BuildSpeak每日 builder 文摘
今日归档生词本关于
🎙 播客Unsupervised Learning· 2026 年 7 月 9 日· 8,274 词 · 约 41 分钟

Ep 90: AI Pioneer Jürgen Schmidhuber on the State of AI Today

SPACE 播放 / 暂停·←→ 上一句 / 下一句
Speaker 100:00 - 00:01
Sounds like you think a crash is coming.
Speaker 100:00 - 00:01
听起来你觉得一次崩盘要来了。
Speaker 200:01 - 00:04
That would be one of these stock market crashes.
Speaker 200:01 - 00:04
那会是这类股市崩盘之一。
Speaker 100:04 - 00:05
What about, like, robotics?
Speaker 100:04 - 00:05
那比如说,robotics(机器人技术)呢?
Speaker 200:05 - 00:13
Robot hardware is really inferior compared to human bodies, and there's no human made technology that compares to this hand. It seems a
Speaker 200:05 - 00:13
Robot hardware(机器人硬件)和人体相比确实差得很远,而且人类制造的技术里,没有任何一样能和这只手相比。看起来这
Speaker 100:13 - 00:14
lot more than a hardware problem.
Speaker 100:13 - 00:14
远不只是个硬件问题。
Speaker 200:14 - 00:20
But you can't have AGI without hardware like that. You can't have AGI just behind the screen.
Speaker 200:14 - 00:20
但没有那样的硬件,你就不可能拥有 AGI(通用人工智能)。你不可能只靠屏幕背后的 AGI。
Speaker 100:20 - 00:20
Are we
Speaker 100:20 - 00:20
我们是
Speaker 200:20 - 00:22
kinda close to that? It will come, but it will take a
Speaker 200:20 - 00:22
不是已经有点接近了?它会到来,但还需要一段时间。
Speaker 100:23 - 01:02
Jurgen Schmidhuber has been set as the father of AI by the New York Times, Forbes, and more. He is behind some of the most important advances in the field that really power the AI revolution today, and it was a real privilege on unsupervised learning to get to sit down with him and talk about everything that's top of mind in the ecosystem today. We talked about what's missing in models today, and what he thinks is required to get artificial scientists that can push them forward. We talked about why he thinks today's CapEx boom is massively overdone, and why he's very optimistic on AI technology, but deeply pessimistic on the model companies, and how he doesn't think recursive self improvement will actually be a moat for the companies. We also talked about AI safety, and why he's notably less worried than a lot of others in the field.
Speaker 100:23 - 01:02
Jurgen Schmidhuber 被 New York Times、Forbes 等媒体称为 AI 之父。他推动了这个领域中一些最重要的进展,而这些进展正是当今 AI 革命的核心驱动力;在 unsupervised learning 节目里,能够与他坐下来交流、谈论当下整个生态系统中最受关注的话题,真的是一种难得的荣幸。我们谈到了如今模型缺少什么,以及他认为要实现能够推动模型继续前进的 artificial scientists(人工科学家)需要具备哪些条件。我们还谈到了为什么他认为当下的 CapEx 热潮被严重高估了,为什么他对 AI 技术本身非常乐观、却对 model companies 深感悲观,以及为什么他不认为 recursive self improvement(递归式自我改进)会真正成为这些公司的 moat(护城河)。我们也聊到了 AI safety,以及为什么相比这个领域里的很多其他人,他明显没那么担心。
Speaker 101:02 - 01:14
It's just an awesome opportunity to get to sit down with a legend in the field and ask him all these questions. I think folks will really enjoy hearing his perspective. Without further ado, here's Jorgen. Well, thanks so much for for coming on
Speaker 101:02 - 01:14
能有这样一个机会,坐下来和这个领域的一位传奇人物聊聊,并向他提出这些问题,实在太棒了。我想大家一定会很喜欢听听他的视角。闲话不多说,下面有请 Jorgen。那么,非常感谢你来到
Speaker 201:14 - 01:18
the podcast. Really excited about this. It's my pleasure, Jacob.
Speaker 201:14 - 01:18
这个 podcast。我对此真的非常期待。是我的荣幸,Jacob。
Speaker 101:18 - 01:34
Well, I feel like there there's so many different things I wanna talk about today. You've obviously been a pioneer of of a ton of different parts of this current AI moment. I figured where I'd start at the highest level was, as I understand it, you've had this goal for a long time, which was to build an AI smarter than yourself. How close are we to that right now?
Speaker 101:18 - 01:34
我觉得今天我想聊的事情实在太多了。显然,你是当前这波 AI 浪潮中许多不同方向的先驱人物。我想先从一个最高层面的问题开始:据我了解,你长期以来一直有一个目标,那就是打造一个比你自己更聪明的 AI。我们现在离这个目标还有多远?
Speaker 201:34 - 02:12
From a cosmic perspective, we are as close as we were in the 1970s when I first formulated that wish. So we are very close. But is it going to be a couple of years or a couple of decades? I'm not totally sure about that because true AI is not just the AI behind the screen, which is working very well and which is passing the Turing test now. True AI is also, you know, real robots, real machinery outside of the stream and the real world and the physical world, and that's not working as well.
Speaker 201:34 - 02:12
从宇宙尺度来看,我们和我在 1970 年代第一次形成这个愿望时一样接近。所以,我们已经非常接近了。但到底是还要几年,还是还要几十年,我并不完全确定,因为真正的 AI 不只是屏幕背后的 AI——那部分现在运行得很好,而且如今已经能通过 Turing test。真正的 AI 还包括,你知道,stream 之外、真实世界和物理世界中的真实机器人、真实机械,而那部分进展就没有那么好了。
Speaker 202:12 - 02:27
So the hardware in the real world has lots of limitations that human bodies don't have, and we still have a way to go to be able to compete with human bodies and physical AI.
Speaker 202:12 - 02:27
所以,现实世界里的 hardware 存在很多人体所没有的限制;而要在 human bodies 和 physical AI 方面真正与人体竞争,我们仍然还有一段路要走。
Speaker 102:27 - 02:31
Have there been any current AI results over the last few years that have surprised you?
Speaker 102:27 - 02:31
在过去几年里,当前 AI 的一些成果中,有哪些让你感到意外吗?
Speaker 202:31 - 03:21
Not really, not to the extent that it surprised people who had no contact to neural networks and artificial neural networks in the previous decades, and suddenly there was a chat CBT moment, and suddenly people started being interested in that thing, which they had never seen before. So they they didn't know that there was a long history of large language models and an even longer history of basic insights and algorithms for training these large language models that goes back to the previous millennium. So for a guy who was in center of all of that, it was much easier to predict that than for someone who had totally different interests.
Speaker 202:31 - 03:21
其实没有,至少没有像它让那些过去几十年里从未接触过 neural networks 和 artificial neural networks 的人那样让我意外。后来突然出现了一个 chat CBT 时刻,突然之间,人们开始对这个他们此前从未见过的东西产生兴趣。所以他们并不知道,大型语言模型有着悠久的历史,而支撑这些大型语言模型训练的基本洞见和算法则有着更久的历史,可以一直追溯到上一个千年。对于一个身处这一切核心的人来说,预测这一点显然要比那些兴趣完全在别处的人容易得多。
Speaker 103:21 - 03:40
Well, you know what? I definitely wanna hit on recursive self improvement in meta learning because I feel like you've know, it's been a huge focus of yours for a while. It obviously seems to be a main focus of of a lot of the major labs these days, and you pioneered a bunch of the research here. How do you kind of articulate the you know, where we are today on the on the path of of getting to RSI and and kind of what still needs to be solved?
Speaker 103:21 - 03:40
嗯,你知道吗?我肯定想聊一聊 meta learning(元学习)里的 recursive self improvement(递归式自我改进,RSI),因为我感觉这显然已经是你长期以来非常关注的重点。如今这也明显成了很多主要实验室的核心方向之一,而且你在这方面开创了不少研究。你会怎么概括我们今天在通往 RSI 的这条路径上处于什么位置,以及还有哪些问题需要解决?
Speaker 203:40 - 04:48
So in 1987, this was about using meta evolution evolutionary programming for evolving better programs that learn to do better kinds of evolution, meta evolution, I call it. And it was very Darwinistic in many ways. And so over time, you had better and better learning algorithms, learning to combine code from previous programs in better and better ways to solve problems better and better. And then in, And then 1994, we had reinforcement learning techniques, self referential machines that were able to that basically were using a universal programming language to generate arbitrary self modifications of the code that was running the machine, interacting with some environment. Then in 2003, that was mathematically optimal way of generating self improvements by a machine that has some software, and it's interacting with an environment.
Speaker 203:40 - 04:48
所以在 1987 年,这项工作大致是在用 meta evolution(元进化)的 evolutionary programming(进化式编程)来演化出更好的程序,而这些程序又会学习如何进行更好的进化,我把这叫做 meta evolution。它在很多方面都非常 Darwinistic(达尔文式)。因此,随着时间推移,就会出现越来越好的学习算法,学会以越来越好的方式组合先前程序中的代码,从而越来越好地解决问题。然后到了 1994 年,我们有了 reinforcement learning(强化学习)技术,以及 self referential machines(自指机器);它们基本上使用一种 universal programming language(通用编程语言)来生成对正在运行、并与某个环境交互的机器代码的任意自我修改。再到 2003 年,我们提出了一种数学上最优的方式,让一台拥有某种软件并与环境交互的机器生成自我改进。
Speaker 204:49 - 05:47
And this environment sometimes punishes you or provides a reward, and you want to maximize the sum of all the rewards in your life until the end of your life, and you want to minimize the sum of the pain signals. And then there's an initial software in that machine, and this machine then can, in principle, write programs that modify the initial software. But before it modifies itself like that, it first has to prove, which means there's a proof search in the software. It has to prove that the modification that will be caused by the execution of this particular program is useful in the sense that it will lead to more expected reward than the alternative, which would be not to execute this program. Then it has to generate a formal proof, which was called the Godel machine.
