# Demis Hassabis on shipping momentum, better evals and world models
- Transcript ID: `fbfdb04c9cf6beb9`
- Source tier: `podwise_public_transcript`
- Original language: `en_or_source_language_unlabeled`
- Translation status: `agent_translated_reviewed` (25/25 chunks)
- Original source: https://podwise.ai/dashboard/episodes/4934670
- Raw evidence: `data/transcripts/youtube/njDochQ2zHs.partial.json`
> 中文是派生翻译，英文是留存的源文本。中文未复核时，以英文和原始音视频为准。
## 00:00:00
**中文**
今天与我们一同参加节目的，是 Google DeepMind 首席执行官德米斯·哈萨比斯。
**English / Original**
Today we're joined by Demis Hassabis, who's the CEO of Google DeepMind.
## 00:00:03
**中文**
我们几乎每天都在发布新东西。即使在公司内部，也很难跟上。
**English / Original**
We're pretty much releasing something every day. It's hard to keep up, even internally.
## 00:00:07
**中文**
我感觉我们交付了很多东西：DeepThink、IMO 金牌模型、Genie 3；它们获得了极其热烈的反响。
**English / Original**
I feel like we're shipping lots of stuff. DeepThink, IMO Gold Medal, Genie 3, which the reception has been super incredible.
## 00:00:14
**中文**
我们想构建所谓的世界模型，即一种真正理解世界物理规律的模型。
**English / Original**
We want to build what we call a world model, which is a model that actually understands the physics of the world.
## 00:00:21
**中文**
构建技术，然后真正把它交到人们手中，是一种如此美妙的组合。
**English / Original**
Building the technology and then also actually putting it into people's hands is like such a Beautiful combination.
## 00:00:27
**中文**
我们开始看到这些模型逐渐汇聚为一种我们称为“全能模型”的东西，它能做一切事情。
**English / Original**
We're starting to see sort of convergence of those models together into, you know, kind of what we call an omni-model, which can do everything.
## 00:00:34
**中文**
因此，我们还宣布与 Kaggle 合作推出 Game Arena。
**English / Original**
So we also announced our partnership with Kaggle to launch Game Arena.
## 00:00:38
**中文**
顾名思义，最好的模型会彼此较量。它们变得更好时，测试会自动变得更难。我认为，随着我们越来越接近 AGI，这只是可能需要的许多新基准之一。我想说，这是行业中一个令人惊叹、令人兴奋的时刻。大家好，欢迎回到《Release Notes》。
**English / Original**
As the name suggests, the best models get pitted against each other. As they get better, the test will get harder automatically. And I think it's just one of probably many new benchmarks that are going to be required as we get closer to AGI. It's an amazing, exciting moment in the industry, I would say. Hey, everyone. Welcome back to Release Notes.
## 00:01:01
**中文**
我是 Google DeepMind 团队的洛根·基尔帕特里克。今天我们与 Google DeepMind 首席执行官德米斯·哈萨比斯交谈。
**English / Original**
My name is Logan Kilpatrick from the Google DeepMind team. Today, we're chatting with Demis Hassabis, who's the CEO of Google DeepMind.
## 00:01:07
**中文**
德米斯，感谢你来到这里。
**English / Original**
Demis, thanks for being here.
## 00:01:08
**中文**
很高兴聊聊过去几个月我们取得的所有发布成果和进展。
**English / Original**
Excited to chat about all of the releases and progress that we've had in the last few months.
## 00:01:12
**中文**
很高兴来到这里。
**English / Original**
Great to be here.
## 00:01:13
**中文**
先谈谈这股前所未有的势头吧。我感觉我们交付了很多东西：DeepThink、IMO 金牌模型、Genie 3，它们获得了极其热烈的反响；过去两个月还有大约 50 项别的东西，因为进展太快，我觉得我们都已经忘掉了。我想听听你对正在发生的这种进展和势头的高层看法。
**English / Original**
Let's start with just the unprecedented amount of momentum. I feel like we're shipping lots of stuff. DeepThink, IMO Gold Medal, Genie 3, which the reception has been super incredible, like 50 other things in the last two months that I feel like we've already forgotten about because things are moving so quick. I'm curious to just like get your high-level sense of that progress and momentum that's happening.
