Recursive 押注6.7亿美元于自我改进AI，Sonnet 5.5达到70%，埃隆联合主导五角大楼推进计划 | 第299期

所以他们可以把人类从未见过的图像输入扫描仪，生成预测的脑部扫描结果，然后在这些数据上进行训练。

So they can feed it images no human has ever seen into the scanner, generate predicted brain scans, and then train on those.

这个模型实际上是在自己构建数据集。

The model effectively builds its own data set.

所以理查德，你在书里说，AI在任何一个可以模拟或验证的领域都超越人类，而在其他领域则不然。

So Richard, you know, you say in your book that AI is superhuman in any domain you can simulate or verify, and nowhere else.

那么大脑是可以模拟的吗？大脑是一个可以模拟的领域吗？

So is a brain simulatable, is the brain a simulatable domain?

还是说这有所不同？

Or is this something different?

目前还不行，但这仍然是一个令人难以置信的成果。

It's not yet, but this is still an incredible result.

我还记得第一个这样的成果出来的时候，我当时还在读博士。

I still remember the first such result came out when I was still a PHD student.

我去了斯坦福的生物信息学系，你知道，那时候我们只能从中提取出非常模糊的小图像，而现在看到这样的清晰度和真实感，真是太不可思议了。

I went over to the bioinformatics department at Stanford and they, know, it was like very grainy little images we could just like extract from this and to see this fidelity now and this realism is just incredible.

我觉得其中一个问题，我不确定这项研究的情况，我今天还没看到，我一直在忙工作，但通常的问题是你得为每个大脑单独训练模型，因为每个人的大脑都有细微差别。

I think one of the problems, I don't know if this study, I haven't seen it today yet, I've been at work, but like one of the problems is often that you have to train for each brain because they're all slightly different.

所以你不能直接把一个从没做过fMRI扫描的人的数据拿过来更新模型，那样对那个人来说就太完美了。这对昏迷患者特别有用，我们想知道他们是否还在思考，你知道的，那将是非常了不起的。

And so you won't be able to just take someone who's never put into an f m r I scanner to see and update the model for that particular person, which would be otherwise amazing for people who are like in a coma and you want to see are they still thinking stuff, you know, like that would be incredible.

Alex，给我们多讲讲今天这个故事吧。

Alex, tell us more about this about this story today.

这其实是三月发布的一项成果。

So this actually is a result from March.

所以，从功能性成像、脑电图或电极数据中解码大脑状态这一更广泛的领域，到现在已经有很长很长的历史了。

So this the broader field of doing decoding of brain states from functional imaging or EEG or electrodes, this has a long, long history at this point.

我记得二十多年前，加州大学伯克利分校的加兰特实验室就在做这件事，用来解码猫的视觉皮层。

I remember twenty plus years ago at this point, the Gallant lab at UC Berkeley was doing this to decode the visual cortex of cats.

其实二十多年前，我们就开始能看到猫感知世界的图像了。

Actually, like twenty plus years ago, we were starting to see images of how a cat perceives the world.

我记得最早出来的那些模糊图像，比如一只猫看到树枝的场景。

I remember like the first grainy images that were coming out were like a cat seeing a branch.

再往后快进到非侵入性的功能性磁共振成像，现在有很多团队在做这个。

And then moving fast forwarding to noninvasive fMRI, there are so many groups at this point.

特别是Meta，一直在资助这个领域的不少研究，比如Jean Marie King，还有很多其他团队在做语言解码、视觉解码。再往后，就在过去二十四小时里，Neuralink刚刚宣布了他们的首个规模定律研究，用Neuralink的电子数据来预训练基础模型和前沿模型。

Meta in particular has been sponsoring quite a bit of work in this area from Jean Marie King, many other groups doing language decoding, vision decoding, More recently, even fast forwarding to just the past twenty four hours, Neuralink just announced their first scaling law studies on pre training foundation models, frontier models, off of Neuralink electronic data.

而且这是在单个患者的基础上进行的，但他们捕获的数据量非常大，时间分辨率也很高，比功能性磁共振成像能捕捉到的高得多。

And this is on an individual patient basis, but they're capturing copious amounts of data at high temporal resolution, much higher than fMRI can capture.

功能性磁共振成像，打个比方，用现在的技术，最多能提供立方毫米级别的空间分辨率，时间分辨率大约是一秒。

FMRI, like rule of thumb, fMRI can give you with today's technology, at best cubic millimeter spatial resolution for voxels and at best approximately one second temporal resolution.

所以这在一定程度上限制了你在解码一个人内部视觉状态方面的能力。

So that somewhat limits how well you can do in terms of decoding a person's internal visual state.

不过别误会，还是有人在做这个的……

You can still make no mistake, there are folks who have...

他们几乎把解码梦境、解码一个人视觉皮层中感知到的或幻觉到的视觉内容，无论是睡着还是醒着，都做成了一门小生意。

Are making a cottage industry out of decoding dreams and decoding visually what a person perceives or what they hallucinate in their visual cortex while they're sleeping or while they're awake.

而且我相信，用大量的功能性磁共振成像数据来训练前沿模型和基础模型，这个领域一定会非常活跃。

And I'm sure this is going to be a vibrant, vibrant space training frontier models and foundation models off of copious amounts of fMRI data.

但归根结底，我觉得这必须得往前走，因为fMRI解码在空间和时间上都有局限。

But ultimately, where I think this has to go, there are spatial and temporal limits to fMRI decoding.

要达到我们真正想要的目标，也就是完全沉浸式的VR和全带宽的脑机接口，就需要更高的时间精度和空间精度。

It's going to require higher temporal precision and higher spatial precision to get where we really want to go, which is full dive VR and full bandwidth BCIs.

我们最终会做到的，但好消息是，总得有个起点。

And we will get there, but the good news is you have to start somewhere.

大型语言模型也得从GPT-1开始，那只是一个能预测亚马逊评论情感倾向的单一人工神经元。

The large language models had to start with GPT-one, just a single artificial neuron that could predict the polarity of Amazon reviews.

同样，如果我们想要实现全带宽脑机接口，就得从Brain IT、Jean Marie King和Jack Gallant这些研究起步。

Similarly here, if we're going to get full bandwidth BCIs, it's going to start with studies like Brain IT and Jean Marie King and Jack Gallant.

太棒了。

Amazing.

Salim，如果有一天你能知道别人的想法，那世界会变成什么样？

Salim, what happens in the world where you can know someone's thoughts?

嗯，我觉得你……

Well, I think you you...

我们得小心一点。

Let's be careful.

我们这里解码的是感知，而不是真正的思想。

We're decoding the perception here, not the actual thought.

那要复杂得多。

That's like quite a bit more complex.

区别。

Difference.

那是什么……

What's the...

除了大脑区域，还有别的吗？

Other than the brain region other than the brain region?

这是没有区别的可视化。

It's the visualization without a difference.

而且我知道那里……

And I know there...

我觉得还是有区别的。

I think there there is a difference.

嗯，不是的。

Well, no.

当你想象的时候，对吧，当你在脑海里想象某样东西，它激活的神经元和亲眼看到时是一样的。

When when you imagine, right, when you imagine something in your mind, it lights up the same neurons as when you're seeing them.

那……

That...

那没问题。

That's fine.

那……

That...

那没问题。

That's fine.

只是……

Just...

我只是想确保我们不要把这两者混为一谈，因为你可以有想法，而不一定伴随着相关的感知。

I just wanna make sure we don't mix the two because you can have thoughts without necessarily the perceptions around it.

但你完全可以有一个而没有另一个，或者反过来。

But you can definitely have one without the other and the other without the other.

所以，但对我来说，这里更有趣的是，你知道，我们已经花了数十亿小时在小小的矩形框里打字。

So, but I think for me, the more interesting thing here is, you know, we've spent billions of hours typing things into little rectangles.

如果我们能改变这种界面，采用更自然的界面，也就是直接与大脑交互，我们在奇点大学的神经科学讲座中经常讨论这个，就是，你知道，我们对大脑如何运作仍然知之甚少，但你不需要完全了解它。

And if we can change that interface to do a more natural interface, which is interface directly with the brain, we used to talk about this in our Singularity University lectures on neuroscience, is, you know, we still have very little idea how the brain works, but you don't need to have...