Speaker 204:49 - 05:47
这个环境有时会惩罚你,有时会提供奖励,而你的目标是最大化从现在到生命终点为止所获得的全部奖励之和,同时最小化全部痛苦信号之和。那台机器里先有一套初始软件,原则上,这台机器随后可以编写程序来修改这套初始软件。但在它那样修改自己之前,它首先必须进行证明,也就是说,软件里要执行 proof search(证明搜索)。它必须证明:执行这个特定程序所导致的修改是有用的,也就是与不执行这个程序这个备选方案相比,它会带来更高的期望奖励。然后它必须生成一个 formal proof(形式化证明),这就是所谓的 Godel machine。
Speaker 205:47 - 06:07
It is less practical than certain other things that we did, like neural networks that change their own weight matrix by running and learning algorithm on the network itself. That is what we started in 1992, and that's currently more or less the most popular kind of self modification and self reference.
Speaker 205:47 - 06:07
这比我们做过的某些其他东西更不实用一些,比如 neural networks(神经网络)通过在网络自身上运行学习算法来改变自己的 weight matrix(权重矩阵)。这正是我们在 1992 年开始做的,而到现在,这或多或少已经成了最流行的一类自我修改和自指方式。
Speaker 106:07 - 06:37
Today, as you kind alluded to, like one of the limitations is you have, you know, humans defining the start and end of these, of of of these trials versus, you know, kind of a mathematical way to determine, hey. Is this going to be helpful beforehand? And I guess, obviously, the the trade off of that being that it's very compute intensive, right, to do to do some of these proofs. I guess as you think forward to what the the path to RSI might be, do you think it will involve this kind of, ability to to kind of do these proofs beforehand, or is it like the modification of weights? Or what what yeah.
Speaker 106:07 - 06:37
今天,正如你刚才提到的,其中一个限制是:这些 trial(试验)的开始和结束通常还是由 humans(人类)来定义,而不是用某种数学方式事先判断,嘿,这个东西提前看是否真的会有帮助。我想,很明显这里的权衡在于,要做这类证明在 compute(算力)上会非常昂贵,对吧。那当你展望通往 RSI 的路径时,你觉得它会涉及这种事先做证明的能力吗?还是会更偏向于修改 weights(权重)?或者会是什么样?
Speaker 106:37 - 06:42
I guess what percent would likelihood would you think that the answer lies in in in one of those?
Speaker 106:37 - 06:42
我想问的是,你觉得答案落在这两类方案之一中的可能性大概各占多少百分比?
Speaker 206:42 - 07:28
Yeah. So I would say most of the current self improving systems are scaled back versions, toned down versions of the griddle machine of 2003, the mathematically optimal thing. And they are more like what we had earlier, neural networks, where you have the weights and the program of a neural network is basically the weight matrix of that program. And then you have certain types of neural networks that are general purpose computers. Recurrent networks, for example, they are general purpose computers because on a recurrent neural network, you can implement the processing unit of your Apple laptop or something like that.
Speaker 206:42 - 07:28
对。所以我会说,当前大多数自我改进系统,其实都是 2003 年那个在数学上最优的 Godel machine 的缩减版、弱化版。它们更像我们更早做的那些 neural networks:你有权重,而一个 neural network 的程序,本质上就是那个程序的 weight matrix。然后有某些类型的 neural networks 是 general purpose computers(通用计算机)。比如 recurrent networks(循环网络),它们就是通用计算机,因为在一个 recurrent neural network 上,你可以实现像你的 Apple laptop 那样设备里的处理单元。
Speaker 207:28 - 08:14
So if you have certain instructions that allow you to modify the weights itself, then you can basically run arbitrary learning algorithms on the network, which has to see the errors or negative reward signals that are coming in. So that has to be part of the input, which is essential. And that's all what we did in the early nineties and back then. Compute was so expensive, 10,000,000 times more expensive than today, that we could do only little tiny toy experiments. But today, you can really show nicely that methods like that can learn to generalize and learn new tasks much faster than if you don't have this meta learning capacity.
Speaker 207:28 - 08:14
所以,如果你有某些指令允许你修改权重本身,那么你基本上就可以在这个网络上运行任意 learning algorithms(学习算法),而这个网络必须能够看到输入进来的 errors(错误)或 negative reward signals(负奖励信号)。所以这些信号必须成为输入的一部分,这一点至关重要。这些就是我们在九十年代初做的全部工作。而在当时,compute 的成本比今天高出 10,000,000 倍,所以我们只能做一些非常小的 toy experiments(玩具实验)。但到了今天,你已经可以非常清楚地展示,这类方法能够学会泛化,并且比没有这种 meta learning capacity(元学习能力)的系统更快地学会新任务。
Speaker 208:16 - 08:50
And this is currently the most popular way of doing recursive self improvement. However, one has to admit that it is limited because there, the learning algorithm is invented through gradient descent. So everything is differentiable and then the whole network lies through gradient descent to generate rate changes that are better than what's caused by a grain descent. Yeah. Yeah.
Speaker 208:16 - 08:50
而这目前是进行 recursive self improvement 最流行的方法。不过,人们必须承认它是有限的,因为这里的 learning algorithm(学习算法)是通过 gradient descent 发明出来的。所以一切都是可微的,然后整个网络通过 gradient descent 来产生比 gradient descent 本身导致的变化更好的速率变化。对。对。
Speaker 208:50 - 09:01
So that has the limitations of grain descent, so it's not like the optimal good machine, but it works really nicely in practice.
Speaker 208:50 - 09:01
所以它带有 gradient descent 的局限性,因此它并不是那种最优的“好机器”,但它在实践中确实运作得非常好。
Speaker 109:01 - 09:21
I think a big question people have around recursive self improvement is that in retrospect, is it going to feel like, hey, it was this gradual and boring improvement, or is there some huge discontinuity around the horizon where suddenly models take off in capabilities? What's your gut instinct around that when we get there and are looking back?
Speaker 109:01 - 09:21
我觉得,人们围绕 recursive self improvement 的一个大问题是:事后回看时,这会不会让人感觉,哦,这其实只是一个渐进而乏味的改进过程;还是说,在地平线附近会出现某种巨大的不连续性,让模型的能力突然起飞?你对此的直觉是什么——等我们真的走到那一步、再回头看时,会是什么样?
Speaker 209:21 - 09:46
From a cosmic perspective, it will look just like a stick. There was no self improvement and no real AI, and suddenly there was AI. But then from a cosmic perspective, this is also true for all of civilization. Civilization started roughly thirteen thousand years ago, and before that, there was no artificial intelligence. And then only thirteen thousand years later, there was artificial intelligence.
Speaker 209:21 - 09:46
从宇宙尺度的视角来看,它看起来就像一根棍子一样陡直:之前没有 self improvement,也没有真正的 AI,然后突然就有了 AI。不过从宇宙视角来看,这对整个文明也是成立的。文明大约始于一万三千年前,而在那之前,并没有 artificial intelligence。然后,仅仅又过了一万三千年,就有了 artificial intelligence。
Speaker 209:48 - 10:32
And basically, the first guy who had agriculture thirteen thousand years ago and domestication of the animals, there was almost the same guy who had the first And and and the 13,000 of civilization, they are just one millionth of world history, which is about 13,800,000,000 years. So it's just a a flash in world history. In hindsight, it will look like that. Civilization was almost occurred almost at the same time where AI occurred because over these past thirteen thousand years, more and more stuff was automated, and more and more of agriculture was automated, and more and more of human labor was automated. And at some point, thinking started to become automated.
Speaker 209:48 - 10:32
基本上,一万三千年前第一个拥有 agriculture(农业)和动物 domestication(驯化)的人,几乎就和第一个拥有——而且这 13,000 年的文明史,只占 world history(世界历史)的大约百万分之一;世界历史大约是 13,800,000,000 年。所以这在世界历史里只是一瞬间。事后看来会像是这样:文明的发生几乎和 AI 的出现发生在几乎同一时间,因为在过去这一万三千年里,越来越多的事情被自动化了,越来越多的农业被自动化了,越来越多的人类劳动被自动化了。而在某个时刻,思考也开始被自动化了。
Speaker 210:32 - 11:02
A couple of hundreds of years ago, so the first calculators, and then the calculators become became less expensive, and then they became faster, and now you can calculate much more than a hundred years ago for the same price. And then suddenly, it was there. Now from from this global perspective, it's really like there was nothing, suddenly, there was a lot. From the perspective of a guy who is living through that age, it looks like a lot.
Speaker 210:32 - 11:02
几百年前,最早的 calculators(计算器)出现了;然后 calculators 变得越来越便宜,也变得越来越快,而现在,用同样的价格,你能完成的计算远远超过一百年前。然后突然之间,它就在那里了。现在,从这种全球视角来看,真的就像之前什么都没有,突然之间就有了很多东西。而从一个生活在那个时代的人的视角来看,这一切则显得经历了很多过程。
Speaker 111:02 - 11:15
I think one of the the hopes, I guess, of of RSI is that, you know, it may end intensive over time, right, to actually continue making some of these developments. So I guess we'll see, but that I think would be one of the hopes there, right?
Speaker 111:02 - 11:15
我想,RSI 的一个希望,大概是,随着时间推移,它也许会变得 end intensive,从而真的能够继续推进其中一些发展。所以我想我们只能拭目以待,但我觉得那确实是其中一个希望,对吧?
Speaker 211:16 - 11:57
Yeah, absolutely. And whenever you are talking about intelligence, you basically are talking about laziness. An intelligent being wants to be lazy and wants to achieve whatever it does with the least possible effort, with the least possible energy consumption. So all of our self improving systems, they have this extra reward for being efficient. In other words, They have an extra cost for every time they wake up a neuron and use energy to wake up that neuron or use energy to do other things.
Speaker 211:16 - 11:57
对,绝对如此。而且每当你谈论 intelligence(智能)时,你基本上谈的就是 laziness(懒惰)。一个有智能的存在会想要偷懒,会想用尽可能少的努力、尽可能低的能量消耗,去达成它要做的事情。所以我们所有的 self improving systems(自我改进系统)都会因为高效而获得额外奖励。换句话说,它们每次唤醒一个 neuron(神经元)、使用能量去唤醒那个 neuron,或者使用能量去做其他事情时,都会承担额外成本。
Speaker 211:57 - 12:32
So all the costs, the computational costs and the other energy costs have to be taken into account in the function and the objective function that our self improving systems are optimizing. Yeah. To the extent that they are good at that, they will do the same thing with less and less resources, computational resources, other resources. So a natural consequence of intelligent behavior is that whatever is being done is done more and more efficiently. As you
Speaker 211:57 - 12:32
所有成本——计算成本以及其他能量成本——都必须被纳入我们的自我改进系统所优化的函数和目标函数中。对。只要它们在这方面做得足够好,它们就会用越来越少的资源来完成同样的事情,包括计算资源和其他资源。所以,智能行为的一个自然结果就是:无论在做什么,都会做得越来越高效。正如你
Speaker 112:32 - 12:55
reflect on maybe the broader work that's going on at the AI labs, I mean, obviously, there's tons of money and compute going into a bunch of the research questions that the labs are are going after. And I'm wondering if you were if you were kind of running one of labs or I'm sure you talked to folks there or a few about advising them, what do you think they're maybe that you advise them to do differently today from what they're doing?
