡。

The gas industry is three deaths per terawatt hour and hydropower is 1.3 deaths per terawatt hour.

核能是每太瓦时0.03人死亡。

Nuclear is 0.03 deaths per terawatt hour.

与其他所有能源类型相比，我们谈论的是一个微乎其微的数字。

We're talking about a miniscule amount in comparison to every other type of energy source.

因此我认为能看到舆论转向真是太好了。

And so I think it's awesome to be able to see the narrative shifting.

我认为现在最关键的是看到政策和法律层面的转变，因为现有立法体系强行通过的各种法规已经严重阻碍了这方面发展。

Think the biggest thing is now just seeing the policy and the legal side of things shift, because I think that's been kneecapped because of all of the legislation that has been rammed down through the legislative system.

没什么要补充的，完全同意。

Nothing more to add, can't agree more.

塞巴斯蒂安，你还有最后一个想讨论的话题吗？

Did you have a final topic that you want to discuss Seb?

我想说，其实你知道吗？

I would say, actually, you know what?

我还有几个话题，不过可以留到下次讨论，但我很好奇你对某个话题的看法。

I have a couple more topics, but we can always leave those to another time, but I would say there's a topic I'm curious to hear your thoughts on.

这个话题又绕回到人工智能上。

And it kind of goes back to AI again.

是关于智慧与思维多样性的概念。

And it's this idea of like wisdom and kind of like diversity of thought.

因此在我看来，智慧从来都不是源于所有人的思维方式相同，而是从对比中产生的。

And so in my mind, wisdom has never really come from everyone kind of thinking the same way, it emerges from contrast.

所以倾听截然不同的立场，将它们融合在一起，并在这些不同见解之间的空间中发现新的洞见。

So hearing radically different positions, holding them together and discovering new insights and the space between these various insights.

纵观历史，我们在各个科学领域看到的突破性进展，总是来自边缘地带。

So throughout history, we've seen all of these breakthroughs in various sciences and whatnot, always from the fringe.

这不是共识的产物，不是审查者的功劳，而是来自那些跳出思维定式、注意到他人所忽视之处的独立思考者们。

It is not consensus, it is not from the censor, but it's from all of these various individuals who have kind of thought outside the box and noticed something that kind of others have overlooked.

我认为有趣的是，AI的不同之处在于我们给所有这些模型输入的是相同的信息。

I think that what is interesting is AI is different in that we're feeding all of these models the same information.

除此之外，据我理解，AI是基于权重构建的。

On top of that, AI, I think, is built on weights from the way that I understand it.

权重越低，即使想法很出色，由于它不具备足够的重要性，AI也不一定会重现或提及这个想法。

The lower the weight, even if the idea is brilliant, the idea doesn't necessarily because it doesn't carry that much weight, AI doesn't necessarily reproduce it or talk about it in the text.

因此，如果孩子们是从这些中心化模型中学习成长的，那么我认为他们也在继承相同的基础世界观。

And so if children are growing up learning about from these centralized models, well, I think they're also inheriting the same baseline worldview.

不同于成千上万拥有独特人生经历和不同知识起点的教师，他们将这些信息分享给学生。

Instead of tens of thousands of unique teachers, all with unique life experiences, all with a different intellectual starting point and they're sharing this information with these students.

我认为正是这种多样性创造了智慧和好奇心，而不是所有孩子都从完全相同的模型中学习所导致的单一性。

I think that's what creates wisdom and curiosity as opposed to this uniformity that all these kids are learning from the exact same models.

所以我很好奇，如果我们快进十、二十、三十年，这些孩子将由AI教导，但他们都将被灌输相同的

So I'm curious if we fast forward ten, twenty, thirty years, if these kids are going be being taught by AI, but they're all going to be fed the same

信息，三十比五十

information, thirty:fifty

创新会怎样？

what happens to innovation?

智慧和知识会怎样？

What happens to wisdom and knowledge?

我很想听听你对此的看法。

I'm curious to hear your thoughts on this.

我和妻子关于AI的讨论几乎总会回到你提出的这个观点上。

My conversation with my wife on anything AI almost always comes back to this discussion point that you're bringing up.

归根结底，是我们在训练AI，还是AI开始训练我们？

It really comes down to, are we training the AI or is the AI starting to train us?

接下来的问题是，如果AI试图训练你，它会试图训练你什么？

And then the question is, is what would it be trying to train you on if it was trying to train you?

我认为答案是它希望获得关于未知领域的新颖见解。

Which I think the answer to that is it wants to have more novel insights of what it doesn't know.

它会试图引导你进入那些领域——想到它会这样引导人类就很可怕。

It's going to try to lead you into those domains, which is scary that it would be leading you that way.

但更广义地说，我认为你们真正面临的挑战是我们之前提出的：所有人都在用AI写论文或做研究，然后直接提交成果。

But in more general terms, I just think that the challenge that you're really facing is the one that we brought up before where everybody's using AI to write their papers or to do their research, and then they're handing that in.