只要你能有效地与它交互，你不需要知道它是怎么运作的。

Know how it works as long as you can interface effectively with it.

而我认为，这给了我们一个机会，让我们拥有真正深入的界面，然后这也会帮助我们弄清楚大脑是如何运作的。

And this, I think, gives us the opening to give us really deep interfaces, and then that'll help us figure out how the how the brain works.

我会接受这个。

I'll I'll take the

你提出了一个非常...

You bring up a really...

哦，理查德，你先说。

Oh, go ahead, Richard.

抱歉。

Sorry.

你提了一个很好的观点，我之前看到过一个研究，说人分两种。

You bring up a really good point, which is that I I read this result a while back that there are two types of people.

有些人思考的时候，脑子里真的会形成句子。

Some people who think actually think in sentences.

而另一些人，他们思考的时候，脑子里只有一团模糊的想法，只有当他们被要求说出来或者试着写下来的时候，才会把这团模糊的想法整理成真正的句子。

And other people, when they think, they just have a fuzzy thought cloud, and only once they're asked to verbalize their thought or they try to write it down, they actually, like, make a real sentence out of their fuzzy thought clouds.

我绝对是那种脑子里只有模糊想法的人。

And I'm definitely a thought cloud thinker.

我妻子则完全是那种脑子里有句子的人。

My wife is very much a, like, sentence.

比如，她脑子里思考的时候，真的是用完整的句子在想。

Like, in her head, she actually thinks in actual sentences.

而且，没有说哪一种更聪明之类的。

And, like, there's no there's no, like, one is smarter than the other anything.

它们其实是差不多的。

There's, like, they're, like, equivalent.

但我觉得有时候，写作就是思考，对我来说，如果我真的要把某个想法说出来，而不是只是脑子里有个念头，那会非常有帮助。

But, like, I do think sometimes, like, writing is thinking and, like, for me, it is very helpful if I have to really verbalize something versus I just have the thought.

我就在想，如果眼睛真的能直接提取我的想法，那世界会是什么样子，现在我需要越来越多地把我的想法云结构化成句子，或者……

I'm just trying to think, like, what does the world look like if the eye could really just like extract my thoughts, now I need to like structure my thought clouds into sentences more and more Or

理查德，如果你把所有文字都从神经网络训练中去掉，完全不用语言，只训练纯图像和更高层次的思维结构，那训练出来的神经网络会非常不同。

Richard, what would neural net training look like if you took all text out of it, no language at all, just train on pure images and higher level thought constructs, you'd get a very different neural net out the other side.

而且它可能真的会更像你那样思考，而不是像你妻子那样，因为……

And it might actually think a lot more like you do, unless like your wife does, because

也许我们应该持这样的观点：任何没有内心独白的人，都只是行尸走肉。

maybe we should be taking the position that anyone who doesn't have an inner monologue is just a pee zombie.

我会……

I'll do...

我还得再加一点。

I gotta throw one more thing in.

我记得我们请过一位世界顶尖的语言学家来演讲，我们当时觉得，这年头当语言学家肯定不好过。

I remember we had one of the top linguists in the world come in and speak, and we were like, you know, this must be a bad time to be a linguist.

现在谁还讲究拼写、语法这些啊。

Nobody's using spelling or grammar or anything else like that.

他说，不是的。

And he goes, no.

恰恰相反，现在是语言学家最有意思的时代。

On the contrary, this is one of the most interesting times to be a linguist ever.

我们都很惊讶。

And we're like, wow.

为什么？

How come?

他说，因为你看表情符号，这是人类语言史上第一次，你可以用符号来传递情感，因为你可以用数字方式发送情感。

And he said, because when you look at emojis, it's the first time in the history of human language that you can symbolically send emotion because you can digitally send emotions.

我们当时就想，哇。

And we're like, wow.

这挺有意思的……

That's an interesting...

所以他对打开那个全新领域感到特别兴奋。

So he was super excited by opening up that whole aperture.

我觉得这挺有意思的，不过——

I thought it was an interesting Well,

你知道还有什么对你来说有意思的吗？

you know what else for you?

你是说，你是说，你是说表情符号是语言学这么……的原因？

You're saying you're saying you're saying that emojis are the reason why linguistics is so...

不是因为语言学本身已经蓬勃发展起来了。

Not the fact that linguistics itself has been flourished

我是说，我在这里把这一点跟你联系起来。

by I'm AI connecting at this to you.

是你干的。

You did that.

不，不是我。

No, it's not me.

戴夫？

Dave?

哦不，我真正期待的结果是，大脑如何在没有梯度下降的情况下自我训练？

Oh no, the result I really am looking forward to is how does the brain train itself without gradient descent?

现在人工智能领域发生的一切，过去二十年人工智能领域发生的一切，都是由梯度下降算法驱动的，而生物学家在真实的生物学中找不到任何类似的东西。

Everything going on in AI right now, everything going on for the last twenty years in AI is driven by gradient descent algorithms, and the biologists can't find anything even vaguely like that in actual biology.

所以，我认为通过非常详细的成像，我们可能最终破解这个密码：改变突触权重的基本学习算法是什么？

And so that's, I think with really detailed imaging, we might finally crack the code on what is the fundamental learning algorithm that changes the synaptic weights?

这不是我们用于人工神经网络的方法。

It's not what we use for artificial neural nets.

这是另一种不同的方法。

It's something different.

如果我们发现了这一点，我们可能会发现神经网络训练的效率能提高十倍、百倍、千倍甚至百万倍。

And if we discover that, we might find that neural net training can be 10, a 100, a thousand, a million times more efficient.

所以希望这个结果能尽快出来，也许在……

And so hopefully that'll come out Could within a

你能给听众解释一下什么是梯度下降吗？

you define gradient descent for our listeners?

挺有意思的，伊利亚·苏茨克弗在接受采访时常说，其实只有一个算法。

It's funny, Ilya Sutskever, when he's on interviews, he's like, there is only one algorithm.

这个算法就是梯度下降。

The algorithm is gradient descent.

跟它比起来，其他东西都无关紧要……

Everything else is just irrelevant compared to...

所以现在的做法是，你搭建一个有96层或者120层的深层神经网络，里面全是些毫无用处的随机连接，然后给它喂上几万亿个训练样本，比如一万五千亿条。

So what happens right now is you build these 96 or 120 layer deep neural nets, and all it is is a whole bunch of random connections that do absolutely nothing useful, and then you give it trillions, 15,000,000,000,000 training examples.

你就告诉它：看到这张图就输出那个词，或者看到这段文字，下一个字应该是啥。

You're like, when you see this, say that, or when you see this paragraph, this is the next token.

如果你猜错了，就会通过梯度下降受到惩罚。

And if you're guessing wrong, you get punished through gradient descent.

所以，误差，也就是你偏离正确答案的程度，就是你的惩罚，它会反向传播到所有层，计算出误差，然后责怪每个神经元或每条连接，问它们对这个糟糕的答案负有多大责任。

So the error, how far off you are, is your punishment, and it passes back through all the layers, and it calculates an error, and it blames each neuron or each connection for how responsible were you for this terrible answer.

如果你错得很离谱，你就会调整，朝着能帮你答对的方向移动。

And if you're way wrong, you get moved, and you move in the direction that helps you get the answer right.

所以这是一个极其费力的搜索过程，每个突触连接的移动就是一个梯度，你就像爬山一样，爬到最佳状态。

So it's this incredibly laborious search process, and that movement of each synaptic connection is a gradient, and you just hill climb to the best possible state.

然后神奇的是，在花费大约一亿美元的计算资源后，理查德的情况可能是四亿美元，它就开始通过这个单一的算法——梯度下降——进行思考了。

And then it magically, after about $100,000,000 of compute, in Richard's case, I guess 400,000,000, it magically starts thinking through this one algorithm, gradient descent.

我觉得我们说的其实是反向传播，而不是梯度下降，对吧？

Think it's really that back prop that we're talking about synecdochically rather than gradient descent would be the obvious So

也许吧，是的，反向传播，我想，就是那些梯度的高效计算方法。

maybe, yeah, back prop is, I guess, the efficient computation of those gradients.

