Building the Automated AGI Lab: Core Automation's Jerry Tworek and Rohan Anil
发布时间 来源
Episode 设置
Jerry，前OpenAI副总裁，以及Rohan，谷歌大脑和Gemini的关键人物，共同创立了Core Automation，怀揣着一个宏大的愿景：取代Transformer架构，开创一个真正能从经验中学习的人工智能新时代。他们认为，尽管Transformer对当前的人工智能热潮具有奠基性作用，但其局限性使得彻底的变革成为必然。
Jerry承认Transformer的深远影响和经济价值，它使得可扩展的预训练和强化学习（RL）成为可能。然而，他认为该领域目前专注于渐进式改进（即让它们更便宜/更高效），而非从根本上更强大或更具表现力的发展。他相信，核心瓶颈是架构本身。Transformer尽管能力强大，但在现实世界的适应性方面表现不佳。它们在受控的实验室环境中进行训练，但在部署时却面临“更混乱的”数据分布。它们的“上下文学习”（in-context learning）在范围和持续时间上都很有限（例如，Codecs只有20分钟），而持续微调则面临灾难性遗忘和数据效率低下的问题。Jerry强调，当前的人工智能系统是有效的“人-LLM混合体”，但未能达到真正的通用人工智能（AGI）——一种能够在无需人类干预的情况下自我改进的模型。他强调“从经验中学习”比单纯的强化学习范围更广，他将足球（类似强化学习）与深层数学推理进行类比。Core Automation旨在开发新的算法，以架构的方式呈现，使模型能够在测试时、基于真实世界用户数据、在更长的时间跨度内持续学习和适应。
Rohan表示赞同，指出Transformer尽管在预训练（数据压缩）方面效率很高，但在推理方面却效率低下，尤其是因为其“浅层计算深度”和逐令牌生成的特点。他认为思维链推理（chain-of-thought reasoning）是一种效率低下的权宜之计，旨在弥补这种架构上的局限性。Rohan提倡一种整体方法，将预训练和强化学习相结合，以创建更高效的学习过程。他强调优化方法的关键作用，认为更强大的优化器可以释放更复杂、更深层架构的潜力，而这些架构目前很难训练。他还将当前的数字硬件与生物学习进行对比，暗示要实现显著的效率提升可能需要更具模拟特性、更注重硬件的设计。
两位创始人解释道，他们决定创办Core Automation源于他们在大型实验室中观察到的停滞，这些实验室目前由于竞争压力专注于现有Transformer技术的短期扩展。相比之下，Core Automation旨在成为“世界上最自动化的实验室”，旨在促进快速实验并突破架构研究的界限。他们计划自动化“内核生成”（即优化硬件计算的低级代码）等瓶颈，这目前需要稀有的人类专业知识和大量时间。例如，优化QR内核可以带来60倍的加速，但超出了LLM的当前能力。通过大幅加速迭代周期——可能每天执行几十甚至几百个实验——他们相信他们可以有效地探索巨大的可能架构空间。他们的最终目标是找到一种能够展现有意义的、长期适应性的“卓越架构”，甚至可能通过观察他们自己的模型每天在Core Automation科学家的工作中变得更出色来实现。
Jerry, former VP at OpenAI, and Rohan, a key figure at Google Brain and Gemini, have founded Core Automation with a bold vision: to replace the Transformer architecture and pioneer a new era of AI that truly learns from experience. They argue that while Transformers have been foundational to the current AI boom, their limitations necessitate a radical shift.
Jerry acknowledges the profound impact and economic value of Transformers, enabling scalable pre-training and reinforcement learning (RL). However, he contends that the field is now focused on incremental improvements (making them cheaper/more efficient) rather than fundamentally more powerful or expressive. The core bottleneck, he believes, is the architecture itself. Transformers, despite their capabilities, struggle with real-world adaptability. They are trained in controlled lab environments but face "messier" distributions in deployment. Their "in-context learning" is limited in scope and duration (e.g., 20 minutes for Codecs), while continuous fine-tuning suffers from catastrophic forgetting and data inefficiency. Jerry highlights that current AI systems are effective "human-LLM hybrids" but fall short of true AGI – a model capable of improving itself without human intervention. He emphasizes that "learning from experience" is broader than just RL, drawing parallels between football (RL-like) and deep mathematical reasoning. Core Automation aims to develop new algorithms, expressed architecturally, that enable models to learn and adapt continuously at test time, on real-world user data, over much longer horizons.
Rohan concurs, pointing out that Transformers, while efficient for pre-training (compressing data), are inefficient for inference, particularly due to their "shallow computational depth" and token-by-token generation. He sees chain-of-thought reasoning as an inefficient workaround to compensate for this architectural limitation. Rohan advocates for a holistic approach, combining pre-training and RL to create a more efficient learning procedure. He stresses the critical role of optimization methods, arguing that stronger optimizers can unlock the potential of more complex, deeper architectures that are currently difficult to train. He also contrasts current digital hardware with biological learning, suggesting that significant efficiency gains might require more analog, hardware-aware designs.
Both founders explain their decision to start Core Automation stems from a perceived stagnation in larger labs, which are currently focused on short-term scaling of existing Transformer technology due to competitive pressures. Core Automation, in contrast, aims to be the "most automated lab in the world," designed to facilitate rapid experimentation and push the boundaries of architectural research. They plan to automate bottlenecks like "kernel generation" (the low-level code that optimizes computations on hardware), which currently requires rare human expertise and significant time. For example, optimizing a QR kernel can yield a 60x speedup but is beyond the current capabilities of LLMs. By dramatically accelerating the iteration cycle – potentially executing dozens or even hundreds of experiments daily – they believe they can efficiently search the vast space of possible architectures. Their ultimate goal is to find a "superior architecture" that demonstrates meaningful, long-term adaptability, perhaps even by observing their own models becoming better at the core automation scientists' work each day.
摘要
Jerry Tworek led reasoning at OpenAI, convinced that scaling reinforcement learning was the path to AGI. Rohan Anil co-led Gemini pre-training and built the Shampoo optimizer. Now they've teamed up at Core Automation on a contrarian premise: the transformer has carried us as far as it can, and the bottleneck to smarter systems is no longer scale — it's the architecture itself. The missing capability is continual learning, models that adapt at test time, which transformers can't do. In-context learning taps out fast (Codex needs compacting after ~20 minutes) and fine-tuning invites catastrophic forgetting. Rohan argues pre-training and RL should be optimized end-to-end, and that transformers spend computation inefficiently. They lay out why the largest labs won't chase alternatives while locked in the coding-agent race, and why building the world's most automated lab starts with automating kernel generation—the one place frontier models still lose to a high-taste human.
Hosted by Sonya Huang and Pat Grady, Sequoia Capital
00:00 Introduction
01:46 Appreciating Transformers
02:44 Scaling Hits Limits
04:54 Why Architecture Matters
05:32 RL Reality Check
07:32 Test Time Learning
09:52 Economics Of Scaling
12:47 Why Start A Company
14:24 Rohan On Transformers
19:11 Computational Depth Problem
20:32 When Transformers Top Out
23:22 Beyond Reinforcement Learning
26:41 Optimization And Efficiency
34:24 Building An Automated Lab
39:45 Kernel Automation Roadmap
GPT-4正在为你翻译摘要中......
中英文字稿 
如果我踢足球，比如说，这看起来很像强化学习。我会多次踢球，每次都会稍微调整一下，然后看看能否达到我想要的效果。这当中有一些自我强化的过程。而当我学习数学时，这是一种完全不同的体验。这更像是去理解一些困难的概念，并在脑海中深入思考，直到我把它们连接起来。这两种方式在某种程度上都是从经验中学习，只是方式非常不同。我们可能比以往任何时候都更依赖计算机来从经验中学习。但强化学习并不是经验学习的终点。在未来的几年里，研究人员会找到更好的方法来利用这些数据。
▶ 英文原文 ⏱
If I play football, for example, it looks very, very close to reinforcement learning. I kick a ball a lot of times. And every time I adjust it a little bit and I see if it roughly matches what I, what I wanted and I, there are some self-reinforcement happening. When I learned mathematics, it's very different type of thing. It's, it's like reading about hard concepts and thinking about them very deeply inside my head until things click and I, until, until I have them connected. And both of those in some way are learning from experience. They are just very different. We probably are spending the most computer than ever on learning from experience. But the reinforcement learning is not the end of learning from experience. And there will be better approaches that researchers will be coming up in the coming years on how to, how to use that data.
感谢您今天加入我们。你们两位是现今旧金山最热门的新实验室之一——Core Automation的创始人。在创办Core Automation之前，你们分别领导过一些AI时代最重要的研究项目。Jerry，你曾是OpenAI的副总裁，负责草莓和推理团队的工作。Rohan，你则是Gemini的四位预训练负责人之一，在此之前，还主导了Google Brain许多基础性的AI研究，并在Google和Anthropic担任解决问题的专家。因此，你们两位在最前沿研究领域都有丰富的经验，我非常期待深入了解你们的工作。
▶ 英文原文 ⏱
Thank you so much for joining us today. The two of you are the founders of Core Automation, one of the hottest neo labs in San Francisco right now. And before starting Core Automation, you led some of the most important research projects of the AI era. Jerry, you were VP at OpenAI, where you worked among, amongst other things on running the strawberry and reasonings teams. And Rohan, you were one of two of the four pre-training leads at Gemini. And before that led a lot of the fundamental AI research at Google Brain, and we're the fix-it guy across Google, and then at Anthropic. And so between the two of you, you've seen more than your fair share of what the world looks like in terms of doing frontier research. And so I'm very, very excited to dig in.