Speaker 112:32 - 12:55
在思考 AI labs 里更广泛正在进行的工作时,我是说,很显然,大量资金和 compute(算力)正投入到这些 labs 所追求的一系列研究问题中。我在想,如果由你来运营其中一家 lab,或者我想你肯定也和那边的一些人聊过、甚至给过他们建议,你觉得他们今天有哪些事情应该做得和现在不一样?
Speaker 212:55 - 13:06
So today, you use large trained systems, large language models that already have read a lot of papers about coding.
Speaker 212:55 - 13:06
所以今天,你会使用那些已经训练好的大型系统,也就是 large language models,它们已经读过大量关于 coding 的论文。
Speaker 113:06 - 13:07
Of course.
Speaker 113:06 - 13:07
当然。
Speaker 213:07 - 13:41
And then now a modern way of using a pre trained model like that is you let it code something, and you only improve the way it codes or recodes its own code. So that is now a pretty obvious way of doing it, and quite a few labs are interested in exactly that. And of course, you have to have safeguards such that it doesn't really write its own code in a way that is totally stupid and makes things worse than better rather than better. But there are ways of of dealing with that.
Speaker 213:07 - 13:41
现在,使用这类 pre trained model(预训练模型)的一种现代方式是:让它先去写一些代码,然后你只改进它写代码的方式,或者让它重写自己代码的方式。所以这已经是一种相当明显的做法,而且有不少 labs 对此正好非常感兴趣。当然,你必须设置 safeguards(防护措施),以确保它不会真的以一种完全愚蠢的方式去改写自己的代码,结果把事情变得更糟而不是更好。不过,这类问题是有办法处理的。
Speaker 113:42 - 13:47
No. That makes sense. Mean, I guess, is there anything you'd kind of doing differently there?
Speaker 113:42 - 13:47
不,这说得通。我的意思是,我想问的是,在这方面你会不会做一些不同的事?
Speaker 213:47 - 14:09
It's a reasonable approach. It's not the most general approach in the sense that if you rely on anything that is pretrained on human generated data, then, you know, you are ignoring many of the other possibilities. So look at our current large language models. They are super biased towards humans. Why?
Speaker 213:47 - 14:09
这是一个合理的方法。但它不是最一般化的方法,因为如果你依赖任何基于 human generated data(人类生成数据)预训练出来的东西,那么,你其实就忽略了很多其他可能性。所以看看我们当前的 large language models。它们对 humans 有非常强的偏置。为什么?
Speaker 114:09 - 14:10
Of course.
Speaker 114:09 - 14:10
当然。
Speaker 214:10 - 14:53
Because they are trained on all the data on the World Web. All the data on the World Wide Web is there for the only reason that at least one guy, one person at some point thought this is interesting from a human perspective, And then all this material that at least some guy thought is interesting is used to train the systems, and they are, for that reason, super biased towards human language, towards videos that humans find interesting, towards behavior that humans find interesting, and so on. So they are very aligned in a certain way with humans. Maybe they are more aligned with certain humans than with others. Nevertheless, there's a tremendous human bias.
Speaker 214:10 - 14:53
因为它们是在 World Wide Web 上的全部数据上训练的。而 World Wide Web 上的所有数据之所以会存在,唯一的原因是:至少曾经有某个人在某个时刻认为这东西从 human perspective(人类视角)看是有趣的。然后,所有这些至少被某个人认为“有趣”的材料,又被拿来训练这些系统。因此,它们就会强烈偏向 human language、偏向 humans 觉得有趣的视频、偏向 humans 觉得有趣的行为,等等。所以从某种意义上说,它们确实与 humans 高度 aligned(对齐)了。也许它们与某些 humans 的对齐程度高于另一些 humans。尽管如此,其中仍然存在巨大的 human bias(人类偏置)。
Speaker 214:54 - 15:40
Now, of course, if you have an artificial scientist who is living in some unknown environment and tries to build a model of the world by just predicting the consequences of its actions and then use the model of the world for planning, such a such an agent will have to create through its own actions the data that trains the wild model. So suddenly, you you have something like an artificial scientist who, through its own actions, generates the data on which one model is being trained, and this is much more like what humans do, what babies do. Babies don't learn by downloading the web or something. No. They learn by predicting the consequences of the actions.
Speaker 214:54 - 15:40
当然,如果你有一个 artificial scientist(人工科学家),它生活在某个未知环境中,并试图仅仅通过预测自己行为的后果来建立一个 world model(世界模型),然后再利用这个世界模型进行 planning(规划),那么,这样一个 agent(智能体)就必须通过自己的行动来创造用于训练该 world model 的数据。于是突然之间,你得到的就像是一个 artificial scientist:它通过自己的行动生成数据,而其中一个模型正是在这些数据上被训练出来的。这就更像 humans 的做法,像 babies 的做法。婴儿不是靠“下载整个 web”来学习的。不是。它们是通过预测自己行为的后果来学习的。
Speaker 215:40 - 16:46
So if they move their fingers like that, then the video changes, which comes in through the cameras, and they learn to predict these changes. And that's how they learn about the physics of the world and about how the world works and about how they their fingers work and everything. And they they are being trained a lot of data that is not on the World Wide Web, you know, that is collected through their own experiments, and and it's exactly the kind of data that the baby needs to better understand what it can do and that it needs to to maximize its own reward. So now you you see that that all of the data that is collected on the World Wide Web, that seems to be a lot, but it's just a tiny, tiny, tiny fraction of all the possible data that you could collect out there through your own experiments. So the future of AI is going to lie in such systems that, through their own actions, through artificial curiosity, as I called it in 1990, collect all the data that is used to train the world models.
Speaker 215:40 - 16:46
所以如果他们那样移动手指,视频就会变化;这些视频通过摄像头输入进来,而他们会学着去预测这些变化。他们就是这样学习世界的物理规律、学习世界如何运作、学习自己的手指如何运作,等等。而且,他们接受训练所用的大量数据并不在 World Wide Web 上,你知道,这些数据是通过他们自己的实验收集来的;而这恰恰是 baby 为了更好地理解自己能做什么、并为了最大化自身 reward(奖励)所需要的数据。所以现在你会看到,World Wide Web 上收集到的全部数据看起来很多,但和你通过自己的实验在外部世界中可能收集到的全部数据相比,那只是一小、一小、一小部分。因此,AI 的未来将会属于这样的系统:它们通过自己的行动,通过我在 1990 年称之为 artificial curiosity(人工好奇心)的机制,收集所有用于训练 world models(世界模型)的数据。
Speaker 216:46 - 17:13
And these world models, they will not depend on human language, and they will be very focused in many ways on this particular robot that is collecting the data. And, yes, of course, it is going to communicate with other robots living in different environments, and it's going to incorporate that knowledge so that it can better generalize and so on. But, you know, suddenly you have a situation where where the systems are going to be much less human biased.
Speaker 216:46 - 17:13
而这些 world models 将不会依赖 human language(人类语言),并且在很多方面都会非常聚焦于这个正在收集数据的特定 robot。当然,是的,它会与生活在不同环境中的其他 robots 交流,并吸收那些知识,从而更好地实现 generalize(泛化)等等。但是,你知道,突然之间你就会进入这样一种局面:这些系统将会少得多地带有人类偏见。
Speaker 117:13 - 17:38
No. It's interesting. I a lot of what you you just described, obviously, there there's companies that are trying to build, you know, automated labs for, you know, material science discovery or for biology or, you know, even in in robotics, there's all these folks doing, you know, teleoperated data. But obviously today it's kind of humans deciding what the data that should be gathered is or designing those kinds of experiments and trying to feed it back. And I guess the dream is to have a fully AI driven loop of that discovery, right?
Speaker 117:13 - 17:38
不。我觉得这很有意思。你刚才描述的很多东西,显然已经有公司在尝试构建,比如用于 material science 发现、biology,甚至 robotics 的 automated labs;还有很多人在做 teleoperated data(遥操作数据)之类的事。但显然,今天依然主要是 humans 在决定应该收集什么数据,或者在设计这类实验,并试图把结果反馈回系统里。我想,那个梦想就是拥有一个完全由 AI 驱动的 discovery 闭环,对吧?
Speaker 217:38 - 18:11
That's true. Yeah. Twenty years ago, I wrote this paper about the formal theory of fun and creativity, which is about exactly that. So what should an artificial scientist or any scientist or any artist or any comedian do when its main needs, like eating three times a day, are satisfied and it has extra time to do stuff? What do you do?
Speaker 217:38 - 18:11
这没错。是的。二十年前,我写过一篇关于 fun 和 creativity 形式理论的论文,讲的正是这件事。那么,一个 artificial scientist,或者任何 scientist、任何 artist、任何 comedian,在它的主要需求——比如一天吃三顿饭——都已经满足,并且还有额外时间可以做事的时候,它该做什么?你会做什么?
Speaker 218:11 - 18:48
Well, you listen to music, or maybe you compose your own music, or you generate art, or you are a scientist who not only is trying to solve problems given to you from by other people. No, you also try to invent your own new problems. Ask your own questions, not just answer questions given to you by somebody else, but invent your own new good questions. That's what scientists do. And all of science is about two things, not only taking an existing question and investing a lot of time into solving it and finding an answer.