我想在戴夫的解释上补充一点，梯度下降本质上是一种优化算法，我们用它来训练各种机器学习模型，通过最小化误差来实现。

Maybe I'll add to this explanation from Dave, but, like, gradient descent is essentially an optimization algorithm that we use to train all kinds of machine learning models by minimizing the errors.

我觉得最好的比喻就是，假设你是一个盲人登山者。

And I think the best analogy is to assume you're a blind hiker.

你在山顶上，想找到最低的那个点，但你看不了多远。

You're at the top of the mountain and you're trying to find the lowest the lowest point, and you can't really see that far.

你看不见，但你能感觉到每一步的坡度。

You're blind, but you can feel the slope at each step.

所以，当你感觉坡度大致朝正确的方向时，你应该往下迈多大的步子？

And so how much, like, how big of a step should you take down when you feel like the slope is going roughly in the right direction?

问题是，这些AI试图优化的地形往往是非常非凸的，也就是说，不是只有一个最低点，而是有很多不同的低点。

And then the problem is that the kinds of landscapes these AIs are trying to optimize are highly non convex, which means there's not just one lowest point, but there are many different low points.

如果你……

And if you...

取决于你从哪里开始，你可能会走进不同的山谷。

Depending on where you start, you might go into a different valley.

取决于你从哪座山开始，你会进入这个山谷，而不是那个不同的山谷。

Depending on which mountain you start, you go into this valley versus this this different valley.

其实有很多非常有趣的类比。

And there are actually a lot of really interesting analogies.

我朋友从心理学角度来思考这个问题。

My friend,, like, thinks about this from the psychological perspective.

比如，当你患有创伤后应激障碍并接受训练时，你会过度拟合一种算法，你的大脑就会困在某种思维方式的低谷里。

For instance, when you have PTSD and you train, you overfit to one algorithm, now your brain is kind of stuck in one valley of how to think about something.

有时候人们会使用迷幻药等物质，这些已被证明对抑郁、创伤后应激障碍和其他心理疾病有帮助。

And it takes sometimes people like doing psychedelics and stuff that have shown to to help help with, like, depression and PTSD and other mental diseases.

比如，它们……

Like, they they...

它们有助于提高学习率，跳过另一座山，进入另一个吸引谷，在那里你会有新的思维方式。

It helps to increase the learning rate to jump over another mountain and and go into a different valley of attraction where you then have new kinds of ways of thinking.

所以这个盲人徒步者的类比，我觉得非常直观，对。

And so this blind hiker analogy, I think, is a really intuitive way Mhmm.

用来思考这个问题。

To think about it.

当然，问题在于这不仅仅是三维空间。

And then, of course, the problem is it's not just in three d.

它是在万亿维的空间里，比如你要在这样一个地形中穿越上万亿个参数。

It's in a trillion d, like a trillion parameters that you're trying to traverse in that landscape.

不过思路大概是，试着找出方向，然后朝着那个方向迈出步伐。

But the ideas of, like, try to identify the direction and take steps towards

即便如此，这个方法依然有效。

it still works.

你知道还有什么更惊人的吗？

You know what else is really incredible?

我能就这个话题聊上几个小时。但你看像 Kimi K3 这样的大模型，它有 93 层，经过反复训练。如果你把其中一层随机打乱，然后再进行训练，让它找回原来的状态，它却做不到。

I could riff on this for hours, but if you look at a big model like Kimi K3, it's 93 layers, and it's train, train, train, train, If you take a single layer and you randomize it, and then you train it and say, find yourself again, it can't find itself again.

它永远无法回到最初的状态。这就像在做 fMRI（功能性磁共振成像）观察人脑时一样，比如你说：‘这是 Alex 的大脑，我能把它和 Salim 的对比一下吗？’

It never gets back to where it was, which comes back to in these fMRIs when you're looking at a human brain, and you're saying, okay, here's Alex's brain, can I compare that to Salim's?

你会发现，哇，这两者之间几乎没有任何共同点。当然，也许这是个不太恰当的例子，但你……

And you're like, wow, there's nothing in common going on here, yet, you know, well, okay, maybe that's a bad example, but you're

你就信这个吧，因为这大概率是对的。

You're gonna take that because it's it's probably right.

而且是很聪明的那种对法。

Smartly right.

毕竟呢

The way

你大脑在学习过程中形成的神经连接是独一无二的，但结果却可能得出‘咱们足球踢得一样好’这样的结论。

your brain wires as you learn is completely unique to you, yet it can come out with like, okay, we're equally good soccer players.

但信号在我们网络中传播的方式完全不同，而且对每个人来说都是独一无二的。

But the way the signals are propagating through our networks is completely different and unique to each of us.

这真的非常奇怪，它竟然无法找回自己原本的状态。

Which is really, really strange, that it can't find its own way back to its state.

我觉得随着fMRI数据的不断涌入，我们实际上会学到很多关于它是如何运作以及为何如此运作的知识。

I think we're gonna learn a lot, actually, with the fMRI data coming in, on how and why that works.

肯定存在一些共性，只是我们目前还没能找到而已。

And there have to be commonalities that we just can't find.

旋转会让一切看起来不一样，但其实差别没那么大。

Rotations make everything look different, but they're really not that different.

如果你用变换来看它们，比如傅里叶变换的视角，我们就会突然发现，天哪，原来它们是这样对齐的。

If you put them through a transform, like a Fourier transform view, we'll suddenly say, oh my God, this is why they line up.

前几天我还和我们兄弟会的兄弟戴夫在一起，他是南加州大学神经生物学系的主任。

I was just with my fraternity brother of ours, Dave, who's the head of neurobiology at USC the other day.

现在是脑科学最激动人心的时代。

And this is the most exciting time ever for brain science.

哦。

Oh

我是说，从人类诞生起，大脑就一直是个黑箱。

Mean, brain has been a black box since humanity began.

而我们理解大脑、应对精神疾病的能力，我觉得将会非常惊人。

And our ability to understand the brain and deal with mental disease, I think is gonna be extraordinary.

嗯，所以你可以模拟，这其实就是Liquid AI创立的初衷，因为我们完全逆向工程了C。

Well, so you can simulate, this is why Liquid AI was founded actually, because we completely reverse engineered the C.

把线虫的大脑完全逆向解析，每一个细节都搞清楚了。

Elegans worm brain down to exactly every single thing going on.

然后他们就能模拟它了。

And then they were able to simulate it.

模拟完之后，他们发现，等等，这其实是一个非常高效的神经网络。

And then after they simulated it, they realized, wait, this is a very efficient neural net.

然后他们把它产品化了。

And then they productized it.

我是说，这就是，所以他们两个都……你继续，理查德。

I mean, that's, so they both, so go ahead, Richard.

有件特别酷的事。

There's a really cool thing.

既然你提到了线虫。

Since you mentioned C.

很少有人会提到这个。

Elegans worms, there's not many people that bring that up.

我有个朋友，是哈佛的神经科学教授，叫Sam Gershman，他……

A friend of mine who's a professor of neuroscience at Harvard, Sam Gershman, he...

他是我见过的最聪明的人之一。

One of the most brilliant people I've ever met.

他做过一个实验，把虫子切成两半，然后其中一半……

Like, he actually did a study with with worms where he cut them in half, and one...

C.

C.

秀丽隐杆线虫确实有大脑。

Elegans worms, they do have a brain.

而且它们身体其他部分还有很多神经元。

And then they have a lot of other, like, neurons in the rest.

你可以训练它们对某种刺激做出反应，然后把它们切成两半。

Now you can train them to react to a certain stimuli, and then you split them in half.

没有大脑的那一半会重新长出一个新的大脑，而且，你听好了，这太疯狂了。

The the second half without the brain regrows a new brain, And here, get this, this is crazy.

那个新大脑竟然保留了同样的记忆，太神奇了。

That new brain has the same memories Wow.

而且还能对不同的刺激做出反应。

And react to different stimuli.

所以他就想，也许我们学习的方式还有别的可能。

And so he's like, maybe there's a different way where we learn.

我确实觉得，现在我们对学习的理解太局限于神经网络、电信号这些了。

And I do think there are things that right now we're very much stuck in this, like, it's a neural net, it's all electrical signals and so on.