我们先从你开始，Jerry。最近你发了一条非常犀利的推文。你说，取代Transformer的第一步是深刻理解它们已经为我们带来的成就。这是对Transformer的悼词吗？这是什么意思？非常感谢Sonja邀请我们。我觉得最近我的许多采访都是在解释我的推文、我想传达什么。理解Transformer就像是要知道它擅长什么，这样你就不会去解决它已经解决得很好的问题。你需要专注于它的弱点，并理解它的优缺点。在很多工作中，人们做的事情就是努力让Transformer更便宜、更高效。
▶ 英文原文 ⏱
Let's start with you, Jerry. You tweeted a very spicy take recently. The first step to replacing Transformers is appreciating deeply how far they were able to carry us. Is that a eulogy for the Transformer? What does that mean? Thank you very much for inviting us here, Sonja. I feel like a lot of my interviews these days is explaining my tweets and what did I mean? But appreciating Transformer means like understanding what it does well, so you're not solving the problems that it is solving well. You have to focus on its weaknesses. You have to understand good parts and bad parts. And it's very easy in a lot of the work what people are doing in architectures is trying to make Transformers cheaper and trying to make Transformer more efficient.
我很少看到有人在思考如何让Transformer模型更强大、更具表现力。实际上，看到模型的弱点和看到模型的强项几乎是一样的道理，这只是更深入地了解Transformer的结构而已。但我认为，我们现在正处于一个阶段：我们已经非常擅长训练非常大的模型，并且掌握了两种算法——大规模的预训练和大规模的强化学习。我常常在想，机器学习的下一步是什么。我认为，目前更好模型和更智能系统的瓶颈在于其自身的架构。
▶ 英文原文 ⏱
I very rarely see people thinking about how do we make Transformers more powerful, trying to do more expressive. But seeing someone with weak parts and seeing someone with strong parts are almost the same thing. It's just understanding the shape of Transformer a little bit more. But I think right now we are in this stage. We got really, really good at training really, really big models. We mastered two algorithms. We mastered pre-training at a large scale and we mastered reinforcement learning at a large scale. And I'm asking myself a lot what is next in machine learning. And I think at this moment what the bottleneck is to better models and to smarter systems is the architecture itself.
在过去六年里，我们一直在尝试为本质上是两种相同操作的MOE和注意力机制添加更多的参数，现在是时候重新审视我们所乘坐的这趟列车了。当我思考我们目前所处的位置以及正在做的事情时，我想到了编解码器和云代码为我们带来的帮助。我对这些系统、编码、工作流程自动化以及我们在过去六年中扩展而建立的产品系统非常感激。我认为，这是一种初步的思考方式：如果我们想要进行替代，就需要审视我们所处的位置，以及已经解决了哪些问题，以便开始思考下一阶段是什么，还有哪些问题尚未解决。
▶ 英文原文 ⏱
It is this moment to revisit the train we've been riding for the last six years of trying to add more and more parameters to essentially two of the same operations, which is MOE and attention. And when I'm thinking about it, like where we are today and what we are doing, I am thinking a lot about what codecs and what cloud code are doing for us. And I'm really, really appreciative of those systems and of the coding and of the workflow automation and of the systems of the products that we have today that we essentially have built over those six years of scaling. And I think, I think this is, this is the first step of thinking, like what is the, if we want to work on the replacement, we need to like, see where we are, what problems we have solved to like, start seeing what the next stage is, what problems we haven't solved yet.
我们到底漏掉了什么？每次我使用编解码器并成功完成一个任务时，我总是在想，为什么我没有尝试更努力地去推动这个事情。每次我上班时，我都会使用编解码器来做很多事情，但我依然要来到工作岗位，依然要让它为我做某些事情。我总是问自己，我在那里究竟有什么用？核心自动化的名字和理念是希望自动化任务。为什么这些事情还没有实现自动化？为什么编解码器不能为我完成所有事情？
▶ 英文原文 ⏱
What, what, what kind of, are we missing? And this is, this is kind of whenever, whenever I use codecs and I am successful at a task. I also start thinking, why did I, why didn't I try to push that thing harder? Whenever, whenever I come to work, there are a lot of things I do with codecs, but I still come to work. I still ask it to do certain things for me. And I'm always asking myself, why am I even needed there? Why is core automation is name and its concept is we want to, we want to be automating tasks. And why, why, why are those things not yet automated? Why, why, why is not codecs doing everything for me?
这段话可以翻译成中文如下：
“这是一个关于我们想把研究引向何处的问题。针对这个研究，我在思考我们需要什么样的模型、什么样的系统以及什么样的品质，这些是我们现在所没有的。而这是我最近常常思考的问题。你提到的观点是，从架构上找问题，我认为这是一种逆向思维。那么是什么让你有这样的观点？是什么让你认为是架构的问题？基本上，问题是什么？这可以追溯到之前的影响。我认为的问题在于，模型是在实验室中训练的，但却在真实世界中应用。这就是存在的根本矛盾。”
▶ 英文原文 ⏱
And this is, this is the question of like the research where we want to go. And with that research, I'm trying to think what kind of models, what kind of, what kind of systems do we need? What kind of qualities do we need? What that, that, that's, we don't, we don't have today. And that's, that's what I'm thinking a lot these days. And you have this starting premise of the architecture is the issue, which I think is a contrarian point of view. So what, what led you to that point of view? What did you see that made you think the architecture was the issue? It's fundamentally, what is the issue? It's, it's, it's, it like comes back from the, from, from the previous implication. What, what I think is the issue is that the models are being trained in the lab and are being deployed in the, in the real world. That is that, that, that is the fundamental, fundamental tension that is, that is there.
我的失望有一部分来自于我个人的故事。当我们在 OpenAI 开始进行强化学习扩展的研究和进展时，我坚信扩展强化学习是通向通用人工智能（AGI）的必要步骤。从我开始在 OpenAI 工作起，我一直是强化学习的坚定支持者，认为这就是我们需要关注和做的事情。我看到大型语言模型（LLM），像 GPT-3 和 GPT-4，不断被扩展到更高的水平，而我们在强化学习方面做的却很少。我心中一直有一种信念：一旦我们开始扩大强化学习的规模，我们就能解决所有问题。
▶ 英文原文 ⏱
And, um, like a bit of, a bit of my disappointment can, comes from my, my, my personal story. Whenever we were starting, uh, the research and progress on scaling up reinforcement learning at OpenAI, I basically believe that scaling up reinforcement learning is a necessary stepping stone on a path to AGI, uh, since, since, since I started like, uh, working, working at OpenAI. And I was always reinforcement learning maximalist. I was always believed this is what we need to focus on. This is what we need to do. I've seen LLMs being scaled up to, to, to, to higher and higher levels through GPT-3 to GPT-4, and we're still doing very little RL. And I had this, this internal belief that the moment we start scaling up RL, we'll, we'll, we'll, we'll, we'll be able to solve, solve all the problems.
我们最终开始解决并扩大强化学习的规模。我当时就在其中，我在中心地带。我当时在想，如果你在2024年问杰瑞，我们何时会达到通用人工智能（AGI），我会说2025年将是那一年。这是我们解决一切问题的时刻。我看到我们在不断训练一个又一个模型，这些模型变得越来越好，所有的基准测试成绩都在上升。那么我们是否也在那一刻解决了所有的现实世界任务呢？遗憾的是，并没有。我们仍然有工作要做。我意识到存在这样一个区别：我们用来评估模型的所有基准测试，实际上与我们用于训练模型的内容基本上是一样的。
▶ 英文原文 ⏱
And we eventually started solving, uh, scaling up RL. I was, I was just in there. I was, I was in the center of it. I was thinking here we are, if you ask Jerry in 2024, when do we get AGI? I would say 2025 will be, will be that year. This, this, this, this, this is where we solve everything. And I saw us training model after model, this model was getting better and better. All the benchmark scores were going up. And did we also solve all the real world tasks at that moment? Unfortunately, unfortunately not. We, we, we, we still have work. And I realized there was this bit of distinction. As all the benchmarks that we are evaluating our models, they were essentially the same thing as we were training the models on.
所有的评估和训练相关任务本质上是同一个问题的不同方面，但现实世界中的分布和任务要复杂得多，模糊得多，差异也更大。我们的训练数据并没有真正地反映出实际使用案例。尽管我们基本上已经最大化了所有任务，但如果问任何一个模型训练人员，他们的主要问题是什么，他们会说：我没有足够困难的任务，没有足够的数据来训练我们的模型。然而，即便如此，我们依然无法完全覆盖现实世界的分布。因此，我的结论是，我们需要能够在测试时学习的模型，需要能够在用户的数据上真实世界任务中学习的模型，使之适应真实世界的分布。
▶ 英文原文 ⏱
Like all the evals and training side tasks are the same sides of the coin, but the real world distribution and real world task is much messier, much murkier, much more, much more different. Our, our, our training data didn't really replicate the real world use cases. And despite us basically maximizing all the tasks. If you see, ask anyone training models, Hey, what is one of your main issues? I don't have hard enough tasks. I don't, I don't have what to, what to train our model on yet. We are still not covering the, the entirety of the, of the real world distribution. From that, my conclusion is we need to have models that learn at test time, need to have models that learn with users on their data, on their real world task, on the real world distribution.
在那个时候，当有人问为什么，为什么，为什么，为什么我们今天没有实现这个？为什么，为什么，为什么，为什么变压器模型没有在任何地方学习？基本上，在测试时我们可以进行两种类型的学习。我们可以进行变压器的上下文学习，这种学习没有灾难性遗忘的根本问题。它在数据效率上表现良好，这点很棒，但它的可扩展性不是很好，我们只能做到有限的程度，存在一些机械上的限制，比如你在构建上下文时到底在做什么，但也许我们以后可以再讨论这个问题。
▶ 英文原文 ⏱
And there, when you are asked why, why, why, why, why don't we have that today? Why, why, why, why are transformers not, not learning any anywhere? And there are essentially two types of learning that we could be doing at test time. We could be doing in context learning essentially of transformers, which is, it doesn't have fundamental problems of catastrophic forgetting. It doesn't have that issue. It is pretty data efficient, so that, that is great, but it's not very scalable. We only can have so much of it, this limited and has some more, even more of mechanical limitations of, uh, what actually are you doing when you, when you build context, but maybe, maybe we can, we can come back to it later.