Speaker 218:11 - 18:48
你会听音乐,或者也许你会自己作曲,或者你会生成艺术作品,或者你是一个 scientist,不仅仅去解决别人交给你的问题。不,你还会试着发明你自己的新问题。提出你自己的问题,而不只是回答别人给你的问题,而是发明你自己的、好的新问题。scientists 做的就是这个。而整个 science 关乎两件事,不只是拿到一个现有问题,然后投入大量时间去解决它并找到答案。
Speaker 218:48 - 19:32
No. Inventing the good questions. And the the basic principle is very simple. So an artificial scientist or any scientist, in my point of view, a human scientist as well, is driven by one simple thing, which is try to find through your actions, your own self invented experiments, data that has the property that in the data, there is some regularity, some pattern that you didn't know but can quickly learn. That you didn't know, but you can quickly learn it because it's at the limit of what you already almost understand, near the horizon of what you don't know and what you know.
Speaker 218:48 - 19:32
不。还包括发明好的问题。而基本原理非常简单。所以,在我看来,一个 artificial scientist,或者任何 scientist——human scientist 也一样——都是被一件简单的事驱动的:通过你自己的行动、你自己发明的 experiments,去找到这样的 data:其中具有某种 regularity(规律性)、某种 pattern(模式),是你原本不知道、但可以很快学会的。也就是说,那是你原本不知道的,但你可以很快学会,因为它正处在你几乎已经理解的边界上,位于你未知与已知之间的地平线附近。
Speaker 219:34 - 20:15
And then you generate this data through your own actions, through your experiments, sequences of actions, which are experiments, and the data comes in. And if there is something interesting in the data, what does that mean? It means that there's a pattern in there which you didn't know. All patterns mean there is some way of compressing those things in space and time. You're not talking about time right here to make things not too complicated, but to compress to compress the patterns in a way that you didn't know before, which means that before you understood the regularity and the pattern which is coming in, you needed so many internal bits and hidden units and so on to encode it.
Speaker 219:34 - 20:15
然后你通过自己的行动、通过你的 experiments——也就是一系列行动——来生成这些 data,接着 data 就流入了。如果 data 里面有某种有趣的东西,这意味着什么?这意味着里面存在某种你此前不知道的 pattern。所有 pattern 都意味着,存在某种方式可以在空间和时间中压缩这些东西。这里我们先不谈时间,以免把事情搞得太复杂;但重点是,以一种你之前并不知道的方式去 compress(压缩)这些 pattern。这也就意味着,在你理解正在到来的这种 regularity 和 pattern 之前,你需要那么多 internal bits、hidden units 等等来对它进行编码。
Speaker 220:15 - 20:55
And afterwards, after you learn to see the pattern, after you learn to compress it, you need only so many. And the difference between before and after, that's the funds that you have, you know. That's just a real number which says how much internal joy does the scientists now have from recognizing, from realizing, Oh, there is regularity that I didn't know in the day I was just coming in. It tells me something about gravity. And then this becomes the reward of the controller who's generating the actions that lead to the data, which means now the controller is motivated to generate more and more experiments that lead to insights about the world that it didn't have before.
Speaker 220:15 - 20:55
而之后,在你学会看见这种 pattern、学会压缩它之后,你只需要那么多了。before 和 after 之间的差值,就是你拥有的 funds,你知道的。那只是一个 real number(实数),表示这个 scientist 现在因为识别出、意识到“哦,原来刚刚到来的 data 里存在我此前不知道的 regularity,它告诉了我一些关于 gravity 的东西”而获得了多少 internal joy(内在愉悦)。然后,这就变成了那个生成这些行动、并由此导向这些 data 的 controller(控制器)的 reward;这意味着,现在这个 controller 会被激励去生成越来越多的 experiments,去导向它此前并不具备的、关于世界的新 insight(洞见)。
Speaker 220:55 - 21:34
And everything that it understands becomes boring, and then wants to create more complicated experiments to, you know, after it has grabbed the low hanging fruits, more complicated experiments. And maybe the same, the baby which first learned about gravity when it was a little being, maybe a year old or something. Maybe twenty years later, the same baby is working at the particle collider at the sun and is part of the team that discovers the the Higgs boson or something. And and the only difference is that the experiments are more expensive.
Speaker 220:55 - 21:34
它理解的一切最终都会变得无聊,然后它就会想要设计更复杂的实验。也就是说,在它摘完那些 low-hanging fruits(低垂果实、容易拿到的成果)之后,就会去做更复杂的实验。也许人也是一样:一个婴儿在很小的时候,也许一岁左右,第一次学到重力;也许二十年后,同一个婴儿已经在太阳上的 particle collider 工作,成为发现 Higgs boson 团队的一员之类的。而唯一的区别,只是实验变得更昂贵了。
Speaker 121:34 - 22:04
And I guess is that really the the the blocker as you think about, you know, what what stands between us and getting to this AI scientist? Obviously, like, the ability to stand up experiments, right, and and do them, you know, both at an affordable price point, but also kind of feedback into models is is something that a lot of people are working on, but certainly remains a blocker. And it also seems like there's all sorts of algorithmic problems to be solved to actually create this setup that you said. How do you think about what needs to be solved to make this AI scientist a reality? Then how soon do you think we'll get there?
Speaker 121:34 - 22:04
我想,真正的 blocker(阻碍)是不是就在这里?也就是,当你思考什么阻挡着我们实现这种 AI scientist 时。很明显,搭建实验并实际执行实验的能力——既要价格可负担,又要能把结果反馈回 model(模型)——这是很多人都在努力做的事,但当然仍然是个 blocker。与此同时,看起来还存在各种 algorithmic(算法层面)的问题,需要先解决,才能真正搭起你刚才说的那套系统。你怎么看:要让这个 AI scientist 变成现实,还需要解决什么?以及你觉得我们多久能走到那一步?
Speaker 222:04 - 22:15
Well, we have simple AI scientists. We have had them for a long time. It's just that maybe they haven't seen that chat GPT moment yet. When was that, 'twenty two, 'twenty three?
Speaker 222:04 - 22:15
嗯,我们已经有简单的 AI scientists 了,而且其实存在很久了。只是它们可能还没有迎来那个 chat GPT moment。那是什么时候来着,22 年,23 年?
Speaker 122:15 - 22:16
Yeah, end of 'twenty two.
Speaker 122:15 - 22:16
对,22 年底。
Speaker 222:16 - 22:59
It was based on stuff that was really old. We don't have the same kind of chat GPT moment yet for these artificial scientists, but the artificial scientists, to a certain extent, they already exist, and and they are being used in lots of special applications. For example, in chemistry, you have lots of pairs of inputs and outputs, and you train your neural network to predict to predict these these new substances given the previous inputs. And then over time, if you show it millions of experiments, it allows to become an artificial chemist, an intuitive chemist. So it's not a chemist who understands all these reactions from first principles, from valence electrons or whatever.
Speaker 222:16 - 22:59
它是建立在非常老的东西之上的。对于这些 artificial scientists,我们还没有迎来同样的 chat GPT moment;但在某种程度上,它们其实已经存在了,而且已经被用于很多专门的应用里。比如在 chemistry 里,你会有大量成对的输入和输出,然后你训练 neural network(神经网络)去预测——根据之前的输入,预测这些新的物质。随着时间推移,如果你给它看几百万个实验,它就能成为一个 artificial chemist,一个 intuitive chemist(直觉型化学家)。所以它不是那种从 first principles(第一性原理)、从 valence electrons(价电子)之类出发去理解所有反应的 chemist。
Speaker 222:59 - 23:35
No, it just becomes an intuitive chemist that better understand understands what can be done in chemistry. And then you have certain objectives. For example, you want to create a material that is that is twice as efficient against a certain maybe maybe an insecticide or something. That is twice as efficient as the most the the best thing known so far. And then you can say, okay.
Speaker 222:59 - 23:35
不,它只是变成一个 intuitive chemist,更懂得在 chemistry 里什么是可行的。然后你会设定一些目标。比如说,你想创造一种材料,针对某种东西——也许是某种 insecticide(杀虫剂)之类——它的效率要比目前已知最好的东西高一倍。然后你就可以说,好。
Speaker 223:35 - 24:07
Let's have a desired output like that, which encodes that it should be twice as efficient as the best thing I know. And then you can work the whole chemist backwards and can look at the input side and can say, much should I change my experiment, which is visible at the input side, to get this thing that I would like to see to fulfill my wish? And then the chemist will basically give you a suggestion.
Speaker 223:35 - 24:07
我们来设定这样一个期望输出,把“它应该比我所知道的最佳结果高效两倍”编码进去。然后你就可以反向推整个 chemist 的过程,去看输入端,并且问:为了得到我想看到的、满足我愿望的那个结果,我需要在实验中那些位于输入端、可观测的部分上做多大改动?然后这个 chemist 基本上就会给你一个建议。
Speaker 124:07 - 24:10
Do you feel like within a decade, we'll we'll figure out AI chemistry?
Speaker 124:07 - 24:10
你觉得在未来十年内,我们能把 AI chemistry 这件事搞明白吗?
Speaker 224:10 - 25:03
At KAUST, we have an interesting project where the goal is to to use certain patented structures, metal, MOFs, they're called. The goal is to extract carbon dioxide from thin air, which is important for improving the climate. At the moment, all of that is very expensive, and the goal is to make it so cheap that you really can make a dent and maybe improve the global warming situation. So that is one of the potential applications. There are lots of applications and all kinds of chemistry fields What
Speaker 224:10 - 25:03
在 KAUST,我们有一个很有意思的项目,目标是使用某些已获专利的结构,也就是所谓的 MOFs。目标是从稀薄空气中提取 carbon dioxide,这对于改善气候非常重要。目前这一切都非常昂贵,而目标是把成本降到足够低,这样你才能真正产生实质影响,也许能改善 global warming 的局面。所以这是潜在应用之一。应用其实很多,几乎遍布各种 chemistry 领域。What
Speaker 125:04 - 25:15
about robots? How do you characterize where we are, what's been happening there? Everyone has a stream of an at home robot. Are we close to that?
Speaker 125:04 - 25:15
那 robots 呢?你会怎么描述我们目前所处的位置,那里都发生了些什么?每个人都怀有一个居家 robot 的愿景。我们离那个目标近了吗?
Speaker 225:15 - 25:27
I remember again in the seventies when I told my mom about the future of AI and how AI is going to colonize the entire universe. And she said, just come build me a robot that cleans my kitchen.
Speaker 225:15 - 25:27
我又想起在 seventies 的时候,我跟我妈妈谈起 AI 的未来,说 AI 将会殖民整个宇宙。她却说,你先给我造一个能打扫我厨房的 robot 吧。
Speaker 125:27 - 25:28
The age old dream.