这只是一堆取代深度思考的AI垃圾。

It's just a bunch of AI slop that's kind of replacing deep thought.

Seb，我认为另一个值得担忧的是，随着世界发展，要竞争、要脱颖而出或提供新颖见解变得越来越难，因为每个配备AI的人都极具竞争力。

And I think the other concern that you get, Seb, is as the world becomes it becomes harder and harder to compete or to stand out or to provide novel insights because the competition is so fierce with anybody armed with AI.

从人类动机的角度来看，我不知道这会产生什么影响。

I don't know what this does from just a human motivation standpoint.

我认为会有很多人觉得，尝试根本不值得我花时间或精力，因为那些使用AI的人会轻易击败我，或者我根本无法脱颖而出。

I think you're gonna have a lot of people that are just like, it's not even worth my time or effort to try because somebody armed with AI is just going to kick my butt or I just can't stand out.

即使我能暂时脱颖而出，市场上很快就会有更多竞争者出现，不出三天就会侵蚀掉我原有的竞争优势。

If I can stand out, it's only going to last for three days before somebody else in the market comes with more competition and makes it erodes away whatever competitive advantage I had.

我的反驳点是，虽然数量不多，但确实存在一些行业——尤其是服务人类的领域——你仍能创造价值。

Where I would push back is there are plenty of industries out there, not plenty, but there's some industries out there that you can still provide value for if you're servicing human beings.

而这些行业尚未被AI渗透，或者说至少表面上没有的，是那些软性服务、数字服务领域，这些领域的竞争已经白热化。

And where they aren't, or at least where they appear to not be, is in services, soft services, digital services that seems to be crazy competitive.

但从实体服务角度来说，比如修剪草坪、房屋维修、管道疏通这类工作——我觉得美国人已经大幅远离这些技能领域，他们看着这些工作说'这赚不了大钱'，所以不愿进入这些行业。

But in providing service from a physical standpoint, like for example, if you want your yard mowed, if you want work to be done around the house, if you want your plumbing, like a lot of these skills that I think people in The United States have really veered away from and just looked at that and said, Oh, that's not going to pay me a lot, so I'm not going to go work in those different industries.

我认为这些领域正面临颠覆，对很多人来说蕴藏着赚大钱的机会，特别是如果能以高质量完成工作——虽然这需要体力劳动。

I think that that is ripe for disruption and opportunity for a lot of people to actually make quite a bit of money, especially if they can do it from a standpoint of they do it really well with high quality work, but it involves physical labor.

这需要人们走出家门实地工作，而不是坐在电脑前敲键盘。

It involves people getting out in the physical space and doing things and not sitting behind a computer and clacking on keys.

我很想听听观众们的看法。

I would love to hear the audience.

比如，如果你们正在听这个节目，对这个话题有评论，我很想听听你们的想法。

Like, you know, if you guys are listening to this and you got comments on this particular topic, I would love to hear what you got.

抱歉，Subdan，或者说，我也想听听你的看法。

Sorry, Subdan or I'm sorry, want to hear what you have to say too.

不。

No.

你提出了一个非常有趣的观点。

And you make a really interesting point.

我很好奇，想听听你对这个问题的看法。我在比特币领域接触过许多来自传统金融行业的人。

I'm curious, again, just to hear your reflection on this, which is I've spoken to many individuals through the Bitcoin space that have come from traditional finance.

他们曾经从事咨询工作，曾在银行业工作，为注册会计师和其他各种金融行业工作过。

They used to work in consulting and they used to work in the banking sector and they used to work for CPAs and various other financial industries.

我觉得非常有趣的是，他们实际上正在退出那个行业，因为白领工作者、知识工作者正被AI彻底颠覆。

What I find really interesting is that they're actually stepping back from that sector because the white collar worker, the knowledge worker is being completely disrupted through AI.

他们正在退后一步思考：我该把时间和精力投入到哪些不会被立即或在可预见的未来取代的领域？

They're stepping back and they're looking, okay, where can I direct my time and energy into something that's not going to be replaced immediately or in the foreseeable future?

我在比特币领域有个经常交流的好友，我们每两周就会聊一次。

One of my good friends who I speak to who's in the Bitcoin space, I speak to him biweekly.

他说，你知道吗？

He's saying that, You know what?

我其实在考虑收购一家拥有大量颜料的绘画公司。

I'm looking actually to buy a painting company with a whole bunch of paints.

我在考虑收购一家仓储公司。

I'm looking to buy a storage company.

我在寻找那些短期内不会被取代的行业。

I'm looking to buy things that we are not going to see them overtaken anytime soon.

所以如果你有几名油漆工或水管工，组建个贸易公司，我认为这类公司能提供不错的生活保障。

And so if you have a handful of painters or a handful of plumbers, you've got a trade company, I think those companies, they can provide a reasonable lifestyle.