我们的“上下文学习”能力非常有限，能够处理的数据量也非常小。当我使用编码器时，大约每使用20分钟，我就需要压缩数据并转移它，但这些数据并不多。如果我们只能学习20分钟的数据，真的不算多。第二点是模型的微调。我们可以尝试不断微调我们的模型，但这会带来灾难性遗忘的问题，以及极低的数据效率问题。这些问题都不容易解决，也没有很容易找到解决办法。人们一直在努力寻找解决方案，如果这些问题容易解决，早就有人解决了。
▶ 英文原文 ⏱
But we have, we have in context learning, which is very limited and very, very small amount of data. Whenever I'm using codecs, roughly around 20 minutes of usage, I need to, I need to compact it and move, and move it afterwards, which is not that much, not that much data. If we have all, all we can learn is for, for 20 minutes, it's not that much. And the second thing is fine tuning. We could try to continuously fine tune our models, but then those have the issues of catastrophic forgetting. We have issues of very low data efficiency and neither of those are very solvable. Neither of those are very easy to, uh, to find ways people have been trying. If there was, were easy to solve some already solve it.
所以我个人的看法是，我们需要找到一个可以元学习的算法，这个算法可以在架构层面上表达。这个算法需要能够代表学习是什么样子的，尤其是能够在更长的时间跨度上有效的学习。你认为这个架构会类似于Transformer吗？因为根据我的理解，OpenAI 长期以来一直试图扩展强化学习，而直到Transformer出现，才似乎有了一种可以扩展RL（强化学习）的可扩展基础。那么，我们应该如何去思考扩展这个新系统呢？这是个很好的问题。我认为这些事情几乎是同时发生的。但其中的大部分变化实际上与经济因素有关，因为从技术上讲是可以扩展LSTMs的，只是没有人真的敢于尝试这条路。
▶ 英文原文 ⏱
So my personal belief is we need to find an algorithm that we can, we can meta-learn that we can express on the architectural layer. That can represent how does, how does learning look like? How does learning look like that can work on much, much longer horizons? Do you expect the architecture will look transformer-like? Because my, my, my understanding from the, from the chief seats is that, you know, OpenAI had been trying to scale up reinforcement learning for a long time. And it wasn't until the transformer came about that it seemed like there was an even kind of scalable prior on the world upon which to even scale RL. And so how do you even go about trying to think about scaling up this, this new regime? Yeah, it's a, it's a, it's a great question. I think those two things happened at the same time. But if anything that, that happened there was, was mostly about, about economics, because technically you can scale up LSTMs. Just no one, no one really, really dared to go in that direction.
他们的扩展性确实差得多。论文中讨论了扩展损失，其中比较了LSTM和Transformer，结果表明Transformer的扩展损失表现更好。可以想象一个没有发明Transformer的世界，我们可能会继续扩展LSTM，并有一些模型，但这些模型的训练成本会更高，效果也不如Transformer令人印象深刻。这会导致更差的用户体验，可能也不会有人愿意投入大量资金去训练那些庞大的LSTM模型，因为无法获得相应的市场回报。而Transformer的与众不同之处在于其经济价值：训练Transformer的成本低于它所带来的收益，这可以说是机器学习的奇迹之处。这就是为什么我们要如此重视Transformer的原因。
▶ 英文原文 ⏱
And they did scale much, much more poorly. They're, they're, they're, they're, they're scaling in a scaling loss paper. There is, uh, presented a comparison of LSTMs and transformers and fundamentally the scaling loss of transformers was better. There is a world where we never invented transformers and we would be scaling LSTMs and we would be having some models, but because they would be much more expensive to train and much less impressive as a product. We would have, we would have just this worse experience and maybe no one would be able to convince people to spend as many dollars training that those gigantic LSTMs because we wouldn't get a market return. The, the, the, the majestic thing about transformer, which goes back to like, why, why, why, why do we have to appreciate transformers so deeply is that transformers are economically valuable. The training them, the cost of training them is lower than the revenue that they, that they generate, which is just magic of machine learning.
这段文字的意思是：并不是说这种方法本身就有保证，但对于长短期记忆网络（LSTM）来说，情况可能并不是这样的，这使事情发生了变化。实际上，大多数架构在很多方面都是可以扩展的。我认为，之前人们不愿意扩展的原因之一是，在OpenAI之前的研究者对这种做法持有很大保留，认为其不符合科学性。过去，算法研究强调的是如何在同样的计算预算下得到更好的结果，更注重效率提升。而OpenAI在当时采取了一种与众不同的策略，他们表示不在乎开发出更好的算法，而是专注于研发更具扩展性的算法，并进行更大规模的计算以获得更好的效果。为此，OpenAI长期以来受到了很多业内人士的批评。
▶ 英文原文 ⏱
And it's not, not guaranteed by itself, but for, for LSTMs, it probably wouldn't be that way, which, which, which made it happen. But you can, in many ways you can scale most of the architectures. I think, I think a lot of reasons why people didn't scale things before was because researchers before open AI had a lot of reluctance to scaling. It was often seen as unscientific and research in algorithm was, was providing, how do we become more and more efficient? How do we, for the same compute budget, get better and better results? And it was, it was a bit of a contrarian bet by open AI at that moment to try to say, hey, we don't care about better and better algorithms. We care about more and more scalable algorithms and how do we, how do we pour more and more compute and get better, get better results, which open AI was criticized repeatedly by many people in the community for, for a long time.
但正因为如此，我们才能拥有今天的模型。我认为有很多架构是可以扩展的。我是核心自动化任务的一员，我们的信念是，长期以来很多架构研究的规模都太小了，很多人尝试在小数据集和小计算环境中先测试他们的架构，然后再看看它是否能扩展。但例如，当你在做强化学习的工作时，你会知道要取得有趣的结果，你需要一定程度的计算能力，只有这样才能看到模型的潜力。
▶ 英文原文 ⏱
But thanks to that, we have, we have the models that we, that we have today. And I think there are tons of architectures that can be scaled up. And I am part of the core automations mission. And our, our belief is that a lot of architectural research happened at too small scale for too long time, a lot of people are trying to say, Hey, let's, let's try to first try our architecture on a, on a, on a small data set on a small, in a small compute regime. And then, and then, and then see where, where, where, where it scales only after, after you prove itself. But for example, when you do work on reinforcement learning, you know that to get to any interesting results, you only, you need certain level of compute to even see the capabilities in the model.
强化学习需要一个基本能力的基础才能开始工作。所以，我的观点是，可能有很多架构需要一个基础的计算能力，才能开始做一些有趣的、实用的事情。那么，我可以问你一个可能有些敏感的问题吗？当然可以。如果你需要一个基本的计算能力，这似乎是一个在大型研究实验室内很适合的工作，为什么要创办公司来做这件事呢？这是个好问题，我认为在很多方面，这可能是一个时间的问题。目前的市场处于一个非常特殊的阶段，最大和最成功的实验室无论是偶然还是命运，可能正处于有史以来最激烈的市场竞争中，这使得他们不太愿意尝试不同的路径和替代方案。
▶ 英文原文 ⏱
Reinforcement learning needs a baseline of, of, of ability to only, to only start working. So, so where I am coming from, probably there are many architectures that need a baseline of compute to even start doing anything, anything interesting, anything useful. Can I ask you then maybe a touchy question? Please do. If you need a baseline of compute, that sounds like a job that would be well served inside of a big research lab. Why start a company to go do this? It's a, it's, it's a great question. And it's, I think in many ways, it's likely a timing thing, timing issue market is right now in a, in a very specific place where the biggest and the most successful labs by coincidence or by fate are probably in the most competitive market fight ever right now, which makes them not very keen on trying different paths, trying alternatives.
如果变压器技术是盈利的，并且你能够投入更多的精力和资源来扩展变压器技术以在下个季度取胜，那么很难分出太多注意力和精力去从事那些可能会在一年或两年后重新定义领域的东西。因此，我认为最大的实验室——我基本上和他们都聊过——对尝试变压器技术的替代方案不太感兴趣。而那些不是最大的实验室则尽其所能地模仿最成功的实验室，大家都在尝试训练相同的编码代理。如果看看上周的发布情况，就会发现大家现在都在尝试发布一个编码代理。我觉得我们需要多样化的发展路径和方法，这就是我们在生态系统中试图填补的空白。
▶ 英文原文 ⏱
If transformer is profitable, and if you can spend more efforts and more resources, scaling transformer to win in the next quarter, it's very hard to put at least a lot of attention and a lot of energy to work on something that will, that will maybe. Better on maybe, or maybe we'll, we'll redefine the field in a year or two. So, so I think the biggest, biggest labs, and I talked to basically all of them don't, don't have that much interest in trying the alternatives to, to transformer and the labs that are not the biggest are doing whatever they can to do. What the, what the most successful labs are doing and everyone is trying to train the same, the same coding agent. And if you look at the last week's releases, everyone, everyone is trying to release a coding agent right now. And, and I think we need different paths and different, and different approaches here. So, so that's what the niche in the ecosystem we are trying to, trying to fill in.
罗汉，你曾在发明Transformer的时候就在Brain工作。你同意杰瑞对Transformer的赞词吗？是的，从某种意义上来说，就好像当初Ashish、Noam和其他人提出Transformer的时候。我也在进行在线蒸馏的研究，并且和他们大概在同一时间的内部研究会议上展示了成果。在内部，这并没有引起很大反响，只有少数人真正理解它。很多人只是觉得，哦，这就是又一项研究。当时大家都在寻找应用的方法，而且至少最初的研究非常专注于一个实际问题，那就是翻译。
▶ 英文原文 ⏱
And Rohan, you were at, you were at Brain when the transformer was invented. Do you, do you agree with Jerry's eulogy for the transformer? Yes, in some sense, like once the, first when the transformers, Ashish, Noam and others came up with it. I had like work, I worked on my work on online distillation around the same time we presented it at the same internal research conference. It wasn't a big deal internally. There's only a few people who actually got it. A lot of people were like, oh, that's, it's like, yeah, it's another work. And people were finding ways to, and it was also very focused on, at least the original work was very focused on a real problem, which is translation.