Speaker 125:27 - 25:28
这是个由来已久的梦想。
Speaker 225:28 - 25:55
Yeah. Exactly. Back then, that didn't work, still still doesn't work because robot hardware is really inferior compared to human bodies. And there's no handmade technology, no human made technology that compares to this hand. This hand is full of sensors, millions of little sensors, and little cables that connect it to the control center.
Speaker 225:28 - 25:55
对,完全没错。那时候做不到,现在其实还是做不到,因为 robot hardware 相比 human bodies 真的差太多了。而且没有任何 handmade technology,也没有任何 human made technology,能和这只手相比。这只手里布满了 sensors,数以百万计的微小 sensors,还有把它们连接到控制中心的细小 cables。
Speaker 225:57 - 26:13
I wouldn't even know where to put all these cables in a in an artificial hand. And the the crazy thing is you you harm it. You cut it, and it starts healing itself. This is super advanced technology. We have nothing like that in man made tech.
Speaker 225:57 - 26:13
我甚至都不知道,在一只 artificial hand 里该把这么多 cables 放到哪里。更夸张的是,你伤到它,你把它割破,它还会开始自我愈合。这是超级先进的 technology。我们在人造技术里根本没有任何类似的东西。
Speaker 226:14 - 26:24
And and that's the reason why, you know, why the robots of the movies are all played by humans because humans are much better robots than the the robots that
Speaker 226:14 - 26:24
而这也就是为什么,你知道,电影里的 robots 全都是由 humans 来扮演,因为 humans 作为 robots,要比那些 robots 更出色。
Speaker 126:24 - 26:31
But it seems a lot more than a hardware problem. Right? I mean, even even with the hardware we have, if we if we had good models, I'm sure we could do we could do way more.
Speaker 126:24 - 26:31
但这看起来远不只是 hardware 的问题,对吧?我的意思是,即便用我们现有的 hardware,如果我们有好的 models,我相信我们也能做到多得多的事情。
Speaker 226:31 - 26:38
Right? But, you know, you can't have AGI without hardware like that. Know? You can't have AGI just behind the screen. Yeah.
Speaker 226:31 - 26:38
对吧?但是,你知道,没有那样的 hardware(硬件)就不可能有 AGI。明白吗?你不可能只靠屏幕后面的东西就实现 AGI。对。
Speaker 226:38 - 27:05
You can have a superhuman chess player behind the screen and something that passes the Turing test. But if it does master the real world, it's not an AGI. Know? It's just a maybe a fancy text editor behind the screen or something that has an idea of how this robot should move or whatever, but it's not the real thing. And if you want to master the real world through a real AGI, a physical AGI, so much more has to be done.
Speaker 226:38 - 27:05
你可以在屏幕后面拥有一个超人类的国际象棋选手,或者一个能通过图灵测试的东西。但如果它不能掌控现实世界,那它就不是 AGI。明白吗?它也许只不过是屏幕后面一个花哨的文本编辑器,或者某种知道这个机器人应该怎么移动之类的系统,但它不是真正的那个东西。如果你想通过真正的 AGI、physical AGI(物理形态的 AGI)来掌控现实世界,那还需要做多得多的工作。
Speaker 227:05 - 27:36
So our robots have to become much better than the limited stuff that we have today. How much longer will that take? We can easily predict how how compute per dollar is going to evolve. It's probably going to stay like what we have seen for decades now, factor of 10 every five years, roughly like that, which means also that the guys who are investing a thousand billion dollars into GPUs for data centers today, within the next five years, they are going to lose $900,000,000,000. You know?
Speaker 227:05 - 27:36
所以,我们的机器人必须比我们今天拥有的那些受限的东西强得多。这还要多久?我们很容易预测每美元 compute(算力)会如何演进。它很可能会延续我们过去几十年看到的趋势,大致是每五年提升 10 倍,差不多就是这样。这也意味着,那些今天向 data center(数据中心)里的 GPUs 投入一万亿美元的人,在接下来的五年内将损失 900,000,000,000 美元。你知道吧?
Speaker 227:37 - 28:15
There's no business model that can recuperate that loss, which already is an indication of the coming crash maybe. Yeah. But in in robots Yeah. In in robot technology, how long is it going to take to to get something that is comparable to this hand, which can do both strong grips and very delicate, you know, finger movements that just manipulate tiny little things in a way that is infeasible for what kind of robots can do. That is, at least for me, much harder to predict.
Speaker 227:37 - 28:15
没有任何 business model(商业模式)能弥补这种损失,这本身可能已经是在预示即将到来的崩盘。对。但是在机器人领域——对,在 robot technology(机器人技术)里——要多久才能做出某种可与这只手相比的东西?它既能进行强力抓握,又能进行非常精细的手指运动,去操控极其微小的东西,而这种能力是现在各种机器人根本做不到的。至少对我来说,这就更难预测了。
Speaker 228:15 - 28:23
It will come, but it will take, maybe it will not take just a few years. It will take another few decades maybe.
Speaker 228:15 - 28:23
它会到来,但这可能不只需要几年时间。也许还需要再过几十年。
Speaker 128:23 - 29:14
Well, I I do wanna switch over to business side because you said obviously something very intriguing there, and I've heard you say this before that, like, you know, this massive investment in AI data center build out, you know, ultimately with the improvement in compute performance, you're investing a ton upfront in something that's going to be outdated and worth far less five years from now. The counterargument or what folks would say to that is that there's just going to be infinite need for compute and that ultimately, like, we're going to be, you know, bottlenecked in our ability to to produce enough compute such that, like, even if you have hardware that's legacy and way less efficient, it's actually gonna still people will still want to use it because there's just gonna be such demand to run inference for for all of these different interesting, at least today, digital and in the future physical use cases. And so even if it's less efficient or or not the best compute, it's like every chip on the planet will will need to be used in some way. What's kind of your reaction to that?
Speaker 128:23 - 29:14
好吧,我确实想转到商业层面,因为你刚才说了一件显然非常耐人寻味的事,而且我之前也听你这么说过:像现在这种对 AI data center(AI 数据中心)建设的大规模投资,最终随着 compute 性能的提升,本质上是在前期投入巨资去买一个五年后就会过时、而且价值低得多的东西。对此的反驳,或者说人们会讲的是,对 compute 的需求将会是无限的;最终,我们会受限于自己生产足够 compute 的能力。也就是说,即便你手上的 hardware 已经是 legacy(旧一代)的、效率低很多,人们还是会想用它,因为仅仅为了给所有这些不同的、至少在今天还是 digital(数字)的、未来则是 physical(物理)的 use cases(用例)运行 inference(推理),需求都会大得惊人。所以即使它效率更低,或者不是最好的 compute,也像是地球上的每一块 chip(芯片)都会以某种方式被用上。你对这种说法大概是什么反应?
Speaker 229:14 - 29:27
Yeah. Maybe there will be more and more demand for compute, but somebody has to pay for it. Right? And if it turns out that those guys who are currently paying, that they are losing a lot of money. You know?
Speaker 229:14 - 29:27
对,也许对 compute 的需求会越来越多,但总得有人为它买单。对吧?如果结果是,那些目前正在付钱的人其实是在亏很多钱呢。你知道吧?
Speaker 229:27 - 30:26
At the moment, a couple of companies are investing hundreds of billions per year. So in total, maybe 1,000,000,000,000 per year or something like that into GPUs for data centers. And these companies, which used to be nimble software companies, had a little team improving some city operating system and then rolling it out for billions of people who had their own phones and their own computers to run the operating system. Suddenly, they are providing clouds and data centers, and suddenly they have to become like utilities, like electricity companies, and they have to invest in nuclear power plants and gas turbines and whatever. And suddenly, the cash flow, the free cash flow of these companies goes down from a 100,000,000,000 down to 10,000,000,000 or maybe minus 10,000,000,000 of like, for some of these companies.
Speaker 229:27 - 30:26
眼下,有几家公司每年都在投入数千亿美元。合起来,也许是每年 1,000,000,000,000 美元左右之类的数额,投向 data center 里的 GPUs。而这些公司过去本来是灵活的软件公司,靠一个小团队改进某个城市 operating system(操作系统),然后把它推送给数十亿人,让他们用自己的手机和电脑来运行这个 operating system。可突然之间,它们开始提供 cloud(云服务)和 data center,突然之间它们不得不像 utilities(公用事业公司)那样运作,像电力公司一样,还得去投资 nuclear power plants(核电站)、gas turbines(燃气轮机)之类的东西。于是,这些公司的 cash flow(现金流)、free cash flow(自由现金流)就会从 100,000,000,000 降到 10,000,000,000,甚至对其中一些公司来说可能会变成负 10,000,000,000。
Speaker 230:27 - 31:11
So at the moment, while everybody is trying to get market share, these these services are really not efficient in the sense that the guys who are providing the services, they are losing a lot of money. It doesn't show up in the price earning ratio because the free cash flow, which really should be considered, is not showing up there, But all these companies are getting less and less efficient. So at some point, they wouldn't have to stop. You know? Now they're taking on debt to finance even more data centers, and this will be possible only to a certain extent, and then these companies will become less and less valuable.
Speaker 230:27 - 31:11
所以眼下,当所有人都在争夺 market share(市场份额)时,这些服务其实并不高效,意思是提供这些服务的人正在亏很多钱。这一点不会体现在 price earning ratio(市盈率)里,因为真正应该被考虑的 free cash flow(自由现金流)并没有反映在那里。但所有这些公司的效率都在越来越低。所以到了某个时候,他们就不得不停下来。你知道吗?现在他们还在通过举债来融资、建设更多 data centers,而这只可能持续到某个限度,之后这些公司的价值就会越来越低。
Speaker 231:11 - 31:45
So they because they misinvested their money. That means that that at some point, former later, if you can't pay for all these services, if the entire economy is not set up to pay for all these services because compute isn't cheap enough yet and because people want to compensate for it by buying even more computer computers. If if you have a situation like that, which which doesn't lead to to a profitable economy, then it's going to disappear due to the Yeah. Laws of supply and demand.