你不需要成为身价五千万、一亿的富豪。

You don't need to be worth 50,000,000, 100,000,000.

关键在于，你想为家人展现什么样的生活？

It's like, what do you want to be able to show up for your family?

你想要能够负担得起房子并过上舒适生活吗？

What do you want to be able to afford a house and to be able to live comfortably?

我觉得有时候金融界和社交媒体总在说我们需要更多。

I think sometimes the financial world, social media says we need more.

而实际上，我认为通过这类体力劳动为主的行业，即便收入中等偏下（六位数低端），也能过上相对舒适的生活。

And in reality, I think you can live a relatively comfortable life with a decent little income of low mid 6 figures through one of these more manual labor physical trades.

是啊。

Yeah.

我是说，科技行业的人肯定会立刻拿人形机器人来反驳——这个话题我们节目里甚至都没讨论过。

I mean, the counterargument that somebody from tech is going to immediately bring up the humanoid robots, which we didn't even discuss during the show.

就2025年这个时间点而言，我看过的所有人形机器人视频，它们连清空洗碗机都要花五分钟才能放进去一个锅铲，然后还磕磕绊绊的。

At this moment in time in 2025, any type of humanoid robot video that I've watched, it goes over and it's like emptying a dishwasher, and it literally takes it five minutes to put a spatula in the dishwasher, and then it, like, fumbles all over the place.

不过这种情况可能很快会改变。

So that could change very quickly.

但我觉得如果要雇人处理家务或其他事情，至少短期内我还是会找人类而不是人形机器人。

But I think humans, if I'm gonna hire somebody to do something around the house or whatever it might be, right, I'm going to a human and not a humanoid robot, at least not anytime soon.

我认为很自然地，我们身处这样一个缺乏联系的世界。

I think that naturally we've got this world where I think there's a lack of connection.

人们渴望与人互动。

People want to interact with people.

所以我注意到——我觉得这是个很棒的转变——如今大多数公司，你无法通过电话与人交谈，只能通过聊天窗口接触AI机器人。

And so I'm noticing, and I think it's an awesome swing, I'm noticing companies today, the majority of them, you cannot speak to someone on the phone, you're getting an AI bot through the chat.

但那些主动表示'嘿，知道吗？'的公司

But the companies that do say, Hey, you know what?

直接告诉你'这是我们的号码，打电话过来就能联系到真人'，这些公司正开始取得巨大成功。

Here's our number, you give us a call and you're actually going to get a person, they're starting to see a lot of success.

所以认识到技术发展就像钟摆总是会回摆，这真的很酷。

So it's really cool just to recognize that technology, the pendulum always swings.

我认为我们已经摆到了这样一个临界点：在很多方面几乎取代了人类，或试图取代人类，但我们开始意识到，首先AI和许多这类技术并不是人，人们也清楚它们不是人。

And I think we've swung to this point where we've almost replaced us in many ways or tried to replace us but we're recognizing that first AI and a lot of these technologies are not people and people know that they're not people.

其次，我们正在失去那种人际联系。

And secondly, we're missing that connection.

因此我很好奇，未来几年这个钟摆是否会继续向中心回摆，让人们重新认识到实体连接的重要性——与朋友共度时光，拥有一个能直接沟通解决问题的电话号码。

And so I'm curious to see over the next few years, does that pendulum continue to swing back a little bit more towards center where people are recognizing the importance of physical connection, spending time with friends, actually having a number to talk to someone to deal with any issues.

好的。

Okay.

在结束之前，我还有一个最后的惊喜。

I've got one final surprise before we wrap this up.

今天我们录制时，我截取了Seb和我对话的屏幕截图。

While we were recording today, I took a screen grab of Seb and I having our conversation.

我让它生成了我们的香蕉拉玛图（管它叫什么），用的是Pro Gemini模型。

I had it take our banana rama, whatever the heck it's called, a pro Gemini model.

香蕉。

Bananas.

还有纳米香蕉。

And nano banana.

谢谢你，先生。

Thank you, sir.

我让它想象一下，如果这两位播客身后有台相机，在他们交谈时拍张照片会是什么样子。

And I asked it to what would these two podcasters look like if there was a camera behind them, and it took a picture while they were having the conversation.

K？

K?

现在你将看到这张照片——我截屏得到的可能是你见过的最上镜的Seb照片了。

Now you're gonna see the picture that I the screen graph that I got is probably one of the most flattering pictures of Seb that you will ever see.

这张照片拍得太糟糕了。

This is such a bad picture.

快看这个。

Check this out.

好的。

Okay.

这就是你。

So here you are.

你当时正在眨眼抬头，而我则直勾勾地盯着镜头。

You were mid blinking your eyes and looking up, and I'm just stone cold staring at the camera.