他们就像Beatles乐队一样解决了翻译问题，这是需要开放心态的。我是说，在Google内部，确实有像Noam和其他一些人对扩展语言模型很感兴趣。我认为直到GPT-2和GPT-3出现后，我们才看到了transformers的技术效果显著。至少在我看来，架构设计关注的是如何有效利用计算资源，而transformer就是一种非常高效的计算利用方式。但现在，当我观察行业时，我们在推理时花费了大量计算资源，主要是用于处理tokens。让我问一个问题：如果我想优化一个更好的架构，我想同时关注预训练和强化学习。
▶ 英文原文 ⏱
So they solved like the Beatles TM on translation and it took opening up. I mean, internally at Google, there was definitely like Noam and a few others were definitely interested in scaling language models. I think it is until GPT-2 and GPT-3 that we saw the benefit of transformers working quite well. At least the way I think about architecture is how do we spend computation? And transformer is one way, very efficient way to spend computation. But now that I look at the industry, it's, it's a lot of our computation is inference time and spending it on tokens. Let me ask this question. Like if I want to optimize for a better architecture, I want to look at both pre-training and RL together.
我想寻找能够更有效利用计算能力的架构，而不是像现在这样逐字生成的方式。我认为就像是预训练阶段构建了具有特定上下文长度的变压器（transformer）模型，但强化学习（RL）进来后，发现这还不够，它需要更多的计算能力，于是通过每次添加一个标记来实现。然而，从推理的角度来看，这种每次生成一个标记的方式效率很低。因此，大多数解决方案都在寻找进行推测解码的更好方法，这就像是为了解决我们因只能一次生成一个标记的问题而找的权宜之计。
▶ 英文原文 ⏱
And I would like to find architectures that spend computation much better than current chain of thought token generation in, at a, like to give a much better overview. It's, I think of like pre-training has built the transformer with certain context length and RL comes in and it's like, well, that's not sufficient. I need more computation. Let me do it via adding one token at a time. This is quite inefficient from like inference perspective, you're doing one token at a time. So most of the solutions have been finding to do better ways of speculative decoding. So it's like a bandaid to a problem that we've picked something that's can only generate one token at a time.
因此，在自回归解码方面，transformer在如何有效利用计算资源上存在一些问题。我认为很多人都在训练非常大且密集的模型，而整个行业大约用了两到三年的时间才改进了架构，让我们现在理所当然地认为的事情，当初对许多人来说并不明显。稀疏性、专家混合以及如何通过这些方法提高训练效率是关键。所以，你可能会问transformer的问题在哪里。问题在于如果计算深度不足，我们该如何增加计算深度？仅仅是提出这个问题，就开启了大约二十个新的研究方向，可以让我们修改机制以解决这一问题。
▶ 英文原文 ⏱
So autoregressive decoding, there is problems with the transformer in terms of how do we spend the computation for the longest time. I think most of the world was training very large, dense models and it took like the industry, like roughly two to three years to get to refine the architecture to what we now take for granted was not obvious to a lot of people. Sparsity and mixtures of experts and getting good training efficiencies with them. Right. So then you can ask like, what's wrong with the transformer? Well, it's if the computational depth is poor, how do we increase computational depth? And just posing that question opens up like 20 new directions on how we can modify the mechanism to incorporate it.
所以我明白，做这样的工作是需要时间的。过去，基础研究通常要花五到六年才能进入工业应用。这很大程度上是因为组织知道这很重要。这就是一个赌注。就像Jerry内心坚信强化学习（RL）是必要的一样，而在谷歌，我完全没有那种信念。我是一个预训练的极端主义者，坚持预训练最大的模型。这就是为什么你们很合适。对，就是如此。然后就是那种内心的信念。
▶ 英文原文 ⏱
So I see like to do work like this, it takes time. And usually like fundamental research in the past have taken like five, six years to land into industry. And it's largely from organizational knowing that it is important. This is the bet. Like just like Jerry had the inner belief that RL is needed. Absolutely do not have that belief at Google. I was a pre-training maximalist. Pre-trainer biggest model. That's why you guys are a good fit. Yeah, right. And then so that inner belief.
首先，你需要确保你的架构能够在硬件上高效运行。一个理论上的最佳架构如果无法在实际中应用，对任何人都没有用。这个过程需要从研究的开始到落地生产，将所有的核心和模块编写完成，实现一个端到端的循环。目前只有少数地方有整合的团队在做这些工作。而我认为，我们建立的团队方式是将各领域的专家聚集在一起，而不是分散在不同的部门。这样，我们就能加速整个过程中每个人从整体上看待问题的进程。因此，我对此持乐观态度，这也是我在这里的原因。
▶ 英文原文 ⏱
And second is you need your architecture to run efficiently on hardware. A theoretically optimal architecture is not useful to anyone. It is something when it comes into practice. So you need the research inception to getting it productionized and getting kernels and everything written, the end-to-end loop. And there's only a few places right now which have integrated teams doing that. And I think we have built a team in a way that puts the experts together, not in different silos that like we are accelerating on having everybody look at the problem holistically from end-to-end. So that's like where I'm quite bullish. That's why I'm here.
目前的机制相当差劲。如果我们把问题交给全世界来解决，我担心我们可能需要更长的时间才能替换掉transformer。此外，我认为已经有很多人在抱怨令牌的成本。好像说我们——我们并没有在抱怨。我是说，就像是的，确切地说。我来自Google的思维模式，我们必须服务数十亿的人。因此，我们需要寻找更高效的架构，这些架构既能满足延迟的截止要求，也能在服务的令牌数量上达到平衡。
▶ 英文原文 ⏱
The current mechanisms are quite poor. And if you leave it to the world, I am afraid that it will take us a much longer time horizon before we replace the transformer. And I think a lot of folks are already complaining a lot on token costs. And that seems like as someone-- We're not complaining. I mean, in terms of like, yeah, exactly. I come from the Google mindset where we had to like serve billions of people. So like finding more efficient architectures that fit, have like a deadline on latency and the number of tokens that you can serve.
所以，当我看到这种情况时，能够使用前沿技术的人实在是少之又少。必须有人或某个团体加速改进这一情况，而我们正努力于此。现有技术即使扩大规模，也只适用于一部分人。这就是我们正在做的赌注。我听你说过，变压器的问题是计算深度不足。如果这是问题的关键，能不能告诉我们这意味着什么？为什么会这样？该如何解决？
▶ 英文原文 ⏱
So like when I look at that, like the amount of like the world that can use like frontier tech is very little. And someone or some group has to like accelerate and make this better. And we are taking that shot at doing that. The current technology just scaled up is still only relevant to like a subset of humans. Yeah. And this is the bet we're making. So one of the things I heard you say was the problem with transformers is the computational depth is poor. Yeah. If that's the crux of the issue, tell us what does that mean? Why is that the case? How do you fix it?
我可以给你一个见解。我们训练的大多数变换器实际上是比较浅的，最多只有100层。我们之所以称之为“深度学习”，是因为希望得到更深层的表达。虽然已经有试验在探索增加深度，但还没有人真正证明我们可以学习到极深的表达。通过链式思维推理和让模型自己进行链式思维的强化学习（RL）是一种增加计算深度的方法，因为每增加一个词，就等于增加了一条新的路径。因此，这样可以突破预训练架构所设置的瓶颈。
▶ 英文原文 ⏱
I can give you like one insight. Like most transformers that we train are quite shallow. That's at most like 100 layers deep. Depth is like, it's called deep learning because you wanted deeper representations. There has been experiments on going into depth, but no one has actually shown us learning extremely deep representations. The chain of thought reasoning and RL to do chain of thought by model itself is one way to increase computational depth because every token you add, you add like one more pathway. So then you can get out of like this bottleneck that the pre-trained architecture has set you up on.
您只能做特定数量的层次和序列长度的组合。现在，您可以增加序列长度，从而获得更强的结果。您可以进行推理时间的扩展。然而，推理时间扩展的问题在于模型现在必须生成更多的标记才能获得更好的结果，并且这是一次生成一个标记。由此可见，您可以直接解决其中的许多问题。这只是我们正在研究的一部分工作，目的是让这一过程更高效。
▶ 英文原文 ⏱
You can only do a number of layers times sequence length. Now you can increase the sequence length and you get much stronger results. You can do inference time scaling. Now the issue with inference time scaling is that models now have to produce more tokens to get better results and that's very one token at a time. And from this, you can see like you can directly address many of these things. And this is like a subset of work that we are looking at, right, making this much more efficient.
你对基于Transformer的架构有何展望？如果这不是最终形态，它还能带我们走多远？我们什么时候会看到它的极限呢？我认为这一切都回归到我们为什么要训练Transformer，以及我们能用它们做些什么。我们正在进行预训练，这非常擅长将互联网中的所有知识提炼到Transformers中。然后我们可以通过强化学习（RL）对它们进行训练，也就是说，我们基本上可以将所有我们想要实现的工作流程融入到一个Transformer中。
▶ 英文原文 ⏱
What's your forecast for the transformer based architecture? If it's not the end state, how far can it get us? When do we start to see it topping out? I think it all comes back to what we are training transformers for and what we can do with them. We're doing pre-training, which is very good at distilling all the knowledge from the internet into transformers. And then we can RL them through, which is we basically can bake all the workflows that we want into a transformer.
翻译如下：
所以说，Transformer 模型已经达到了一个极限，它将人类所有的知识和这些知识之间的关系整合在一起，还包括它们如何协同工作，以及如何组合在一起。基本上，任何我们有训练数据的任务都可以放入这个模型中。这可能是一个庞大的模型，用大量计算资源训练，使用的是世界上的所有数据。那么如果我们停止对这个模型的训练，会发生什么呢？这是一个值得常常思考的问题，我一直在思考 Transformer。
▶ 英文原文 ⏱
So what transformer is like capped out is we have all the knowledge of humanity in the model together with the relationships and how do they work together, how they can be combined. And basically any task that we have training data for, we can put into the model. And this can be gigantic model, trained with a lot of compute on all the data in the world. And then if we ever stop training that model, what would happen? The question worth asking often, I'm thinking about transformer.