Speaker 231:11 - 31:45
所以这是因为他们把钱投错了地方。这意味着,迟早有一天,如果你没法为所有这些服务买单,如果整个经济体都没有被构建成可以为这些服务付费的样子——因为 compute(算力)还不够便宜,而且人们还想通过购买更多 computer(计算机)来弥补——如果你处在这样一种局面里,而这种局面又无法导向一个有利润的经济,那么它最终就会因为供需规律而消失。对,laws of supply and demand(供求规律)。
Speaker 131:45 - 32:32
I mean, I guess there's kind of, like, two two questions embedded in that. Right? One is on on the inference itself, you know, are there enough use cases out there that are valuable enough for companies to to want to pay, you know, well well above and beyond the the kind of cost of these data center build outs? And certainly with AI coding, it seems like there's, you know, as as that as the as the first market with a ton of fit, it seems like there's been at least some use cases there that that certainly meet that bar. I you know, certainly on the training side, I think there's a big question, and I know you've talked about this before, and I I think our listeners would love your thoughts on whether these, you know, closed source model providers who obviously are are spending a ton on training, whether there's a business there, right, or whether, you know, over time open source models just catch up and and ultimately, you know, I think there there's there's two sides of that argument, but would love for you to share just how you think about for the closed source model providers.
Speaker 131:45 - 32:32
我的意思是,我觉得这里面大概嵌着两个问题,对吧?一个是关于 inference(推理)本身:外面是否真的有足够多、足够有价值的 use cases(用例),让公司愿意支付远高于这些 data center 建设成本的钱?当然,就 AI coding 而言,作为第一个明显有大量 product-market fit(产品市场匹配)的市场,看起来至少已经有一些 use cases 确实达到了这个门槛。至于 training(训练)这一侧,我认为显然还有一个很大的问题,我知道你之前也谈过,我想我们的听众也会很想听听你的看法:这些 closed source model providers(闭源模型提供商)显然在 training 上花了巨资,那么这里面到底有没有一门生意,对吧?还是说,随着时间推移,open source models(开源模型)会追上来,最终——我觉得这件事确实有两面的论点——但还是很想请你分享一下,你是怎么思考这些 closed source model providers 的。
Speaker 132:32 - 32:37
Is there a business staying maybe three, six months ahead of the open source models?
Speaker 132:32 - 32:37
仅仅比 open source models 领先三个月或六个月,这能成为一门生意吗?
Speaker 232:37 - 33:35
Yeah. As you say, the open source models are one of the major reasons why the big companies cannot simply raise the prices because they are so close behind, and somebody Some commercial large language model breaks another benchmark record here, but a few months later, there's an open source model that catches up with that, which means there's enormous pressure on pricing, which means all these expensive data centers and GPUs where these companies invested in and took on debt to pay for that, they are not being profitable at the moment. So a stupid way would be just wait a little bit. Just wait for five years and compute is going to be 10 times cheaper. You will be able to do the same thing for oneten of the price or wait ten years and you will be able to do the same thing for 1% of the price.
Speaker 232:37 - 33:35
对,正如你说的,open source models 是大公司无法简单提价的主要原因之一,因为它们落后得并不远。某个商业 large language model(大语言模型)在这里打破了一个 benchmark(基准测试)纪录,但几个月后,就会有一个 open source model 追上来。这意味着定价面临巨大的压力,也就意味着这些公司投入巨资建设、并通过举债支付的昂贵 data centers 和 GPUs,目前并没有带来盈利。所以一个“愚蠢但有效”的做法其实就是稍微等一等。等五年,compute 会便宜 10 倍;你将能以十分之一的价格做同样的事。再等十年,你将能以 1% 的价格做同样的事。
Speaker 233:36 - 34:27
But then, of course, the big companies say, Oh, but if I do that, if I wait until the others do it, then I will lose maybe my market share or whatever, or I miss out on the opportunity to be the first to have a true AGI or something like that. And in in that case, once I have a true AGI that can solve everything, then I can take over the world and it will and then the investment will be nothing compared to the value of that. But but I think all of these these outlooks are overoptimistic, and and and probably you will see that that the current way of investing a lot in computers that aren't fast enough yet is going to backfire.
Speaker 233:36 - 34:27
但当然,大公司会说:哦,可如果我这么做,如果我等到别人先做了,那我可能就会失去 market share,或者错过第一个实现真正 AGI 的机会,诸如此类。而如果真是那样,一旦我拥有了一个能解决一切问题的真正 AGI,那我就可以接管世界,到那时,相比它的价值,现在的投资根本不算什么。但我认为,所有这些前景判断都过于乐观了,而且你大概率会看到,当前这种把大量资金投进那些还不够快的计算机的做法,最终会适得其反。
Speaker 134:28 - 34:45
I think related to our earlier conversation, I think there's this belief that, hey, if you can be the first to recursive self improvement or the first to have models that make it easier to develop the next models, that it becomes this kind of self perpetuating cycle that makes it really hard to catch the folks that are in the lead. What do you think of that?
Speaker 134:28 - 34:45
我觉得这和我们前面的讨论有关。现在似乎有一种看法:如果你能第一个实现 recursive self improvement(递归式自我改进),或者第一个拥有能让下一代模型更容易被开发出来的模型,那么事情就会进入一种自我强化的循环,使得后来者极难追上领先者。你怎么看?
Speaker 234:45 - 35:06
Not really because many of the guys who are interested in self improvement, they are open source guys. And and everybody's cooking with the same water, you know? Yeah. Everybody's cooking with the same water. Almost all of the important algorithms in AI, they were not invented at big companies or whatever.
Speaker 234:45 - 35:06
不太是,因为很多对 self improvement(自我改进)感兴趣的人,都是做 open source 的人。而且大家其实用的是同样的基础条件,你知道吗?对,大家用的都是同样的基础条件。AI 里几乎所有重要的算法,都不是在大公司之类的地方发明出来的。
Speaker 235:06 - 35:55
No, they were invented at little labs and without much funding, and especially the Because of self improvement stuff and the basic ideas with that all come from little labs, academic labs, And any company that wants to somehow grab that and make it its own doesn't have a moat because there are all these other PhD students out there who are super excited about Because of self improvement, And there's no way to keep a little advantage there for long enough to make a lot of money. So I think it will be very, very difficult for the huge companies to be very profitable in this business.
Speaker 235:06 - 35:55
不,这些东西是在小型实验室里、在资金并不充裕的情况下发明出来的,尤其是 Because of self improvement 这类东西,以及与之相关的那些基本想法,都是来自小实验室、学术实验室。任何公司如果想把这些东西抓过来据为己有,都不会有 moat(护城河),因为外面还有那么多 PhD 学生对 Because of self improvement 非常兴奋。而且,这种小小的优势根本不可能维持足够久,久到让你赚到很多钱。所以我认为,大公司要想在这个行业里获得非常高的利润,会非常非常困难。
Speaker 135:55 - 36:14
Yeah. Because basically, like, these ideas, you know, permeate throughout the broader ecosystem, you know, and so as a result, you know, when when one lab gets to to some level of RSI, it'll be kind of out in the open as well. And unless there's some massive compute need required for it or some kind of data advantage to distribution, it'll just be kind of available more broadly.
Speaker 135:55 - 36:14
对。因为基本上,这些想法会渗透到更广泛的生态系统里,所以一旦某个实验室达到某种程度的 RSI,它也会在相当大程度上暴露在公开环境中。除非它需要极其巨大的 compute(算力)投入,或者需要某种依赖分发的数据优势,否则它最终都会更广泛地可用。
Speaker 236:14 - 37:24
Nobody knows exactly. Maybe maybe maybe you you are lucky, and you stumble across a really good new way of conducting incremental self improvement, using the enormous resources that your particular company has for now and then achieve the AGI that you need to take over the world stock market and and and finish the whole game. I of course, there's a remote little chance of that happening, but all the indications point the other direction. And instead, AI is getting cheaper and cheaper, and there are more and more people who are, you know, involved in developing new AIs, and many of them are poor PhD students somewhere, haven't even founded their companies yet and so on. And then you really will have this AI for all situation where, you know, everything that seems impressive today, thirty years from now, will seem trivial because you can do a million times as much for the same price.
Speaker 236:14 - 37:24
没有人确切知道。也许,也许,也许你运气很好,偶然发现了一种非常好的新方法,能够利用你所在公司目前拥有的巨大资源,去进行渐进式的 self improvement,然后实现你所需要的 AGI,再去接管全球股票市场,把整场游戏彻底打完。当然,这种事情发生的可能性极小,但所有迹象都指向相反的方向。相反,AI 正在变得越来越便宜,参与开发新 AI 的人也越来越多,而其中很多人不过是某个地方贫穷的 PhD 学生,甚至还没创办自己的公司,等等。到那时,你真的会进入一种 AI for all 的局面——今天看起来令人印象深刻的一切,三十年后都会显得微不足道,因为同样的价格下,你能做的事情会多出一百万倍。
Speaker 237:24 - 37:29
And and it will be, you know, it will be just like with with the smartphones. Yeah.
Speaker 237:24 - 37:29
而且它会——你知道——就像 smartphones 当年那样。对。
Speaker 137:29 - 37:31
Yeah. But it sounds it sounds like you think a crash is coming.
Speaker 137:29 - 37:31
对。但听起来,你似乎认为一场 crash(崩盘)要来了。
Speaker 237:32 - 38:06
Well, not a crash in the sense of a civilization crash, but there will be one of these stock market crashes because at the moment, I think there's a lot of misallocation. Of course, the stock market is going to learn from these misallocations and is going to pursue different approaches, it's not going to be the end of civilization or something. No. But it's going to be a renormalization of what we currently have because at the moment we have just super expensive companies which don't really have much to offer.
Speaker 237:32 - 38:06
嗯,不是文明崩溃意义上的 crash,而是会出现那种股市崩盘,因为目前我认为存在大量的 misallocation(错误配置)。当然,股市会从这些错误配置中吸取教训,也会去尝试不同的路径;这不会是什么文明终结之类的事,不会。但它会是一次对我们当前局面的重新归一化,因为眼下有很多估值高得离谱、但实际上并没有太多可提供价值的公司。
Speaker 138:06 - 38:17
I feel like you've kind of publicly been less worried about AI safety today than maybe some of the other folks in the community. And I'm wondering, these past years, has anything changed your mind at all or are you kind of still in the same camp?
Speaker 138:06 - 38:17
我感觉,相比这个社群里的其他一些人,你如今在公开场合对 AI safety 的担忧似乎少了一些。我想知道,过去这些年里,有什么事情改变了你的看法吗,还是说你基本上还站在原来的那一派?