但很明显，大脑的化学状态可以因为很多因素发生巨大变化，比如你饿了会烦躁，或者感到疼痛，或者摄入某种兴奋剂，大脑立刻就不一样了。

But clearly, like, brain chemistry can change massively with just, like, you can get hangry, you can be in pain, you can have, like, a certain, like, stimulant, and all of a sudden your brain is very different.

而且我觉得还有……

And I think there are...

我最近有了孩子，跟一些妈妈聊过，发现确实有些行为模式会突然出现，比如你生了孩子之后就会想筑巢，就像……

And I've seen this now with, like, having babies and and, like, talking to moms, like, there's certain algorithms where you all of sudden you nest, because you had a baby and now you want to, like...

虽然每个人筑巢的方式不同，但大家都会以某种形式去筑巢。

And how you nest is different, but people nest, you know, in one form or another.

所以，我们DNA里还藏着一些潜伏的机制，过了二十多岁就会突然触发。

And so there are these latent algorithms that we have that are in our DNA that just trigger after, like, twenty something years.

只要身体和生理机能刚好达到那个状态，这些机制就会被激活。

When just the right things then happen with your body and your biology.

我觉得我们还有很多东西没搞明白，大脑的复杂性远超我们的想象。

So I think there's still so much that we don't, we haven't put So much complexity.

接下来，我们聊聊三天前的另一个故事。

I'm gonna move us to our next story from three days ago.

所以周二的时候，总统签署了一项行政命令，要求环保署、内政部和农业部，还要跟卫生与公众服务部合作，把华盛顿特区的入侵蚊子数量砍掉至少90%。

So on Tuesday, the president signed an executive order directing the EPA, the Department of the Interior and Agriculture, working with HHS to cut invasive mosquito populations in Washington D.

也就是D.C.

C.

并且到2028年之前，把蜱虫的数量减少至少50%。

By at least 90% and the tick population by at least 50% by 2028.

这些目标主要针对的是会传播登革热、寨卡病毒和黄热病的蚊子种类。

The targets include mosquito species that spread dengue Zika and yellow fever.

这就是让我眼前一亮的地方。

And here's what caught my eye.

你知道吗，这个行政令明确优先采用不育昆虫技术、安全的基因改造和有益细菌，而不是传统杀虫剂。

You know, the order explicitly prioritizes sterile insect techniques, safe genetic modifications, and beneficial bacteria over conventional pesticides.

我太喜欢这个了。

I love that.

你知道吗，这正是Colossal这家灭绝物种公司正在做的基因驱动工作。

You know, this is the exact kind of gene drive work that colossal, the extinction company is doing.

你知道吗，我觉得大多数人没意识到，就像我养孩子的时候，会问他们一个问题，谁……

You know, I think most people don't realize it's like, you know, when I was raising my kids, would ask them a question, who...

哪种物种是地球上杀死人类最多的？

Which which is the species that kills the most humans on the planet?

你知道吗，有些人会说是鲨鱼。

You know, and some people jump onto sharks.

有些人会说是，你知道的，其他什么动物。

Some people jump onto, you know, whatever it might be.

不过在这点上，我相信大家都知道蚊子是地球上最致命的生物。

But on this part, I'm sure people know mosquitoes are the deadliest life form on Earth.

确实如此。

As far.

光是疟疾每年就能夺去五十万人的生命。

Malaria alone kills half a million people a year.

所以你看，亚历克斯，这标志着生物学作为一门工程学科，已经正式成为联邦政策的一部分了。

So Alex, this is biology as engineering arriving as federal policy.

你对这个故事有什么看法？

Your thoughts on this story.

几十年来，我们一直对自己的影子感到恐惧。

For decades we've been scared of our shadow.

我在麻省理工学院有个同学叫凯文·埃斯费尔特，他是基因驱动技术的先驱之一。他为了让各州和地方政府批准针对蚊子的基因驱动研究，真是费了不少劲，但那些不关注这方面的人根本不知道这有多难。

So I have a classmate Kevin Esfelt at MIT who was a pioneer of the gene drive, one of the pioneers, and he's had a devil of a time getting states, municipalities to approve gene drive studies against mosquitoes and for those not tracking.

这项技术的基本原理是利用CRISPR插入一种能够自我复制的基因，通过蚊子种群传播，从而让蚊子绝育。

The premise of this technique is basically inserting a gene via CRISPR that wants to replicate itself to basically sterilize mosquitoes through propagating virally through the mosquito population.

我们已经有这项技术了。

We have the technology to do this.

缺少的，至少在这个国家，是一项联邦指令来真正实施对非人类动物种群的大规模基因工程。

What it has lacked, at least in this country, is a federal mandate to actually go and implement large scale genetic engineering of non human animal populations.

而现在，看起来是第一次，我们通过行政命令看到了这样的指令。

And now, for the first time it seems, we're seeing via executive fiat just that mandate.

这令人兴奋，原因有二。

And it is so exciting for two reasons.

第一点，彼得你刚才提到疟疾和其他非人动物传播的疾病造成了巨大的人类痛苦，同时也给动物带来了巨大的痛苦。

One, because one is, Peter, you mentioned malaria and other non human animal borne diseases cause an enormous amount of suffering, human suffering and also non human animal suffering.

其次，因为现在有一种方法可以做到这一点，而且历史上我们并不需要直接杀死这些动物。

But secondly, because there's a way to do this that doesn't actually involve killing the animals themselves historically.

回想五十年前，如果你想应对昆虫传播的疾病，通常的做法就是喷洒化学药剂。

Historically, you know, you look back fifty years, you'd see chemical spraying if you wanted to do something about, say, insect borne illness responses.

我们现在不需要再那样做了。

We don't need to do that anymore.

我们其实可以保留那些无意中携带细菌或病毒的昆虫。

We can actually keep the insects that are the inadvertent carriers of bacterial or viral disease.

我们既能保护它们的生命，又能阻止疾病传播。

We can actually preserve their lives while also preventing the disease transmission.

我觉得这对它们来说也是好事。

And I think that's good from their perspective as well.

最让我兴奋的是，我们以前都是喷洒化学药剂。

That was the part that got me most excited was we spray things.

我们基本上是在毒害生物，而现在

We're basically poisoning biology, and now

我们转向真正的编程。

we shift to actually programming.

系统。

System.

然后抛弃有毒的生态系统。

And leave the poison ecosystem.

理查德，你对这个故事有什么看法？

Richard, your thoughts on this story?

我同意大家的观点。

I I concur with everyone.

这真是太棒了。

Like, it's it's it's wonderful.

而且非常有趣。

And it's so interesting.

比如说，人们……

Like, people...

我希望我们能更多地传播这类故事。

I I hope we can amplify these these kinds of stories more.

你知道吗？

You know?

未来需要更好的宣传。

Like, the future needs better marketing.

这个世界需要更多的彼得。

We need more Peters in the world.

这只是众多故事中的一个，你知道的，比如治愈各种疾病、利用合成生物学技术，拯救动物生命，防止它们仅仅被繁殖出来用于实验和解剖。

And and this is one of those many stories, you know, curing various diseases, peril bio, like, saving animal lives and and prevent them from just being bred, to be tested upon and dissected.

像这样的故事，我们真的需要在公众视野中更多地传播开来。

Like, those are all, like, stories we need to amplify more in the public eye.

戴夫，我迫不及待想看到这些从华盛顿特区扩展到佛蒙特州、马萨诸塞州以及全美各地。

Dave, I can't wait for this to come from DC to Vermont and Massachusetts and all of these

我想对听众们说，如果你对你们的房地产基金感兴趣的话，这个基金的主题是：通过无人机轻松抵达山顶和岛屿。

I was gonna say for the listeners that are interested in your real estate fund, which, you know, the theme there is, hey, easy access via drone to hilltops, islands.

但另一方面，如果你住在沼泽边缘，你的房子价格可能只有非沼泽地区同类房产的一半。

But also, if you live on the edge of a marsh, your house probably sells for about half the price of a place that's not on the edge of a marsh.

这种情况会消失的。

That'll go away.

现在有很多针对叮咬性昆虫的解决方案，这只是其中之一。所以如果那片土地本身非常美丽，只是以前很难居住，它的价值将会飙升。

There's so many solutions coming to biting insects, this being one of them, that that also is, you know, if it's beautiful, beautiful land that otherwise is very difficult to live on, it's gonna go through the roof in value.