如果OpenAI和Anthropic停止训练新的模型，并且我们只保留我们今天拥有的变压器模型，并宣称这就是最好的模型，那会发生什么呢？几个月过去了，几年过去了，这个模型变得越来越没用。假设实验室真的记录了地球上每个人的生活，他们在做什么，他们的任务以及所处的环境，并将这些信息放入模型中，让其在一个学习环境中学习。但如果这些信息发生了变化会怎样呢？如果世界上出现新的事件，如果这些新事件之间有新的关系，如果出现新的任务类型，如果出现新的代码库，或者新的工具，该怎么办呢？
▶ 英文原文 ⏱
What would happen if OpenAI and Anthropic stopped training new models and we got the transformer we have today and say, this is it. This is the best model we have. Months pass, years pass, and the model is getting less and less useful. Maybe the lab really recorded every human on earth, what they were doing and what their tasks were and their environments and put them in the model, put them in a learning environment. But then what happens if anything of that changes? If there are new events in the world, if those new events have new relationships between them, if there are new types of tasks, if there are new code bases, new tools to use.
变形模型的价值和实用性很大程度上取决于训练中包含的有价值的信息。当这些信息缺失时，模型的表现就会受到影响。虽然它们有一定的适应能力，但并不强大和灵活。在我看来，这大概是变形模型的极限，这也是为什么我认为它们是一种工具。假如我们有一个人清楚地知道变形模型的局限性，那么他们可以更好地管理和使用模型。他们可以设定任务的提示进行训练，并通过这样的训练让模型在这些特定任务上表现得非常出色。对于任何模型无法完成的任务，你都可以通过增加训练数据来提升模型的表现。然而，这一过程需要训练模型的实验室来执行。
▶ 英文原文 ⏱
Transformers are getting a lot of their usefulness and value through the things that are valuable have to be present in training. And when they are not, they suffer. There's some ability to adapt, but it's not very big and not very flexible. So in my mind, this is kind of the level where the transformers top, which in many ways, what I think is a tool to use for us. If we kind of know, if there's a human who knows the limitations of a transformer, they can schedule that model. They can write a prompt of what is the task that you want. And by doing the training we are doing, you can get very successful at that. And any task the model fails, you can add it through training data and the model can succeed. But that loop has to go for the lab training the model for you.
如果模型在实验室中需要进行基础训练，你觉得在多大程度上这是目标，还是你希望能以某种方式更新模型，而不必回到实验室？你有没有读过Rich Sutton和David Silver的那篇论文《经验的时代》？我很好奇你在多大程度上同意他们的观点，或者是否有不同的看法，哪些地方你的观点不同？强化学习并不是一个特别新的方法或概念。所以在某种程度上，我认为“经验的时代”一直都存在。而且人们对预训练有一些批评，因为预训练显然是另一种看待模型的方式，就是我们使用静态数据，而这些数据大多是别人生成的。
▶ 英文原文 ⏱
And if the model that fundamentally needs to be trained in the lab, how much do you think of it, that this is the goal or you would want to be able to update the model somehow, not having to go back there? Have you read the Rich Sutton and David Silver have this paper, The Age of Experience? Have you read it? I'm curious how much you agree or if you have any different opinions, where your opinions diverge? Reinforcement learning is not a particularly new approach, particularly new thing to do. So in some way, Age of Experience, I think always has been there. And people have been criticizing a bit pre-training because pre-training very clearly is the other way of looking at the models, which is like we have static data, that data is mostly generated by others.
虽然我个人认为，今天的预训练很大程度上是在将其他模型的内容提炼并融入到新的模型中，因为互联网的大多数内容都来自人工智能。但是显然，预训练的过程包括行为克隆、模仿以及对互联网数据的压缩。强化学习并不是一个新概念，人们早就开始思考和实践。过去，强化学习被用来解决游戏问题，如骰子游戏Backgammon，围棋，星际争霸和Dota，现在也被用来解决编程问题。每次应用时，重点都是让模型撰写自身的经验并从中学习和成长。
▶ 英文原文 ⏱
Although I have this personal view that pre-training today is largely distilling other models, other models into the new model, because most of the tokens in the internet are coming from AI. But there is clearly pre-training, which is behavioral cloning, which is mimicry, which is compression of internet data. But reinforcement learning is not something that people haven't been thinking and people haven't been doing. Reinforcement learning was used to solve by Gammon back in the day that it used to solve Go, StarCraft, Dota and to solving programming right now. And every time it comes down to model writing its own experience and learning, learning from that experience.
但是这点是非常清楚的。而且我认为有趣并让人困惑的是，强化学习并不是从经验中学习的唯一方式。我认为在这方面将有更多的发展，可以称之为算法上的创新，但本质上是我们从经验中学习方式的创新。强化学习只是其中一种方式，是一种数学上的表达方式。尤其是现在，我们使用它的方法，比如并行展开以减少方差，以及在不同版本的模型中进行比较，这与我们真实的经验学习方式并不相同。
▶ 英文原文 ⏱
But it is very clear. And what I think is interesting and what I think is still perplexing to people, that reinforcement learning is not really the only way to learn from experience. And there will be more and there will be a little bit more of, I think, you can call it algorithmic, but essentially innovation of how we learn from experience. Just because reinforcement learning is only one way to do it, it's a mathematical formulation. And especially right now, how we are using it, it really like those parallel rollouts for variance reduction and for comparing how the model does in parallel versions of the world, which is not how we do it, not how we learn from experience.
我们从经验中学习得更高效，也学到更多。我们在许多方面运用这些经验。有段时间，我一直试图向人们解释大脑的运作和学习方式，不是大脑中只有一种学习算法。我认为实际上有多种算法，它们共同发挥作用。但如果我踢足球，比如说，这看起来很像强化学习。我踢球很多次，每次都稍微调整一下，看看是否大致符合我的期望，并且有一些自我强化在发生。当我学习数学时，那是一种截然不同的体验。这更像是阅读一些困难的概念，并在我的脑海中深入思考，直到突然领悟并将它们连接起来。
▶ 英文原文 ⏱
We learn from our experience much more efficiently and much more. We use those in many, many, many ways. At some moment, I've been trying to explain to people that what brain does, how we learn, it's not that there's one learning algorithm in a brain. I think there are multiple, actually, and they work together. But if I play football, for example, it looks very, very close to reinforcement learning. I kick a ball a lot of times and every time I adjust it a little bit and I see if it roughly matches what I wanted and there's some self-reinforcement happening. When I learn mathematics, it's a very different type of thing. It's like reading about hard concepts and thinking about them very deeply inside my head until things click and until I have them connected.
两者都在某种程度上是从经验中学习的，但它们的方式非常不同。总结一下我对从经验中学习的思考：我们已经这样做了一段时间。我们可能比以往任何时候都更多地使用计算机来进行从经验中学习，但强化学习并不是从经验中学习的终点。在未来几年内，研究人员将会提出更好的方法来在任何未来环境中使用这些数据。很有趣。Rohan，我很好奇，因为你的很多工作都围绕在优化和效率上，我们怎样才能实现数量级上更高效的计算和数据使用的学习算法呢？
▶ 英文原文 ⏱
And both of those in some way are learning from experience, they are just very different. So summarizing my thinking of the learning from experience is that we've been doing it for a while. We probably are spending the most computer than ever on learning from experience, but reinforcement learning is not the end of learning from experience. And there will be better approaches that researchers will be coming up in the coming years on how to use that data in any future settings. Interesting. Rohan, I'm curious, since a lot of your work has been around optimization and efficiency, how do we get to a orders of magnitude more efficient, I guess, more compute efficient and more data efficient learning algorithm?
我会从测量开始。我认为我们现在定义的预训练主要是关于压缩。我们关注困惑度（perplexity），然后测量我们如何降低困惑度。接着我们发现，通过扩大规模、增加参数数量和投入更多计算能力是可行的方法。每次我们将计算能力按对数规模增加时，我们在这些指标上都会取得一点点epsilon的改进。我觉得，这样做对建立一个先验模型是可以的，但我认为这样看待问题是错误的。我们应该看到整体过程。我们训练这些模型到底是为了什么？应该关注最终结果。比如，我训练了这个模型并交给Jerry。Jerry会进行强化学习（RL），并彻底摧毁我所有的困惑度指标。也就是说，这是我们目前为止解决这个问题的最佳方式。
▶ 英文原文 ⏱
I'd start with measurement. I think pre-training as we define it right now is about compression. We look at perplexity and then measure how do we decrease the perplexity. And then we find that scaling and increasing parameter count and putting more compute is the way. And every time we increase compute in log scale, we get this epsilon more improvement in these metrics. This is, I think this is fine for building the prior, but I think this is the wrong way to look at the problem. We should be looking at the end to end. What are we training these models for? Look at the outcome. Like, for example, I trained this model and give it to Jerry. Jerry will do RL and destroy all the perplexity metrics that I have created, right? Like, so then it's sort of like it was the best way we had so far to attempt to solve the problem.
我认为实验室和其他人都做得非常出色，他们提供的情报价值极高，这让我的工作充满乐趣。但要达到这一点，我们经历了启动的过程。我们必须将预训练和强化学习结合起来。这就像一个训练过程，可以说是一种学习算法。从优化的角度来看，我的故事是，我在2016年开始在谷歌研究用于逻辑回归的优化。当时被这个问题吸引，专注于开发一些用于 Sibyl 的解算器。Sibyl 是谷歌在深度学习兴起之前使用的大规模线性解算器，后来被神经网络取代了。
▶ 英文原文 ⏱
And I think the labs and everyone else have done a great job in producing intelligence that's super valuable and makes my work so much fun. But it was the bootstrap process to get there. We have to combine pre-training and RL together. And that's like where one order of magnitude improvement would come from. And that's like a training procedure. You can say it's a learning algorithm. In terms of optimization, my story is I started optimization at Google for logistic regression back in 2016. And got nerd sniped by it, worked on some solvers for what we used to call Sibyl, which was the large scale linear solver that was used at Google before neural network took off and then replaced this.