Speaker 238:17 - 39:25
I remember in the 2010s, we had lots of these safety conferences and alignment conferences and people writing letters that certain types of recursive self improvement should be forbidden. Of course, I never signed any of these letters, but back then, was already a big deal, at least for the machine learning specialists and how can we prevent AIs from becoming dangerous. Now I think all of these approaches were misguided in many ways, and it's kind of obvious that Well, first of all, this whole alignment business means that there is somebody who decides what is the nature of this alignment. How do I align these AIs to the needs of humans? But if you have 10 different humans in one room, they all will have different needs, and they will have different opinions about what is good for humans and what is not good.
Speaker 238:17 - 39:25
我记得在 2010s,我们开过很多这种 safety conference 和 alignment conference,还有人写公开信,要求禁止某些类型的 recursive self improvement。当然,我从来没有在这些信上签过名,但在当时,这已经是一个大议题了,至少对 machine learning 领域的专家来说是这样:我们该如何防止 AI 变得危险。现在我认为,所有这些方法在很多方面都走偏了,而且这其实很明显。首先,整个 alignment 这套说法本身就意味着,必须有某个人来决定这种 alignment 的性质是什么:我该如何让这些 AI 与 humans 的需求保持一致?但如果你把 10 个不同的人放进同一个房间,他们每个人的需求都会不同,他们对于什么对 humans 有益、什么无益,也都会有不同看法。
Speaker 239:25 - 40:07
And and and so this whole perspective is dominated by the idea that you give one objective function to this AI, which is supposed to optimize, you know, and that's then the alignment that you create through that. On the other hand, for you know, since since 1990 or so, we have built these artificial scientists, which are not really aligned much with anything because they, all the time, they invent their own new objective functions. They invent their own new problems all the time. They they they change their own objective functions. And really, starting in 1990, we had systems like that.
Speaker 239:25 - 40:07
而且,而且,而且,整个这种视角都被这样一种想法所主导:你给这个 AI 一个 objective function(目标函数),它就应该去优化它,你知道,然后你通过这种方式实现所谓的 alignment(对齐)。另一方面,从大约 1990 年开始,我们就已经构建出这些 artificial scientists(人工科学家),它们其实并没有和任何东西特别对齐,因为它们一直在发明全新的 objective function。它们一直都在发明全新的问题。它们会改变自己的 objective function。实际上,早在 1990 年,我们就已经有了这样的系统。
Speaker 240:07 - 40:29
So this whole premise of having systems that don't change their objectives didn't make any sense to me. It was very naive in my point of view. So I didn't sign those letters. I just couldn't. And and today, we we have these extreme cases where people are using AI to fight against each other.
Speaker 240:07 - 40:29
所以,这种“系统不会改变自身目标”这一整套前提,在我看来根本说不通。在我看来,那是非常天真的。所以我没有签那些联名信。我就是没法签。而且今天,我们已经看到一些极端案例:人们正在用 AI 来彼此对抗。
Speaker 240:29 - 41:29
If you look at the war between Ukraine and Russia, it's about AI based drones on this side, fighting against AI based drones on the other side. And there's no alignment whatsoever. And of course, you cannot expect in our world a universal government that is going to align all these AIs because all these different secret services and militaries, they all have different objectives, which are often totally conflicting with each other. It seems to me that all these efforts of the 2010s were kind of naive. Now, on the other hand, I think that If you want to get really smart AIs, you have to give them the opportunity to set themselves their own goals, you know, like little artificial scientists that that now ask a new question about what happens if I do this, and how does the world respond if I do that?
Speaker 240:29 - 41:29
如果你看 Ukraine 和 Russia 之间的战争,一边是基于 AI 的 drones(无人机)在作战,另一边也是基于 AI 的 drones 在作战。这里根本谈不上什么 alignment。当然,你也不能指望在我们的世界里出现一个普世政府,把所有这些 AI 都统一对齐起来,因为各种 secret services(情报机构)和 militaries(军方)都有各自不同的目标,而且这些目标往往彼此完全冲突。在我看来,2010 年代的这些努力多少有些天真。另一方面,我认为,如果你真的想获得非常聪明的 AI,你就必须给它们机会,让它们自己设定目标,就像小型 artificial scientists 一样,会主动提出新问题:如果我这样做会发生什么?如果我那样做,世界会如何回应?
Speaker 241:29 - 42:03
So only if you give them the freedom to ask their own questions and to solve their own self inventive problems, only then they will become really smart. On the other hand, they will become less predictable. That is, of course, the issue that is being addressed here. And and now people are asking, But in a world like that, where you have all kinds of AIs that set themselves their own goals, like they have done in my lab for many decades, isn't that super dangerous and so on? Think it's not going to be more dangerous than what you have now with humans.
Speaker 241:29 - 42:03
所以,只有当你给予它们自由,让它们提出自己的问题,并解决自己发明出来的问题时,它们才会真正变得聪明。另一方面,它们也会变得更难预测。当然,这正是这里正在讨论的问题。现在人们会问:但是,如果在那样一个世界里,存在各种各样自己给自己设定目标的 AI——就像它们在我的实验室里几十年来一直做的那样——那不会超级危险吗,等等?我认为,这不会比你现在面对的人类更危险。
Speaker 242:03 - 42:41
Humans also set themselves their own goals all the time and and come up with new questions to to ask. And and, yes, it is true that the kid my kids are unpredictable. I'm not sure what they are going to do in the future, but at least I can come contribute as a as a parent to to making them useful members of society. You know? And whenever they come up with with bad experiments, like, for example, let me take this magnifying glass and and focus the sunlight on this little ant here, then I'm going to punish them and say, that's bad.
Speaker 242:03 - 42:41
人类也一直在给自己设定目标,也一直在提出新的问题去追问。而且,没错,我的孩子确实是不可预测的。我不确定他们未来会做什么,但至少,作为父母,我可以为把他们培养成社会中有用的一员作出贡献。你知道的。每当他们想出一些糟糕的实验,比如说,让我拿起这个放大镜,把阳光聚焦到这只小蚂蚁身上,那我就会惩罚他们,并告诉他们,这样不对。
Speaker 242:41 - 43:32
You shouldn't do that. And and that's how I taught my kids to become reasonable members of society. And the same thing we are going to do with our robots and machines. And in the long run, once they are much smarter than we are, I think we have a certain we can expect a certain kind of protection through the fact that they are scientists, because artificial scientists will be super interested by life and by their own origins and civilization and by everything that led to their current state. And they will they will be fascinated by life and they will be greatly motivated to protect the source of interesting patterns and and and they will be motivated to protect it rather than destroy it.
Speaker 242:41 - 43:32
你不应该这么做。我就是这样教我的孩子成为社会中讲道理的一员的。对于我们的 robots(机器人)和 machines(机器),我们也会做同样的事。从长远来看,一旦它们变得比我们聪明得多,我认为我们可以期待某种保护,这种保护来自这样一个事实:它们是 scientists(科学家),因为 artificial scientists 会对生命、对它们自身的起源、对文明,以及对一切促成它们达到当前状态的事物产生强烈兴趣。它们会为生命着迷,并且会非常有动力去保护这种有趣模式的来源,它们会更倾向于保护它,而不是毁灭它。
Speaker 243:32 - 43:44
So From from this high level perspective, think there's a very strong reason why you shouldn't be too afraid of, you know, Arnold Schwarzenegger Terminator scenarios. We always like
Speaker 243:32 - 43:44
所以,从这种高层次视角来看,我认为有一个非常强的理由说明,你不必过于害怕那种 Arnold Schwarzenegger 式的 Terminator 场景。我们总喜欢——
Speaker 143:44 - 44:01
to end our interviews with with a bunch of, like, you know, quick fire questions where we we we shove a bunch of things into the end. And so maybe to start, actually, I'd love just to give our listeners context on you've obviously done so many interesting things over your career. How are you kind of splitting your time today and thinking about, like, what's most interesting to work on going forward?
Speaker 143:44 - 44:01
——在采访结尾加上一组类似快问快答的问题,就是把一堆问题都集中塞到最后。所以也许先从这里开始吧,我很想先给听众一些背景信息:显然,你的职业生涯中做过很多非常有意思的事情。你现在大致是如何分配自己的时间的?以及你如何看待未来什么方向最值得投入、最有意思去做?
Speaker 244:02 - 44:18
I'm a conservative guy, and I think the most interesting way of going forward is still the one that I pursued in the 70s and 80s when I tried to build this general purpose AI that learns to become smarter than myself such that I can retire. I'm still working on the same old thing.
Speaker 244:02 - 44:18
我是个比较保守的人,我认为继续前进的最有意思方式,仍然是我在 70 年代和 80 年代所追求的那条路:构建一种通用目的的 AI,让它通过学习变得比我自己更聪明,这样我就可以退休了。我现在仍然在做这件老事情。
Speaker 144:18 - 44:31
You've obviously, you know, have this kind of network of of a ton of amazing folks that have come through and worked under you and then got on to do great things as well in addition to you. Like, what have you learned about what makes a really good researcher?
Speaker 144:18 - 44:31
很明显,你拥有这样一个由许多非常了不起的人组成的网络,他们曾经来到你这里、在你手下工作,之后也都去成就了很多伟大的事情,除此之外你自己当然也是如此。比如说,关于什么样的人会成为真正优秀的研究者,你学到了什么?
Speaker 244:31 - 45:40
Many of the best PhD students that I have had the pleasure to work with, they usually focus on a particular question. So, yeah, that's the general big problem that we want to solve, but then there's this particular little thing that currently doesn't work. And then you have to look at the details and how do these weights change on this neural network, is trained by this particular learning algorithm, and why doesn't the network do what you want it to do? Then you have to debug a little tiny thing, and suddenly you see the devil is in the detail, and there is a little detail which you have overlooked so far. And if you fix that one, then suddenly everything works out, and suddenly you have a breakthrough, and then the same PhD student can come up with one research paper at a major conference after another, just because it's such a rich source of additional improvements that was triggered by this little, devil in the detail solution.