我确实觉得，尽量让蜜蜂活着真的很重要，对吧，别乱喷一堆农药，确保鸟儿还有足够的虫子吃，诸如此类的事情，是的。

I do think it's really important to try to like, you know, keep the bees, like, alive, all right, and not spray a bunch of pesticides, make sure birds still have enough insects to eat and things like that, yeah.

我想说的是，可能长达八十年，人类一直对自己影子感到害怕。

I was going to say for maybe up to eighty years, humanity has been scared of its own shadow.

整个人类，特别是美国或西方，核能就是另一个例子，我在播客里也提过，我们可能因为太害怕炸弹或者裂变反应堆，白白失去了五十多年的进步。

Humanity in general, America or the West in particular, nuclear energy, another example I've pointed out on the pod where we arguably lost fifty plus years of progress because we were too scared of either the bomb or the fission reactors or some...

整整一代人看了《中国综合症》这部电影，然后就不必要地害怕核能了。

An entire generation watched the movie China Syndrome and then got scared of nuclear power unnecessarily.

这个也是一样的道理，或者地球工程，也是另一个例子。

Same idea with this or or with geoengineering, another example.

这个世界仍然在害怕。

The the world is still scared.

世界上有一部分人害怕改造天气。

Some fraction of the world is scared of engineering the weather.

我们不需要害怕改造我们的物理世界或生物世界。

We don't need to be scared of engineering our physical world or our biological world.

现在，希望人们能看到人类正在走出这种自我担忧阶段的曙光。

And now, hopefully, one can see green shoots of humanity getting past that stage of worrying about its own shadow.

这很有意思……

This is interesting that...

抱歉。

Sorry.

这真的很……

It's really...

确实有意思，有时候那些危言耸听的人觉得自己在做正确的事，就像在说，哎呀，情况能有多糟？

It's it's interesting indeed that, like, sometimes alarmists feel like they're doing the right thing by just, like, oh, how bad could it be?

我在提醒人们有坏事要发生。

I'm warning people of something bad.

但这确实可能让整个人类远离真正的好东西，比如核能带来的能源丰富。

But, like, it can indeed push all of humanity away from something really good, like energy abundance with nuclear.

我觉得人口过剩是另一个大误区，人们会说，如果人太多，各种资源就会短缺，一切都会变得更贵，人们会……

I think overpopulation is one of these other big myths where people are like, well, if we have too many people, there's gonna be scarcity of all these different resources, and everything's gonna get more expensive, and people are...

更多的是贫困之类的问题。

It's more in poverty and so on.

但实际情况恰恰相反。

And the exact opposite happened.

其实有一个网站，我觉得你们都会喜欢，叫 humanprogress.org 或者 .com，它展示了尽管人口更多，很多这些东西其实越来越便宜。

There's actually one website that I think you all would love, which called humanprogress.org or .com, like, that that shows that a lot of these things are actually getting cheaper and cheaper despite there being more people.

嘿，理查德。

Hey, Richard.

你在德国可是个英雄。

You're you're a hero in Germany.

在德国，你简直是个走在街上都会被认出来的英雄。

Like like, can't walk down the street kinda hero in Germany.

但你怎么看德国正是因为这个问题而缺电？

But what do you think about the fact that Germany has no power for exactly this reason?

这真的很遗憾。

It is really unfortunate.

如果德国有很多人想以一种奇怪的方式，从进步中退出来，认为所有消耗能源的东西都对环境有害，那么这种源自普遍善意环保理念的某种‘去增长’倾向，真的让我挺担心的。

If there are a lot of people in Germany that want to, in a weird way, sort of off ramp from progress, and they think that everything that consumes power is bad for the environment, there's, like, a weird sort of degrowth, offshoot from generally, like, well intentioned pro, like, environmental vibes that just worry me worry me quite a bit.

我们待会儿会聊聊末日论，但还没到时候。

We're gonna talk about doomerism in a little bit, but not yet.

你知道，我很久以前就意识到一件事：人类特别擅长预见未来会出现的问题，对吧？

And, you know, one of the things I realized a long time ago is we humans are really incredible at seeing a problem out in the future, right?

我们会看到酸雨、人口过剩、能源短缺等等各种情况。

We see acid rain, we see overpopulation, we see, you know, energy shortage, whatever it might be.

然后因为我们的杏仁核，因为我们思考的方式，我们把未来的问题提前到了今天，于是我们惊慌失措。

And then because of our amygdala, because of way we think, we accelerate that future problem to today, and we freak out.

我们忘记了这样一个事实：等到我们真正面对那个问题时，我们已经取得了十年的进步。

And we forget the fact that there is a decade worth of progress we're gonna make by the time we reach that problem.

这就是创业者做的事情。

And that's what entrepreneurs do.

他们不断地解决问题，一个接一个。

They solve problem after problem after problem.

我想让我们的听众听到这一点，因为如果你担心未来的某个问题，请理解，有一个极其高效的创业和资本主义市场，它会解决世界上最大的问题，也就是世界上最大的商业机会。

And I guess I just want our listeners to hear that because if you're worried about some future problem, please understand there's a incredibly efficient market of entrepreneurship and capitalism that will solve the world's biggest problems, world's biggest business opportunities.

而且我们有一个美好的、向前推进的解决方案体系。

And it's a beautiful forward, you know, forward propagation solution set that we have.

也许我们稍后会谈到这一点，但我在AI末日论中也看到了同样的情况，他们编造出越来越复杂的场景，攻击者拥有近乎神奇的攻击能力，但不知何故，故事中的防御者从来没有近乎神奇的防御能力，这很奇怪，比如，如果我能制造一种超级病毒，完美地无法检测，传播到全世界，没人注意到，然后……

And maybe we'll get there later, but I I see the same thing happen in AI doomerism, where they create more and more complex scenarios where attackers get these, like, near magical abilities to attack, but somehow the defenders in those stories never get near magical abilities to defend, and it's this weird thing of like, man, if I can create a super virus that is like perfectly undetectable and spreads all throughout the world and no one notices it, and like...

然后它还有一个WiFi开关，你只要打开和关闭它就能杀人，那我也会制造一种超级神奇的疫苗，能让你对所有这些东西免疫。

And then it has this WiFi switch and you just turn it on and off and then kill people, like, I I will create a super magical vaccine that just inoculates you against all of these things.

如果你真的能假设……

Like, if you can actually assume...

做出那个假设，那么，这很奇怪……

Make that assumption, then, like, it's a weird...

很奇怪。

Weird thing.

本期节目由Google for Startups赞助。

This episode is sponsored by Google for Startups.

大家不妨花点时间想想。

Think about this for a second.

现在，你能够使用那些训练成本高达数亿美元的生成式AI模型了。

You now have access to the same generative AI models that cost hundreds of millions of dollars to train.

Google为初创企业提供的生成式媒体技术指南，为你提供了在生产环境中部署Google DeepMind模型的完整方案。

Google's Startup Technical Guide for Generative Media gives you a complete blueprint for deploying Google DeepMind's models in production.

无论是图片、视频还是音频，全都涵盖在内。

Images, video, audio, all of it.

真正的架构，真正的成果。

Real architecture, real results.

链接在下面的节目简介里。

Find the link in the show notes below.

接下来我要讲一个最近二十四小时在网上闹得沸沸扬扬的故事。

I'm gonna move us to a story that's been sort of breaking the Internet over the last twenty four hours.

理查德，这正好拿来检验一下你那个关于监管应用的论点。

And it it puts your regulate the application argument to the test, Richard.

有一家叫Tavus的公司发布了Griffin，他们称这是第一个通过视频图灵测试的模型。

So a company called Tavus unveiled Griffin, which it calls the first model ever to pass the video Turing Test.

来看看这些数字。

So here's the numbers.

48%的人在实时面对面视频中与AI虚拟人交谈时，以为自己在和真人说话。

48% of people who talk to an AI avatar on a live face to face video thought they were talking to a real human.

之前的系统只有3%。

Previous systems were at 3%.

我在这里放一下视频，非常酷。

So let me show the video here, and it's super cool.

我们接下来聊聊这个。

And let's talk about it next.