然后我问我自己，我想在神经网络方面做些什么工作呢？答案很明确，我想理解训练算法并改进它。然后有一天，有个人，名叫Vinit Gupta，就出现在我的办公桌前。他说，我听说你在优化方法方面很有一套。我们有一个点子，在白板上讨论出了一种名为Shampoo的算法。你能帮助我们实现并确保它能在大规模的神经网络训练中正常工作吗？于是我开始研究这个项目。我的经理Yonghui Wu一直支持我，直到2024年我任期结束。
▶ 英文原文 ⏱
Then I asked myself, like, what do I want to work on with neural network? And it was quite clear. Like, I want to understand the training algorithm and make it better. And then someone, Vinit Gupta, just showed up one day at my desk. It's like, I heard you're really good at writing optimization methods. We have this idea that, you know, like we worked out on a whiteboard and what turned out to be the shampoo algorithm. Can you help us implement scale, make sure it works at large scale for neural network training? So I was working on this. I think my manager, Yonghui Wu, who's like supported it throughout that to like till my end of my tenure in 2024.
但是，像社区和大多数其他人一样，对这个想法并不是那么感兴趣。而对我来说，这却是最令人兴奋的事情，因为我觉得我正在加入计算并让训练变得更好。这是我必须解决的问题。我愿意花费我所需的时间来完成它。但是人们总是假设，哦，为什么，像，什么是上限？你仍然可以使用Adam，这没问题。为什么我们要把所有时间花在其他事情上，而不是优化上呢？但从某种意义上说，优化就像你有一个模型，你在优化它，你想要更好地优化它。
▶ 英文原文 ⏱
But largely like the community and most of the other people were not as excited by this idea. And for me, this is the most exciting thing because I was like, I'm putting in computation and making training better. This is the thing I have to figure out. I will spend as much time I would take to do it. And then people were making this assumption, oh, why, like, what's the upper bond? You could still use Adam. That's fine. Like, why are we, you could spend all the time on everything else, not optimization. But in some sense, optimization is like, like you have a model, you're optimizing it. You want to optimize it better.
现在，让我们联系一下之前讨论过的一些关于架构的内容。我们在架构方面做了很多工作，其目的是为了让这些网络可以进行训练。从某种意义上说，这就像硬币的两面：优化和架构是相辅相成的。你可以使用更强的优化器来训练一个更难优化的模型，从而获得更好的性能。或者，你可以在更容易优化的模型上使用较弱的优化器，依然能得到不错的性能。 因此，这中间存在各种权衡。
▶ 英文原文 ⏱
Now connected to some of the stuff that we talked about architecture. What has happened is that a lot of the work that we've done in architecture is to make these networks train. And in some sense, it's like two sides of this coin and optimization and architecture go together. You could have a stronger optimizer, train a much more harder to optimize model and get better performance. Or you can use a weaker optimizer on easier to optimize models and get decent performance. So there are these trade-offs that appear all over.
对于我来说，我花了很多时间在这上面。我记得我们把它用于Gemini 1.5 Flash项目。后来，社区开始对此更加感兴趣。有一篇关于这个主题的文章发表了，我们还积累了一整套关于“洗发水”的文献，也就是你在洗澡时会用到的那些东西。然后事情变得很清楚，似乎这对当时的情况大约有两倍的提升。但即便如此，如果你仔细看洗发水，它在实际效果上还是比较薄弱。
▶ 英文原文 ⏱
And for me, I spend a lot of time working on it. I think we used it for Gemini 1.5 Flash. And then like the community started getting like more interested in it. There was the soap paper published. We have like an entire literature of like shampoo. So all like bath time, the things that you would use. And then it was quite clear. Like, so that was like maybe a 2x improvement over what was happening. But even then, if you look at shampoo, it's quite weak in what it's doing.
当你进行训练时，如果没有利用所有可用的信息，那么效果可能并不理想。随着你在训练中使用越来越多的信息，你就能取得更好的进展。从某种意义上说，你的优化算法决定了你会发现什么样的架构。我有一些同事，他们的观点在文献中不太流行，可能全球只有四个人对此感兴趣，但这些想法非常有趣。比如，残差连接在训练神经网络时就非常有用。
▶ 英文原文 ⏱
It's not using all the information that's available to you when you train. And as you use more and more information as part of training, you can get better improvement. And in some sense, like your optimization algorithm defines what architectures you will discover. Like I have colleagues. It's not very popular in the literature. It's only like maybe like four people in the world care about it. Kind of ideas that are extremely interesting. Like residual connections have been extremely useful for training neural networks.
有些人现在已经摆脱了这些问题，并学会了更深层次的表示。但他们需要一种更好的优化方法。对我来说，优化方法和你问的问题很像，就是我们该如何到达那个目标。这需要结合架构，并从头到尾考虑整个问题。这是许多计算效率所在的地方。我也认为强化学习（RL）花费了大量计算资源，但效率不高，因为你得到的反馈很少。
▶ 英文原文 ⏱
There are folks who have now gotten rid of them and learned deeper representations. But they needed a better optimization method. So like for me, optimization methods and the question you asked is just like, how do we get there? It's combined with architecture and thinking about the problem end to end. It was where a lot of the computational efficiency is in. Yeah. And I also see RL as spending a lot of compute, not as efficiently. And so like if you could spend it because you don't get much feedback.
你正在花费更多的时间去竞争，因为你需要解码这一长串的想法链，只为了把这一点信息传递到网络中。这似乎是一个非常高效且容易实现大幅提升的目标。我可以整天谈论优化问题，我非常热爱这一领域。你觉得我们会不会接近或者超越生物学习的效率呢？我认为不太可能，因为这需要我们做出改变。这可能是个比较强烈的说法。至少以我们现在拥有的硬件来看，这似乎非常不可能。我们人类像Jeff Hinden所说的那样，进行模型计算。我们在成长过程中创建了自己的电路，并用现有的硬件建立了自己的学习算法。然后我们去世，所有的一切都随之消失。
▶ 英文原文 ⏱
And you're spending a lot more compete because you have to decode all this long chain of thought to get this one bit of information into the network. Seems quite an efficient and easy target to get orders of magnitude on top of. I can go on talking about optimization all day. I love it. Do you think we'll ever approach or surpass biological learning efficiency? I do not think so because I think we would need to change. Maybe like that was a strong statement. At least with the hardware we have, it seems pretty unlikely. Our biological, like we have something as Jeff Hinden says, model computation. So we built our own circuit as we grow up and we built our own learning algorithm with the hardware. And then we die and then we're gone.
神经网络非常不同。我们的硬件不变，神经网络也不变。但是，它的学习效率很低，你需要大量的神经网络和大量的并行运算来获取少量信息。因此，我认为在我们设计出更类似于人类运作方式的硬件之前，这个过程会更困难。也许应该更偏向模拟技术，搞清楚如何处理模拟电路和纠错，以及如何传递信息。我觉得在这方面我们还是比较安全的。用“安全”来形容这点很有意思。
▶ 英文原文 ⏱
Neural networks are quite different. Our hardware stays, the neural network stays. But it's learning very inefficiently and you need a lot more of them. And a lot of parallelism to get small amounts of information through. So until I think we design hardware to be much more like how humans operate. Maybe more analog. Figure out how to deal with analog circuits and figure out how to do with error correction. Figure out how to get information through. It'll be much harder. I think we are safe. Safe. That's an interesting way to put it.
将预训练和强化学习（RL）进行端到端优化这一想法似乎是一个明确甚至显而易见的陈述。你认为研究机构意识到了这一点，却因为难以摆脱组织架构和流程的限制而难以实现这一目标吗？或者是什么阻碍了他们将二者统一起来？我认为这并不是那么显而易见的。原因在于，这实际上是一个完全不同的优化问题。你有一个先验知识，你在进行模型的预测和更新，这导致了更高的方差。而预训练则需要更大的批量处理和更强的并行计算能力，能花费更多的时间单位来处理。因此，在你深入思考为什么简单的结合不能奏效之前，人们很难自然地将这两种训练模式结合起来。
▶ 英文原文 ⏱
The idea that pre-training and RL should be optimized end-to-end seems like such a clear, maybe obvious statement. Do you think the labs realize this and is it just hard for them to get rid of org charts and process to be able to make that come together? Or what stops the labs from being able to unify the two? I don't think it's as obvious. Okay. Because it's a completely, again, different optimization problem. You have a prior. You're doing roll-outs. You have higher variance. And then pre-training is a much larger batch. More parallelism or compute for the unit of time that you can spend. So it is not an obvious thing for folks to combine these two training procedures until you think a bit more like, why is it that the naive combination doesn't work?
所以这是第一个。第二个是，如果我询问一些这些实验室中最优秀的研究人员，他们可能会说，哦，这有道理，我们应该去探索它。但是这可能不会被放在最优先的位置，因为他们之前必须为下一个周期训练模型。就像Jerry说的，现在有公司在竞争发布周期，因为token（令牌）不是粘性的。因此，在这种环境中，在许多实验室开展长达六个月的长期研究要困难得多。
▶ 英文原文 ⏱
So that's one. The second one is if I poll some of the best researchers in these labs, they would say, oh, this makes sense. We should probably explore it. But it would be probably not in the top bucket because they have to train a model for the next cycle before. It's like, as Jerry said, there are companies now competing for release cycles because tokens are not sticky. So it's much harder to do long-term research of six months in many of these labs in the environment.
所以，看来核心自动化的基本前提之一是，你们在实验室工作时，正值 AI 科学家这一概念逐步成形。Sam 和 Dario 都在讨论 AI 科学家这个概念。看起来，作为研究人员，你们的工作已经发生了根本性的变化。你们可以在这个新时代创立公司。因此，你们可能能够进行比以前更多的实验。那么，你认为研究工作有多大程度是可以自动化的？你们是如何着手打造你们的实验室，以实现（据我所知是你们的使命之一）成为最自动化的实验室的？
▶ 英文原文 ⏱
So it seems like one of the core premises for core automation is, you know, you're starting at the lab at a time when, I know Sam's been talking about the AI scientist. I think Dario's been talking about the AI scientist. It seems like your job as researchers has actually fundamentally changed. And you get to start the company native to that era. And as a result, you may be able to run a lot more experiments than otherwise might be possible. How automatable do you think the research job is? And how are you guys approaching building your lab to be as, I believe your mission, one of your missions is to be the most autonomous lab there is?