Speaker 244:31 - 45:40
我有幸合作过的许多最优秀的 PhD 学生,通常都会专注于一个具体问题。所以,是的,我们想解决的是那个总体上的大问题,但眼下总有这样一个特别具体的小问题还没有解决。然后你就必须去看细节:这个 neural network(神经网络)里的这些 weights(权重)是如何变化的,它是由这个特定的 learning algorithm(学习算法)训练出来的,为什么这个 network 不能按你想要的方式工作?接着你就得去 debug(调试)一个很小很小的问题,突然间你会发现,魔鬼藏在细节里,而那里正有一个你此前一直忽略的小细节。只要把那一点修好,突然一切就都通了,突然你就取得了突破;然后,同一个 PhD 学生就能接连在 major conference(顶级会议)上发出一篇又一篇 research paper(研究论文),就因为这个由“细节中的魔鬼”触发出来的解决方案,成为了后续大量改进的丰富源头。
Speaker 145:40 - 45:45
Do you think the transformer will persist as like the dominant architecture over the next five, ten years?
Speaker 145:40 - 45:45
你觉得在未来五年、十年里,transformer 还会继续作为主导架构存在吗?
Speaker 245:45 - 46:06
I guess some sort of transformer, Bill. But I think it's going to be a little bit more like the efficient transformers that we had in 1991, the linear transformer. Today, it's called the unnormalized linear transformer. I call it the fast way controller. And what is the interesting thing about it?
Speaker 245:45 - 46:06
我猜会是某种 transformer,Bill。但我觉得它会更像我们在 1991 年就有的那些高效 transformer,也就是 linear transformer。今天它被称为 unnormalized linear transformer。我把它叫作 fast weight controller。那它有趣的地方是什么呢?
Speaker 246:06 - 46:30
It scales linearly, which means that if you have 1,000 times more more text, then you need 1,000 times more compute. While the modern quadratic transformer of 2017, if you have 1,000 times more text, you need 1,000,000 times more compute. So everybody is worried about that. It's one of the reasons these data centers are now so super expensive. You have to do so much compute.
Speaker 246:06 - 46:30
它是线性扩展的,这意味着如果你的文本多了 1,000 倍,你就需要多 1,000 倍的 compute(算力)。而 2017 年那种现代的二次复杂度 transformer,如果你的文本多了 1,000 倍,你就需要多 1,000,000 倍的 compute。所以大家都在担心这个问题。这也是为什么这些 data center(数据中心)现在会变得如此昂贵的原因之一:你必须做这么多计算。
Speaker 246:30 - 47:10
Everybody wants to bring down this complexity into a more reasonable linear complexity or maybe log linear complexity. There are quite a few transformer variants that are going in this direction, or there is stuff that is doing the XLSTM, which is a thing that combines aspects of the old linear transformers. You want to get the complexity down, and it's related to what you asked before. Intelligence is about doing the same thing with less effort. Yeah, we want to reduce the effort.
Speaker 246:30 - 47:10
每个人都想把这种 complexity(复杂度)降到更合理的线性复杂度,或者也许是 log linear complexity(对数线性复杂度)。已经有不少 transformer 的变体在朝这个方向走,或者还有一些像 XLSTM 这样的东西,它结合了老式 linear transformer 的一些特点。你就是想把复杂度降下来,而这也和你之前问的问题有关。智能,就是用更少的努力去做同样的事。对,我们想减少这种努力。
Speaker 147:10 - 47:20
Well, I love that. Look. This has been a a fascinating conversation. I'm sure there's a ton of threads that folks will wanna pull on, you know, in in addition to what we've discussed here. So I wanna make sure to leave the last word to you.
Speaker 147:10 - 47:20
我很喜欢这点。你看,这一直是一场非常精彩的对话。我相信,除了我们这里已经讨论过的内容之外,大家一定还会想顺着很多线索继续深挖。所以我想确保把最后的话留给你。
Speaker 147:20 - 47:27
Where can folks go to learn more about your work and, really, anything you wanna point our our listeners to? The mic is yours.
Speaker 147:20 - 47:27
大家可以去哪里进一步了解你的工作?或者说,你还有什么想推荐给我们的听众?现在话筒交给你。
Speaker 247:27 - 47:35
Where can you learn what I think is interesting? Look at my blog. It's easy to Google if you Google for a year again.
Speaker 247:27 - 47:35
你们到哪里可以了解我认为什么是有意思的?去看我的 blog 就行。如果你把我的名字和某个年份一起拿去 Google 搜,很容易就能找到。
Speaker 147:35 - 47:37
Yeah. We'll we'll link to it too.
Speaker 147:35 - 47:37
对,我们也会附上链接。
Speaker 247:37 - 48:12
You will find it. And and there are overview pages with links to the original papers on meta learning, on artificial curiosity, on the formal theory of fun and creativity, also on the history of our field. The field of machine learning is is about the science of credit assignment and apply that to the field itself and to invent convolutional neural networks and to invent deep learning. All this stuff is nicely listed and explained there. And it also explains what I consider the most important stuff and including what I just mentioned.
Speaker 247:37 - 48:12
你们会找到的。而且那里还有一些 overview pages(概览页),附有原始论文的链接,内容包括 meta learning、artificial curiosity、fun and creativity 的 formal theory(形式理论),以及我们这个领域的历史。machine learning 这个领域,讲的是 credit assignment(信用分配)的科学;也就是把这套思想应用到这个领域本身,进而发明 convolutional neural networks,发明 deep learning。所有这些内容都在那里被很好地列出并解释了。那里也说明了我认为什么是最重要的内容,包括我刚才提到的那些。
Speaker 248:13 - 49:06
The important thing is physical AI outside the screen, so not behind the screen but in the real world. Once we have a robot For hundreds of years, people have talked about self replicating machines, but nobody had any idea how to get there. But now we see an opening. For the first time, we now can have or maybe we'll we will soon have robots that can learn through imitation or reinforcement learning to operate the existing machines, all the existing machines that already are part of our civilization. Once you have a machine that can operate all the machines that humans currently are are operating, then you have a new kind of life.
Speaker 248:13 - 49:06
重要的是屏幕之外的 physical AI,也就是不在屏幕后面,而是在真实世界里。一旦我们拥有了 robot,几百年来,人们一直在谈论 self replicating machines(自我复制机器),但从来没人真正知道该怎么实现。不过现在我们看到了一个突破口。我们第一次有可能——或者也许很快就会——拥有能够通过 imitation(模仿)或 reinforcement learning(强化学习)来学会操作现有机器的 robots,操作那些已经构成我们文明一部分的所有现有机器。一旦你拥有了一台能够操作目前人类正在操作的所有机器的机器,那么你就拥有了一种新的生命形式。
Speaker 249:06 - 50:10
Then you have a way of implementing this ultimate scaling machine because you can have robots that make more of themselves. I've said that for decades, and now it's getting closer to reality. You don't need super smart robots for doing that. Just smart enough to learn to operate all the existing machines, and a collection of machines like that can make more of itself, and something like that can also improve itself, not only make replicas of itself, because all the concepts of machine learning that we already have for AI behind the screen the virtual world, all these concepts we are going to apply to self improving robot societies. And something like that is not only going to work in the biosphere, but also on the moon or on Mercury, where there's a lot of material for building infrastructure and bigger AIs and more AIs and more robots and huge spacecraft and all kinds of stuff that you need to colonize the solar system.
Speaker 249:06 - 50:10
接着,你就有了一种实现这种 ultimate scaling machine(终极扩展机器)的方式,因为你可以拥有能够制造更多自己的 robots。几十年来我一直在这么说,而现在它正越来越接近现实。你并不需要超级聪明的 robots 来做到这一点,只需要它们足够聪明,能学会操作所有现有的机器;而这样一组机器就能够制造出更多的自己。类似这样的系统还能够改进自己,而不只是复制自己,因为我们已经用于屏幕之后、虚拟世界中的 AI 的那些 machine learning 概念,都会被应用到能够自我改进的 robot societies(机器人社会)上。而这样的东西不仅会在 biosphere(生物圈)中发挥作用,也会在 moon 或 Mercury 上发挥作用,因为那里有大量材料可用于建造基础设施、更大的 AI、更多的 AI、更多的 robots、巨型 spacecraft,以及殖民整个太阳系所需要的各种东西。
Speaker 150:10 - 50:23
Well, I think that's the the perfect note to to end on, an incredibly exciting vision for the for the future, and and and, seriously, thank you so much for for taking the time to chat through everything here. It's it's such a privilege to to get a chance to talk, and I know our our listeners will enjoy it too.
Speaker 150:10 - 50:23
我觉得这是一个再完美不过的收尾,一个对未来极其令人兴奋的愿景。也真的非常感谢你愿意花时间把这一切都聊一遍。能有机会进行这次对话是一种莫大的荣幸,我知道我们的听众也一定会喜欢。
Speaker 250:23 - 50:25
It was my pleasure. Thank you, Jacob.
Speaker 250:23 - 50:25
这是我的荣幸。谢谢你,Jacob。
Speaker 150:25 - 50:51
I'm Jacob Efron, and this has been unsupervised learning, A podcast where I get to talk to the smartest people in AI, and ask them tons of questions about what's happening with models, and what it means for businesses in the world. As I hope is clear, I have a ton of fun doing this. It's a nights and weekends project, in addition to my day job as an investor at Redpoint. But our ability to get these incredible guests on really comes from folks like you subscribing to the podcast, sharing it with friends. It's really what ultimately makes this whole thing work.
Speaker 150:25 - 50:51
我是 Jacob Efron,这里是 unsupervised learning。这是一档 podcast,在这里我可以和 AI 领域最聪明的人交流,向他们提出大量问题,聊聊 models 正在发生什么变化,以及这对现实世界中的企业意味着什么。希望大家已经能看出来,我做这件事非常开心。除了我在 Redpoint 担任投资人的本职工作之外,这还是一个利用晚上和周末时间来做的项目。但我们之所以能请到这些不可思议的嘉宾,确实离不开像你这样订阅这档 podcast、并把它分享给朋友的人。归根结底,真正让整件事运转起来的,就是这些支持。
Speaker 150:51 - 50:55
And so please consider doing that, and thank you so much for your support and listening. We'll see you next episode.
Speaker 150:51 - 50:55
所以也请你考虑这样做,非常感谢你的支持和收听。我们下期节目再见。
原文 ↗https://www.youtube.com/watch?v=RKjR8DQ40po
BuildSpeak — 关于本项目BUILT IN PUBLIC · 跟随 builders 而非 influencers