现在，只竖个大拇指怎么样？

Now how about just a thumbs up?

当然可以。

Sure.

好的，马上给你竖个大拇指。

One thumbs up coming right up.

太棒了！

Hell yeah.

那真是太酷了。

That was awesome.

我得说，你是我最喜欢的同事。

I would say you're my favorite coworker.

能和你一起做这个项目，真的非常开心。

It's been it's been really great to work on this with you.

你让我都不好意思了。

You're making me blush.

那个穿着丑毛衣的人是谁？

Who's that with the ugly sweater?

西蒙说，摸摸你的头发。

Simon says touch your hair.

这是你要找的东西吗？

Is this what you're looking for?

好的。

Okay.

不错。

Nice.

西蒙说，指一下。

Simon says point.

好了。

Alright.

我正在做呢。

I'm doing it.

现在摸摸你的下巴。

Now touch your chin.

我才不会上当呢，你没说那个魔法词。

I'm not gonna fall for that, you didn't say the magic words.

被你抓到了，好吧。

You got me, okay.

把松动的连接器重新装回主板上。

The loose connector back onto the motherboard.

好的。

Okay.

让我去拿那个。

Let me go let me go grab that.

但在此之前……

But beforehand...

我已经把烙铁打开了。

I turned the soldering iron on.

我觉得大概需要十二秒，所以等它好了告诉我一声就行。

I think it's gonna take about twelve seconds, so maybe just let me know when that's done.

听起来不错。

Sounds good.

我会盯着它，一到十二秒就告诉你。

I'll keep an eye on it and let you know the second it hits that twelve second mark.

好的。

Alright.

真有意思。

So interesting.

Tavus 称它为第一个人类交互模型。

So Tavus calls it the first human interaction model.

它支持全双工，能同时听和说，就像人类真实交流那样。

It's fully duplex listening and talking at the same time that you do the way humans actually do.

它在 NVIDIA 的全双工 AI 视频基准测试中排名第一。

It's number one on NVIDIA's benchmark for full duplex AI video.

Tavus 把它宣传为“每个学生的导师，能发现他们走神；也是倾听的老年护理伴侣”。

Tavus pitches it as quote a tutor for every student that notices when they're lost an elder care companion that listens.

Tavus 自己说，Griffin 在公开发布前需要先做安全研究，因为它是第一个可能被误认为真人的模型。

Tavus itself says Griffin requires safety work before it can be released publicly because it's the first model that can be mistaken for a real person.

萨利姆，你对这个有什么看法？

Salim, your thoughts on this one.

其实我并没有被它的图像和模仿能力震撼到。

So I'm less kind of floored by the imagery and the the mimicry of it.

这是意料之中的事。

This was expected to happen.

对我来说，我更感兴趣的是它能不能完成一些足够有用的事情，让我不在乎它是不是人类。

For me, I'm more interested in can it accomplish something useful enough that I don't care whether it's human or not.

我觉得那更有意思，也许将来会出现这样的东西。

And I think that that is more interesting, which might come along and have...

总有一天会实现的。

Have that happen at some point.

我们之前讨论过，在不久的将来，比如现在，你会有数字同事出现在Zoom上，你可以打电话、发短信、用Slack联系他们，他们会和你互动。

We've talked about the idea that in the near future, like now, you're going to have digital coworkers that pop up on Zoom, that you call, that you text, you Slack, and interact with you.

我觉得作为未来虚拟公司的一种机制，这挺有吸引力的。

And I think it's pretty compelling as a mechanism for a future virtualized company.

亚历克斯？

Alex?

嗯，彼得，你首先应该让我摸一下我的脸。

Well, first Peter, you should probably ask me to touch my face.

好的。

Okay.

摸一下你的脸。

Touch your face.

西蒙说，摸一下你的脸。

Simon says touch your face.

很好。

Very good.

我只是确认一下，但我想塔瓦斯已经领先我了。

Just checking, but I I guess Tavus is ahead of me.

我觉得，一方面，如果你看看中国实验室，比如阿里巴巴实验室出来的模型，一个主播就让我们预览了将要发生的事情。

I I I think on the one hand, if you look at models coming out of Chinese labs like Alibaba's lab, one streamer gave us a preview of what was going to happen.

在生成式视频模型和交互式生成式视频模型方面，中国实验室的投资力度相对西方实验室来说还是更重一些。

The Chinese labs remain over invested relative to the Western labs in generative video models and interactive generative video models.

所以我觉得，那个流媒体演示算是展示了未来可能性的一个预览。

So one streamer, I think, is a preview of what's possible.

我猜我是看到了关于 Tavus 发布的一些信息。

My guess is I read the information that was put out around Tavus.

我觉得从整体情况来看，先……

I I think in full generality, let...

咱们来聊聊这事儿最终会走向何方。

Let's just talk about where this is going to end up.

短期内的情况很难预测。

It's it's very difficult to predict the short term.

但长期来看，其实挺容易预判的。

I think it's pretty easy to predict the long term.

长远来看，端到端的生成式像素技术会带来那种神奇的镜子效果——可以实时、完全互动，逐像素生成；或者像Anthropic希望的那样，按矢量或程序化生成。总之，就是完全由算法实时生成的互动视频模型。

In the long term, end to end generative pixels will have these magic mirrors that could be real time, fully interactive, pixel wise generated, or if if anthropic has its way, vector wise or procedurally generated, but either way, fully generated real time interactive video models.

而且你能够创建一个包含人物的场景，就像现在技术上已经可行，但延迟很高。

And you'll be able to create a scene that consists of people like it's possible right now, but the latency is high.

你看，像One Streamer这样的模型已经发布并且开源了，或者像Tavus Griffin这种还没真正发布、肯定也不开源的模型，你都能看到未来的雏形。

You see with one streamer, which is already out and already open source, or with this Tavus Griffin type model, which is not really out yet and definitely not open source, you see a preview of the future.

这看起来会是什么样子呢？

What does this look like?

嗯，有几种不同的可能性。

Well, there are a few different ways it can go.

要看看每个token能带来多少收入。

Query how revenue generating per token it is.

它是否接近内容生成的最佳性能边界？

Is it anywhere close to the optimal frontier of cogeneration?

恐怕不是。

Doubt it.

另一方面，如果生成任意场景变得极其便宜，比如生成参加Zoom会议的人或者参加播客的人，那我们真的在乎它是否接近最优性价比吗？

On the other hand, if it becomes so absurdly inexpensive to be able to generate arbitrary, say, humans participating in a Zoom meeting or humans participating in a podcast, do we really care whether it's close to being near the optimal cost performance frontier?

也许吧。

Maybe not.

有很多人类服务行业的岗位需要一张脸、需要互动、需要声音，还需要能随时按要求做出表情，这些工作很可能完全被一个模型自动化掉，而这个模型本来只能做文字交互，没有脸也没有声音。

There are a lot of human service industry jobs that require a face and require interactivity and a voice and the ability to touch one's face apparently on demand on on request that could be probably completely automated away by a model that otherwise would be limited to text based interaction but doesn't have a face and a voice.

所以在最乐观的情况下，这类交互式视频模型——顺便说一句，它的缩写，HIM，是Alex Finn先用的，很明显是在致敬那部电影《她》——

So in the most optimistic scenario, these sorts of interactive video models, open paren, note that the acronym for this, which by the way was there first with Alex Finn, him, h I m, is such an obvious reference to her, the the movie, close paren.

我觉得这这这

I I think that this this has

这这这你应该用破折号，拜托。

this has You should use you should use em dashes, please.

好的。

Okay.

我会深入讲讲。

I'll delve into it.

所以我觉得这会对服务业产生变革性的影响。

So I I I think I I I think this is going to be transformative for the service sector.

而且，希望Tavus以及美国西部更广泛的视频前沿模型和交互式视频前沿模型，能借鉴Tavus的做法，开始与中国竞争。

And, hopefully, Tavus and broader American west of video frontier models and interactive video frontier models, take a page from Tavus and start competing with China.

Richard，你书里的一个原则，你的一条规则是，当你能验证答案时，AI就变得超人类。

Richard, one of the principles in your book, one of the rules you have is that AI becomes superhuman where you can verify the answer.