世界上最自动化的实验室。首先，我认为核心自动化的版本是关于以某种方式给予每个人最大的主动性。这并不是要把人类完全排除在外，也就是一种自动化的方式，而是要让人们在有限的时间内做到更多。就像你走路的时候，可以走一段距离；但是如果你骑自行车，可以走得更远；而如果你开车，就可以走得更远更快。
▶ 英文原文 ⏱
The most automated lab in the world. And to start with, I think the automation, the version of automation by core automation is about giving each human maximum level of agency in some way. It is, we are not trying to really get humans out of the loop, which is like one version to automate, but it is about give humans ability to do the most with their amount of time. Whenever you are walking, you can get some distance. Whenever you get a bike, you can go walk a larger distance. Whenever you are a car, you can go much, much, much, much larger.
当人类开始务农时，他们只能徒手耕作，并在一小块地上劳动。而有了机器之后，人们可以在大得多的土地上耕作。我个人非常喜爱当前的编程助手，我们对此非常满意。在某种程度上，这就是我多年来一直努力的方向，既进行编程研究，也在OpenAI内侧参与各种版本的AI科学家的开发。
▶ 英文原文 ⏱
Whenever humans started farming, they had to farm by hand and work on a small plot of land. When you have a machine, you work on a much, much larger plot of land. And personally, I am both really great fan of the current coding agents and we're happy. In some way, it is what I've been working for many years, both doing coding research and working on various versions of AI scientists inside the side of OpenAI.
最终，我意识到，为了实现那个愿景，创办一家公司是其中一种最佳的方式之一。因为如今进行研究的方式与过去大不相同，单个研究人员现在可以做得更多。归根结底，迭代的速度、研究的速度、处理想法的速度以及获取想法数据的速度都与以往有很大的不同。
▶ 英文原文 ⏱
And in the end, I realized starting a company to realize that vision is one of the best, one of the best ways to realize it because the way you can do research today is very, very different because a single researcher can do much more. In the end, the speed of iteration, the speed of research, the speed of how quickly you can move through ideas and how quickly you can get data on your ideas is something very, very different.
你可以尝试调整现有的结构和团队的工作流程，例如如何收集数据。或者你可以尝试原生地为此构建，正如你所说，可以为流程量身打造，以最大限度地赋能每位研究人员，使他们能够更快地开展他们的想法。所以我们正在这里尝试重建整个深度学习栈，并思考如何以不同的方式进行几乎每一个操作，我们有多种选择。如果我们每天至少能执行其中的一个实验，那就已经比以前的迭代速度快很多了。而这并没有什么物理定律上的根本限制。有一天我们可能会达到每天执行10个实验，也有可能达到每天执行100个实验。
▶ 英文原文 ⏱
And you can try to move the old structures around it and the team's workflows, how data is gathered. Or you can try to build, like you said, you can try to build natively for it, for processes that maximally empower each researcher and allow them to just iterate on their idea much, much quicker. So we are here and we are trying to rebuild the full deep learning stack and try to think how we can do almost each operation differently, what are various options. And if we can execute at least even one of those experiments a day, that's already a pretty good iteration speed versus anything that was done before. And there isn't really like any fundamental laws of physics reason why not. And maybe one day we get to 10 of those a day, maybe one day we get to 100 of those a day.
从根本上来说，对于这个搜索过程和优化过程，我们应该能够在更好的深度学习环境中找到有效的方法。我们团队几乎每个人都充满了主动性和自动化意识。我们正在进行一个实验：测试我们能够将这些能力发挥到多大程度，以及在一个小团队的情况下能做到多少。当我们努力做到最好时，我们可以走多远？何时我们才会知道达到了人工通用智能（AGI）？以前我常说，每个人心中都有自己的标准，不同的人对AGI的定义各不相同。
▶ 英文原文 ⏱
And fundamentally for that search process, optimization process, we should be able to just find things that work in a better deep learning setting. And what we are trying to do is like within almost all of us, we are a team that is very agent-filled and automation-filled. We are trying to do an experiment. Like how far we can push those things and how much an organization that tries to do as much as we can with a small team, how far we can get with that. When will we know that we've reached AGI? At some moment, I used to say it's very much in everyone's heart, whatever they consider AGI.
OpenAI 是一个在经济价值工作中能超过所有人类表现的系统。但回到我之前的论点，如果 OpenAI 停止训练模型会怎样？它还能继续工作吗？它还能保持同样的自动化水平，还是会有所推动？对我来说，AGI 是一个能够在任何情况下自我完善、无需人工参与的模型。我认为，那时我们才能有意义地讨论 AGI，因为这在某种程度上是对前一个定义的补充。因为改进 AI 模型实际上是一项人类可以完成的工作，属于经济上有价值的工作，这点无疑成立。
▶ 英文原文 ⏱
OpenAI is the system that can outperform all humans in economically valuable work. But it goes to my previous statement, what if OpenAI stops training models? Would that still keep working? Would that still keep the automation level the same or would that drive? And for me, AGI is a model that can improve itself without human in the loop in any way. That's I think the moment where we can meaningfully talk about AGI because that is in some way, it is a sub definition of the previous one because improving AI models is actually a job that humans can do. It is economically valuable work and it definitely is the case.
但要通过模型将人类从决策过程中移除，目前依然非常困难。我们还没有接近实现这一点。让我很难找到任何任务是大型语言模型（LLM）能够完全不需要人类参与就能完成的。现在，人类和LLM的结合非常成功，但单靠LLM则效果不佳，甚至根本无法实现。我在2024年和2025年初看到的情况表明，当前的路径并不能达到我们的目标。我认为需要进行非常深入的研究，无论是这家公司还是其他公司，都需要探索如何让我们的模型在测试时进行更深入的学习和适应。
▶ 英文原文 ⏱
But removing humans from loops with models has been actually notoriously difficult so far. We haven't come anywhere close to it. It's very, very hard for me to find any task where LLMs were able to get humans out in the loop. We are the human LLM hybrid is really, really successful right now. But LLMs without humans, not so much, not at all. And what I have seen there in 2024 and in early 2025 is that the current path doesn't get us there. And I think we need some pretty serious research on this company or some other to try to unlock, how do we make our models learn and adapt at a test time on a deeper level than we've been so far.
我觉得我们在整个对话中已经暗示过这个问题了，这有点像那种五个被蒙住眼睛的人试图找大象的情境。关于Core Auto的宏伟计划，您能透露些什么呢？从某种意义上说，我们可以分享我们的六个月路线图，比如构建架构。正如我所说的，不仅仅是架构的优劣，它是否运行良好？我们是否能吸引用户，包括我们实验室的成员自己来使用它，对吧？所以直接来说，我们现在可以做很多事情，但我们想要自动化解决的困难问题是内核生成。
▶ 英文原文 ⏱
I feel like we've alluded to this throughout the conversation, but it's kind of been one of these, you know, five blinded, blindfolded people trying to find the elephant things. What is the grand master plan for Core Auto that you're willing to share? We can share our six month roadmap in some sense, like building architectures. As I said, like wouldn't, it's not just how good the architectures, does it run well? And can we get users, including ourselves as part of the lab to use it, right? So then that's directly like we can do many things now, but the thing that is going to be difficult and we want to automate away is kernel generation.
我们有一套硬件，包括GPU和Blackwells，用来进行训练和推理。我们将构建最优秀的模型，旨在缩短从产生一个非常出色的想法（可以消除架构中的瓶颈）到在GPU上以最高T-flops运行的时间。在某种程度上，目前的代码代理加上人类的努力可以发挥很大的作用。这方面的一个例子是我们与GPU模式合作举办的QR内核竞赛。这个竞赛是为了运行一个非常古老的线性代数操作QR，它用于优化，比如shampoo工作中使用了它，许多其他地方也在使用。
▶ 英文原文 ⏱
So we have a set of hardware, GPUs, blackwells that we have to train and run inference on. We will build the best model that we can to basically reduce the time from having a very cool idea that can make these bottlenecks go away in the architecture to having them run at the highest T-flops on GPUs. And in some sense, like current coding agents plus humans can go a long way. But like an example of this is our QR kernel competition that we hosted with GPU mode. It's for running this very old linear algebra operation QR. It's used for optimization, like shampoo line of work uses it, many other places uses it.
您希望在B200节点上高效运行这个程序。如果您使用CoSolver来处理我们关心的形状，可以提高一些效率。而结合人工干预和搜索循环，效率能提高大约7倍。但这需要一个高水平的人才，也许全球只有三个人能够胜任，还需要花费约10万美元，让这些程序员在四周内达到提升60倍速度的解决方案。如此一来，这些模型目前还远远达不到那种提升60倍的高效水平，确实存在瓶颈。
▶ 英文原文 ⏱
And you want to run this efficiently on a B200 node. If you use CoSolver for the shapes that we care about, you get some efficiency. And then a human plus some search loop can get you like something like 7x. But it requires the high taste human, like there's maybe three people in the world, to spend about $100,000 on these coding agents over a span of four weeks to get to a solution that's 60x faster. So these models today are nowhere close to getting that 60x faster kernel. And there is a real bottleneck.
现在，这是一个单一的问题。可能有三个不同的操作员来操作这个面板，执行矩阵乘法，再将其折叠回去，并重复这个过程。这就是矩阵的安全分解过程。如果你把这个问题交给熵模型、OpenAI模型、Gemini，它们是无法解决的。我们的模型甚至连接近解决这个问题都谈不上。所以对我们来说，这就像是我们一直在讨论的一个话题，是我们正在努力接近的一种状态，因为这是实现更高效架构的核心步骤。
▶ 英文原文 ⏱
Now, that was a single problem. It has like perhaps three different operators work on this panel, do this matrix multiply, fold it back in and do this repeatedly. That's what the secure factorization of a matrix would look like. And if you give this problem to entropic models, OpenAI models, Gemini, it just wouldn't solve it. It just, it's not, our models are not even close to solving this problem. So for us, it's like something that we've talked about, something that we're getting close to as sort of like getting to that point because that's our inner loop to having more efficient architectures.
为什么要用内核？是因为它能最大化计算性能的智力输出吗？在某种程度上，我有过三个项目，其中两个已经在业界落地。首先是次级方法。内核是一个瓶颈，因为你需要高效运行它。如果你在像谷歌这样的大公司工作，你不能花费十倍的计算资源只获得两倍的效果。因此，我只能花费大约20%的预算来获得两倍的提升。这样所有人都开心。我不喜欢交易。因此，我认为这就是市场的运作，你投入的要少于得到的。
▶ 英文原文 ⏱
- Why kernels? Is it just because like maximize intelligence per flop of compute? - In some sense, I've had like three projects. Two of them have kind of landed in the industry. So first is secondary methods. Kernels was a bottleneck because you have to run it well. If you were at a place like Google, you cannot spend 10x amount of compute and get a 2x win. So I could only spend maybe a budget of 20% and get the 2x win. Everyone's happy. I don't like trade. So like, I think that's the market, right? You spend less than you get.
所以，内核成为了一个瓶颈，因为大多数操作都是新的，我们没有很多人去研究它。在谷歌，只有两个人能够编写这段代码，分别是 Rasmus 和 Peter Hawkins。因为这需要编写深入的 XLA 和 LLO 代码来实现这一功能。他们花了两年的时间来完成这些工作。那时候，我和其中一位同事有另一个想法，就是用额外的内存来替换 Transformer 中的一些参数。
▶ 英文原文 ⏱
So kernels ended up being a bottleneck there because most of the operations were novel that we haven't gotten a lot of people to look at it. There's only like two humans at Google who could write it, Rasmus and Peter Hawkins, because it was deep XLA, LLO code that you have to write to make this work. And then that took them two years to do. The other idea I had with one of my co-workers at that time was like replacing some of the parameters in a transformer with extra memory.
我们称之为n-grammer，即n-gram记忆。我们在2020年开始研究这个项目，并在内部部署了几个版本，但不是大型版本，而是小型版本。为了训练，我们需要一种可以加速稀疏数据汇集和分散的工具。这需要硬件改变并利用新的硬件。然而，这样的硬件从未实现。我曾与TPU团队以及其他一些人安排了会议，在COVID期间我们一直在讨论这件事，想着它会实现。但最终，它从未实现。
▶ 英文原文 ⏱
And we called it n-grammer, n-gram memory. We worked on it in 2020. We had versions of it internally deployed, not the big version, but the smaller version. But there, I needed like something that can accelerate sparse gathers and scatters as part of training. It required a hardware change and making use of the hardware. It never arrived. I had conferences set up with the TPU team, us and a bunch of others. We were talking about it during COVID, like, oh, we're going to have it happen. And it never arrived.
我在Anthropic也使用过TPUs。在我离开的时候，才刚刚开始了解到如何使用它们。不过大约在六个月前，DeepSeq研发了他们的n-gram，这种技术是通过增加内存的改进版本，展示了扩展损失的效果。这表明，其实你不需要MOE（专家混合模型），可以用这些n-gram嵌入来替代。对我来说，这仿佛是个五年的项目，我为他们感到非常高兴。
▶ 英文原文 ⏱
I also was using TPUs at Anthropic. While I was leaving, just barely started the surface of being able to do it. But at the same time, six months before that, DeepSeq wrote their n-gram, which is an improved version of adding more memory, showed scaling loss that, yeah, you don't need MOEs. You could have actually replaced it with these n-gram embeddings. For me, that was like, oh, it was like a five-year thing. And I was very happy for them.
要探索的状态空间中有很大一部分是无法实现的，除非你在编写内核程序。内核程序。在编写内核程序时，你需要一些帮助或解决该内核问题，以获得最高性能。而且性能的上限非常高。这就像QR算法。如果我使用Qsolver的QR，性能是一个情况；如果我使用竞赛获胜者的QR，性能就会快60倍。这是完全不同的一个层面。
▶ 英文原文 ⏱
So much of the state space to explore isn't even possible if you're not writing kernels. Kernels. And you need to be assisted in writing kernels or solve that kernel to have the highest performance. And the roofline is pretty high. So it's like the QR. If I use Qsolver's QR, I get some performance. If I use the competition winner's QR, you get 60x faster. And that is a completely different playing field.
这为你打开了一整套全新的可应用算法。在训练transformers、训练优化器方面，QR对特征值分解和其他许多分析非常基础。但从某种角度来说，拥有这种技能的人并不多，而且他们通常分布在各地——这里一个，那里一个。如果模型本身具备这些能力，那将是非常理想的。
▶ 英文原文 ⏱
Now it opens up an entire new set of algorithms you can apply. And in terms of training transformers, training optimizers, QR is so fundamental in analyzing the for eigen decomposition and many other things. So it is a thing that I think also, if you think about it, only few people have the skill set too. And they're very much not at the same place. It's like one person here, one person there. And it would be ideal if models had those abilities.
好的。也许我会简单总结一下，从更高层次来谈一下我们想要达到的目标。核心自动化实验室旨在构建可以持续学习并从实际应用中自我改进的模型。正如我提到的，我们认为transformer模型无法进行持续学习，也就是说，无法在transformer上实现持续学习功能。因此，我们意识到必须寻找一种不同的架构。我们的目标是找到这种新架构，寻找可以替代transformer的方法。
▶ 英文原文 ⏱
Yeah. Maybe I'll summarize a little bit and talk from the high level of what we want. Core automation is a lab created to build models that continuously learn and learn from deployment. We believe, as I mentioned, that transformers are incapable of continual learning. There's no way how to put continual learning on transformers. So we know we have to find a different architecture. Some of anyway, our quest is to find that new architecture, find that transformer replacement.
我们希望建造一个高度自动化的实验室。我们想能够在大规模上构建实验，尽快地进行迭代，尝试很多新的架构想法，对我们的目标有明确的方向，以高效地探索架构的可能性，争取比其他人更快找到答案。这就是我们的目标。我们在核心技术、大规模训练和尝试新架构上所做的所有工作，都是为了探索这个领域。
▶ 英文原文 ⏱
And we want to build the most automated lab to do it. We want to be able to build experiments at scale, the quickest we can iterate on them, try a lot of new architectural ideas, have strong priors of what we want to do to search the space of architectures efficiently to find, to go to that, go to that place fastest than, than anyone else. That's what we, what we want to do. And all the work we are doing on kernels, on large scale training, on trying new architectural ideas is exploring that space.
所以，你要通过实验来寻找一种更优越的架构。你怎么知道什么时候找到了？你需要看到什么才会说：“就是这个了”？这是个很好的问题。通常有两个角度。在我看来，根据我的经验，每一个成功的研究都会有一个不同寻常的现象出现，是其他研究没有的。有一条线明显有些不一样，带着一丝弯曲并以不同的方式展现出来。
▶ 英文原文 ⏱
So you're going to experiment your way into finding a superior architecture. How will you know when you've found it? What are you looking for to say, aha, this is the one? That's a great question. There are always two angles. In my mind, in my experience, every successful research had a plot that shows something that other other plots don't show. There's one line that is a little bit bending and in a different, different way.
你说的，这是你想要的。但至少，根据我在研究中的经验，情节的发展往往已经是旅程中较晚的阶段了，大多数时候你已经知道自己想要什么，已经了解你在做什么。我有点开玩笑，但事实上我生命中所有最好的情节，都是在梦中出现的，先于它们成为现实。我大概知道自己在寻找什么。只是问题在于什么时候才会真正地"恍然大悟"，如果你明白我的意思。因为多数时候，你大概知道自己在寻找什么，但就是找不到。你尝试了一件事，啊，不行。再试第二件事，啊，还是不行。
▶ 英文原文 ⏱
And you're saying this, this is what you want. But at least, at least my experience with research always has been that plot is already quite late in a journey where most of the time you already know what you want and already know what you are up to. I am a bit, a bit joking, but that is actually true that all the best plots in my life, I have them in a dream before, before actually, they were, they were, they were real. I kind of knew what I was looking for. Just, just the question is, it's like when, when it actually clicks, if you, if you know what I mean, because most of the time you kind of know what you are looking for, but you are, you are not finding it. You, you try one thing and ah, it doesn't work. Try second thing. Ah, it doesn't, it doesn't work.
最终，所有的，所有的正确部分都会自行就位。大多数深度学习系统都非常复杂。因此，通常你需要连续做对五件事情，链条才会开始运作。然后最终，你得到一个图表，看起来和你想的一样。然后，你会知道，我认为我们正在寻找的是能够在这个时候学习的系统。如果我们看到我们的系统有意义的长期适应能力。
▶ 英文原文 ⏱
But eventually the, all the, all the right pieces fall into it. And most of the deep learning systems are very intricate. So usually you have to get five things right in a row for the, for the link to start working. And then eventually you get the plot that looks like, like, like, like you want. And then, and then, you know, so, so I think what we are, what we are looking for is systems that learn at this time. And if we see meaningful, long-term adaptability of our, of our systems.
我们是在开玩笑，但我们确实想要评估我们日常工作的系统。核心自动化科学家的工作每天都在改进。我们甚至想过作为一个团队去度假，看看在那一周里实验室是否能创造出更好的成果。
▶ 英文原文 ⏱
And like, we are, we are, we're joking, but it's, it's, it's a, it's a real, we want to be evaluating our systems of our, of our everyday work. They get better at doing the work of core automation scientists each day. Yeah. We like go on a vacation as a team and see if the lab produces something better for the week, give them.
那么，当你回来后打算做什么呢？嗯，我们再看看吧。把假期延长两次，到四次，直到我们永久度假。嗯，这是个很美好的结束语。Rohan，Jerry，非常感谢你们的加入。你们俩在我们今天所在的领域做出了非常具有变革性的工作。我非常期待看到你们在这个新旅程中创办实验室，也很期待看到你们能取得什么成果。
▶ 英文原文 ⏱
Then what do you do when you get back? Um, we'll see. Extend, extend the vacation two times to the four times until we are on permanent vacation. Uh, that is a beautiful note to end on. Um, Rohan, Jerry, thank you so much for joining us. You've both worked on really, really transformative work for where we are today. And I'm so excited to see you starting a lab on this new journey, uh, and very excited to see what you're able to come up with.
感谢加入我们。托尼，能在这里和你聊天真是太好了。谢谢你。托尼，见到你真好。谢谢。
▶ 英文原文 ⏱
Thanks for joining us. Tony, it's been great to be here and chat with you. Thank you. Tony, it's been great to see you. Thank you.