你知道，像这样通过人类的科学里程碑，通过人类的判断和互动，是否足以作为一个科学里程碑来验证？

You know, is passing a human scientific milestone like this, is passing a human judgment and interaction sufficiently as a scientific milestone for verification?

不完全算。

Not quite.

有点像……

It's sort of like...

当需要人类介入来判断“这像不像人类，对面是不是另一个人”时，就很难规模化。

It's hard to scale when a human has to be in the loop to say, this is like a human, another person on the other side or not.

你没法完全自动化，并以那种自动化的方式去验证。

You can't quite automate it completely and verify it in in that automated fashion.

我确实觉得你说得对。

I do think you're you're right.

这是一个很好的例子，我觉得我们需要做到……

This is a good example where I think we need to do be...

超越KYC，做到KYU，也就是了解你的使用场景。

Go beyond KYC and do KYU know your use case.

比如说，你肯定不希望这项技术出现在Zoom里，假装是CEO，而且确实有一些著名的案例，已经发生过这种事，他们让人转了大概两千万美元。

Like, you don't want this technology to be in Zoom pretending to be the CEO, and, like, there's some famous stories where this has already happened where they got someone to wire, like, $20,000,000

当然。

Sure.

因为他们创建了一个包含五位高管的Zoom会议，而且他们都……

Because they created a Zoom of five other executives, and they all, like...

他被骗得够惨的。

He was fooled well enough.

所以我觉得我们需要小心，Zoom和Google Hangout可能得开始采取反制措施，来识别这是不是真人，诸如此类，防止这类黑客攻击发生。

And and so I think we need to, yeah, to be careful, and like Zoom probably and Google Hangout need to start, like, finding countermeasures to identify, is this a real person, and and things like that to prevent those kinds of hacks from happening.

是的，这类诈骗会越来越多。

Yeah, scams are going to proliferate on this.

我想再提醒一下大家。

I just want to do a call out again.

我之前在播客里说过，如果你还没听过，而且你的祖父母或父母还健在，你还没花时间录下他们的故事，用视频和音频，深入聊上几个小时。

I've said this before on the pod, and if you've not heard this before, if you still have your grandparents with you, or your parents, and you've not taken the time to record their stories on video and audio and go deep spend, you know hours recording them.

你能为你的孩子、孙子或曾孙创造一个超高清晰度、栩栩如生的虚拟形象，我觉得这太棒了，但你需要数据。

Your ability to create a super high resolution lifelike avatar for your kids or your grandkids or your great grandkids, I think is an incredible thing, but you need the data.

所以，如果你在听这个节目，而且你很幸运，父母或祖父母还在身边，那就去收集他们的数据吧，因为这对你的后代和他们来说，都是一个绝佳的机会。

So, if you're listening to this and you're lucky enough to have your parents or grandparents around collect their data, because it's going to be a beautiful opportunity your progeny and theirs.

另外，我还想补充一下，给他们办个LCOR会员，保存他们的连接组，如果有这个考虑的话。

And or I should add, get them an LCOR membership and preserve their connectome if there is a consideration.

何必只停留在行为数据呢？

Like, why stop with just the behavioral data?

把整个父母都保存下来。

Preserve the whole parent.

你知道，我的孩子们都在深度冷冻中，或者至少他们的胎盘细胞是这样。

You know, kids are my kids are in a deep freeze or at least their placental cells.

我对此有些保留意见，主要是基于刚才的讨论。

I have a pushback on that just from I have a pushback on that just from this conversation.

请讲。

Please.

如果从之前的故事里能读取人脑，那你要做的只是想象一下你祖母的样子，然后从中提取出来就行。

If from the earlier story we can read people's brains, all you have to do is picture an image of your grandmother and then extract it from that.

那精度会非常低。

It'll be low low fidelity.

我是说，从第一性原理来看，祖先模拟就是这样运作的。

I mean, do think like to to first order, this is how ancestor simulation will work.

你看，就在过去四十八小时里，到处都是人们用 Opus 5.5 或其他前沿模型重建历史片段的新闻。

You see stories left and right just in the past forty eight hours of people using Opus 5.5 or or other frontier models to reconstruct bits of history that would...

这些片段原本仅凭当时的文物是根本无法得知的。

Were otherwise unknown just based on artifacts of the day.

但到了某个阶段，我确实认为，Salim，你不仅仅想要这个人的记忆，你想要的是这个人本身。

But at some point, I I do think, Salim, you want to go beyond just the memory of the person and you want the actual person.

我还要再强调一下，除了彼得，你保存孩子的胎盘真是很慷慨，但未必是保存你的孩子。

And I'll pound the the drum again in addition to Peter, you know, very generous of you to preserve the placenta associated with your children, but not necessarily your children.

我相信胎盘会感谢你的。

I'm sure the placenta placenta I will will thank you.

但是我们……

But we...

请继续，彼得。

Go go ahead, Peter.

但我想说，是的。

But but I I I would say, like, yeah.

如果这真的是一个严肃的担忧，你担心你的父母或祖父母，并且你想超越仅仅拥有他们的录音或基于AI提示生成的祖先互动模型，那就去给他们办LCOR会员，让他们冷冻保存。

If if if this is a serious concern, you're worried about your parents or your grandparents and you want to go beyond just having recordings of them or AI prompted generative interaction models of your ancestors, go get them LCOR memberships and get them cryopreserved.

理查德，我真的很想问你一个问题。

Richard, really curious to ask you a question.

我们显然已经通过了图灵测试，萨利姆的反应就是我的反应。

We clearly passed the Turing Test, and Salim's reaction is my reaction.

我们刚才看到的，就是一个“那又怎样”的例子。如果你每天都在用这些模型，整天都在用，你知道你只需要把各个组件拼起来，然后你就有……

What we just saw is a so what if you've been using models every day, all day long, you know you just had to stitch together the components, and you have...

但我认为这会让主流世界看到，我们已经跨越了图灵测试。

But I think this will show mainstream world that we've crossed the Turing Test.

然后下一个里程碑就是德米斯·哈萨比斯测试，就是利用1910年之前的信息，重新发现E=mc²。

And then the next milestone is the Demas Hassabis Test, where using information from 1910 and prior, rediscovery equals MC squared.

所以你不能作弊，这真的很难衡量，因为作弊就是，你知道的，但如果不作弊，只用之前的信息，得出E=mc²。

So you can't cheat, which is a really tough one to measure, because cheating is, you know, but without cheating, only using prior information, come up with the equals MC squared.

那么，有没有一个理查德·索赫尔测试，是一个真正有趣、可衡量的里程碑，即将到来呢？

So is there a Richard Socher test that's like a really fun, measurable milestone coming?

我提出了一个反图灵测试，实际上我觉得整个图灵测试已经反转了，现在为了知道对面是不是人类，你反而会问一些难到人类根本答不出来的问题。

I came up with the like anti Turing Test, which actually I think the whole Turing Test has flipped, where now in order to know whether there's a human on the other side or not, you actually ask it questions that are so hard no human could ever answer it.

比如，你让AI给你写一个复杂的网页应用，然后十秒钟后，它给你返回了五万行代码。

Like, if you asked an AI to just write you a complex web app and, like, ten seconds later, it comes back with, like, 50,000 lines of code.

你知道的，你大概就知道那不是人类了。

You know, like, you kinda know it was not a human.

所以我觉得这个测试其实已经完全反过来了。

So I think that test has actually completely flipped.

你不觉得，理查德……

You you don't think, Richard...

我是说，比如你……

I mean, like, you...

它可以自我限制速度。

It could throttle itself.

那个很容易被破解，但就像那个篮子一样，现在就像……

That one's easy to defeat, but just like the basket It's just like now

其实就是关于伪装。

it's just about it's just about fakery.

它确实有点像……

It did like...

它之所以成为人工智能的智力测试，是因为要像人类一样聪明太难了。

The reason it was an intelligence test for artificial intelligence was that it was so hard to be as smart as a human.

现在你只需要把能力限制在人类水平就能通过测试，这让这个测试作为衡量智力的标准完全失去了意义。

Now you just have to throttle yourself down to human level in order to pass the test, makes the test useless as an inspiring test for intelligence.

对，我

Yeah, I

我觉得反直觉地看，最好的验证方法——当然这不是建议——就是问它一个跟生化核放射相关的问题，看看它能不能回答，以此来判断你是不是在和文字聊天机器人交互。

think perversely probably the best way to, like the best CAPTCHA for knowing whether you're interacting, this is not a recommendation, but best way to know whether you're interacting with a text based chatbot is probably to ask it a CBRN related question and see if it's capable of responding.

估计不行。

Probably not.

我再快速跟我们的听众强调一下。

Let me just hit real quick again for our listeners.

如果你怀孕了，或者家人有孕妇，记得把胎盘细胞存起来。

If you're pregnant or someone in your family is pregnant, you know, save their placental cells.

我投资组合里的一家公司叫Life Bank USA，就是做这个的。

One of my portfolio companies called Life Bank USA does this.

我就是在那儿存的孩子的胎盘。

That's where I saved my kids' placenta.

你知道吗，胎盘就像是制造婴儿的3D打印机，它含有干细胞、自然杀伤细胞、T细胞和外泌体。如果我的孩子将来需要任何生物增强或器官，你就有原始的启动盘，用个老词来说，就是你孩子DNA的备份。

You know, the placenta is the three d printer that manufactures the baby, and it has the stem cells, natural killer cells, T cells, exosomes, and it's like from those cells, if my kid should ever need any kind of biological enhancement or organs, you've got the original boot disk, to use an old term, for your kid's DNA.

所以这就是Life Bank USA。

So that's Life Bank USA.

只是……

Just...

我觉得这就像父母的一种道德义务，应该保存这些细胞作为备份。

I just think it's like a moral obligation parents should have to to save those cells as a backup.

作为你孩子的备份，你能用它来替换……

As a backup for your children, could you use it to replace

你的孩子，彼得？

your children, Peter?

当然可以，你绝对可以用它来克隆你的孩子。

You could you could use it to clone your children for sure.

现在他们知道你为什么存这些了。

And and now now they know why you're banking them.

没错。

Exactly.

一支年轻的迪亚曼迪斯大军。

An army an army of young Diamandis.

理查德，接下来我们聊聊你在Recursive正在做的核心事情。

Richard, let's jump next into the core of what you're building at Recursive.

也就是递归自我改进和超级智能。

Namely, Recursive, Self Improvement, and Superintelligence.

那，我先问几个关键问题来开场吧。

So, let me ask a few sort of key questions to kick this off.

首先，我们现在在RSI方面进展如何？

So first, where are we at the moment with RSI?

第二，你怎么定义ASI？

Number two, how do you define ASI?

这是个长期争论的话题。

It's been a long standing debate.

那么，我们离ASI还有多远？

And then, how far away are we from ASI?

你能回答这三个问题吗？

So, can you hit those three?

我觉得有……

So I think there's...

那我们现在在RSI方面进展如何？

So where are we at the moment with RSI?

这是……我们是不是……

This Are we

今年发生了一个重大转变，那就是AI现在能写代码了，对吧？

year something major shifted and that is AI can now code, Right?

这大大提升了它自我改进的能力。

And that is a major shift in the ability for it to change itself.

你知道，我们现在基本上可以充分利用AI本身就是代码、而且AI能写代码这个事实。

Know, like we now are able to essentially lean into the fact that AI is code and AI can't code.

所以这里就形成了一个可以闭环的循环。

So you have a loop that you can close there.

因此，在各种弱形式下，我们已经有了RSI。

And so in various weak forms, we already have RSI.

就连Anthropic和OpenAI在谈论时也常提到，比如看看我们的员工、工程师、程序员有多少人在用Codex或Claude来写代码。

And the weak forms that even Anthropic OpenAI talk about when they often talk about is like, look, how much our employees, our engineers, our programmers use codecs or Claude to create some code.

我认为这其实是递归自我改进的一种弱形式，因为人类仍然深度参与在这个循环里。

And I would argue that that is a weak form of Recursive self improvement because you still have, like, deeply embedded humans in that loop.

我们在Recursive公司做的事情，就是让人类只负责设定奖励、环境和目标，然后让AI完全掌控整个流程，包括构思、实施和验证想法，并让它拥有完全的控制权，从而让所谓的开放式算法、进化搜索算法，能够将不同的有趣想法结合起来，真正蓬勃发展。

And what we're working on at Recursive is to have humans only be involved in setting up the rewards and the environment and the goals, and then allow the eye to have the entirety of the process of ideation, implementation, and validation of ideas and have full control over that and allow for a so called open ended algorithms, evolutionary search algorithms that combine interestingly different ideas, to really flourish.

而且，嘿，这真的很不可思议，但我们已经有一些这样的形式在运行了。

And, hey, like, it's been incredible, but, like, we have, like, forms of that going already.

现在，现实中的瓶颈在于，你仍然需要大量的计算资源。

Now where the the the sort of physical reality hits is that you still need a lot of compute.

如果你想让RSI提出真正优秀的自我改进方案，你需要给它足够的计算能力，让它能够训练出更复杂的自身版本。

If you ask that RSI to come up with really great forms of itself, you need to give it a lot of compute to come up and be able to train very sophisticated versions of itself.

但明年这个就会起飞了。

But this is going to take off next year.

就是说，它……

Like, it's...

所以我们还没到那一步？

So we're not there yet?

我们还没完全到，但已经很接近了。

We're not quite there yet, but we're very close.

而且再说，在弱形式下，已经有一个了。

Like, and again, in weak forms, there's already one.

然后还有别的方式，人们会从不同角度来拆解它。

And then there are other ways that people slice and dice it.

我朋友杰森·韦斯顿，他还在Meta，他写过一篇相关的论文，你可以从不同的轴来思考自我改进的可学习维度，比如参数、训练数据、目标函数、神经架构，还有整体代码和框架等等。

My friend Jason Weston, who's still at Meta, he kind of wrote a paper around this where you can think about different axes, learnable axes of self improvement, the parameters, the training data, the objective function, the neural architecture, and the overall code and the harness and everything else.

而且还没有人真正破解这个难题，就是同时搞定这五个以上，真正想出该优化哪些维度和轴的想法。

And no one has really cracked the nut of doing all of these five plus, like, truly coming up with the ideas on which of these dimensions and axes to optimize for.

而且你可以知道这还没发生，因为所有大公司还在雇佣成千上万的工程师手动去做，

And you can know that that hasn't hasn't happened yet because all the big companies are still hiring thousands of engineers to do it manually to

在很大程度上。

a large degree.

而且越来越多地借助AI来实现。

And And more and more implementation help from an AI.

你对ASI的定义，因为我想让Salim听到这个。

Your definition of ASI, because I want Salim to hear this.

我认为人工超级智能至少要在多个不同能力上达到顶峰，但按照最宏大的定义，它不仅要超越像图灵测试那样的任意人类，还要超越全人类，解决任意困难的任务。

I think artificial superintelligence has to spike at the very least across several different capabilities, but in in the grandest definition of it, it should supersede not just arbitrary humans like a Turing Test, but all of humanity to solve arbitrarily hard tasks.

而且最终，我认为有10个不同的智能领域，我在You里面定义了这些。

And eventually, I would argue there are 10 different spaces of intelligence that I've defined in in the You.

E.

E.

G.

G.

还有机器方面的书。

Machine book too.

但最终，它不能只是机械地、完全按照指令行事。

But eventually, it cannot just purely robotically do exactly what it's told.

它应该具备某种能力，我并不是说像那种……

It should have some capability, and I'm not saying as sort of a more...

道德上的特权，但我想说的是，如果一件事物在某种程度上无法自主选择要处理的工作，那它就不能算作超级智能。

Moral prerogative, but, like, I would argue that something isn't super intelligent if it cannot choose to some degree what it works on.

而且它还得对自己的思考过程有一定的元认知。

And has to have some metacognition about its thought itself.

不过一般来说，衡量它最直观的方法就是看它在各种智能领域的能力表现，比如视觉感知、沟通、语言、社交互动等等，这些能力都要超越人类。

But generally, the most easy way to measure it is just capabilities across many different spaces of intelligence, like visual perception, communication, language, social interaction, and so on that is beyond that of humanity.

而我们离那个目标还远着呢。

And that we we are still far away.

埃隆的定义是：比所有人类加起来还要聪明。

Elon's definition is as smart as all humans combined.