## 00:01:36
**中文**
是的，看到这一点非常棒。我认为，过去几年里，我们一直在逐步提升交付和进展的速度；如今你看到的正是其结果。但我会说，这是行业中一个惊人而又令人兴奋的时刻。似乎每天都有新东西发布。
**English / Original**
Yeah, it's fantastic to see. We've been sort of building up to the speed, I think, of our shipping and progress over the last couple of years. And I think this is that you're seeing the results of that now. But it's an amazing and kind of exciting moment in the industry, I would say. It seems to be, you know, things coming out every day.
## 00:01:54
**中文**
我们几乎每天都在发布新东西。无论是内部还是整个领域，都很难跟上。看到这一点令人兴奋，我也真的为我们最近发布的一些工作感到自豪和满意。
**English / Original**
We're pretty much releasing something every day. It's hard to keep up even internally and the field as a whole. It's exciting to see and I'm really proud and pleased with some of the latest work we put out there.
## 00:02:07
**中文**
你具体如何看待 DeepThink？显然，令我最兴奋的一点是，模型真正可用：Gemini 应用订阅用户可以使用 IMO 金牌模型的一个版本。人们可以真正上手这个模型；我认为这在过去——我们刚才在聊 Genie 的事——
**English / Original**
How are you thinking about DeepThink specifically? Obviously, one of the things I've been most excited about is the models actually available to a version of the IMO gold model is available for Gemini app subscribers. People can actually put their hands on the model, which I think has historically, we were pinging about Genie stuff.
## 00:02:27
**中文**
构建技术，然后真正把它交到人们手中，是一种非常美妙的组合。那么从 DeepThink 的角度看，你是如何思考它的？
**English / Original**
Like building the technology and then also actually putting it into people's hands is like such a. Beautiful combination. So from a DeepThink perspective, how are you thinking about it?
## 00:02:37
**中文**
嗯，我认为思考模型的出现，多少让人想起我们最初在 AlphaGo、AlphaZero 等项目上的游戏工作。事实上，从 DeepMind 创立之初，我们的工作史一直都与基于智能体的系统有关。所谓智能体系统，是指能够完成一整个任务的系统，对吧？
**English / Original**
Well, look, I think these, you know, the advent of the thinking models is a little bit hark back to our original gaming work on things like AlphaGo and AlphaZero. So we always worked on, you know, from the beginning of DeepMind, actually, the history of our work has always been with agent-based systems. So what we mean by that is systems that can complete a whole task, right?
## 00:02:55
**中文**
而在早期，主要是把游戏玩得非常好，并且有一个目标。基本上，你有这样一个模型——比如今天我们有非常强大的多模态模型，可以建模语言和周围的一切；但当年我们有的是游戏模型。然后还需要在其上叠加某种思考、规划或推理能力。
**English / Original**
And mostly in the early days, playing a game really well. And there's an objective. And you basically, you know, you have this model, you know, model of, say we have model, multimodal models, really powerful ones that model language and everything around us. But in those days, we would have game models. And then you need some thinking or planning or reasoning capability on top.
## 00:03:16
**中文**
这显然是通向 AGI 的路径。一旦有了思考能力，你就可以进行深度思考或极深度思考，并进行并行规划。你可以并行地规划和思考，再收敛到最优方案，作出决策，然后进入下一步。
**English / Original**
And this is obviously the way to get to, you know, AGI. And then, of course, once you have thinking, you can do deep thinking or extremely deep thinking and then sort of have parallel planning. You know, you can do sort of planning and thoughts in parallel and then collapse onto the best one and then make a decision and then move on to the next one.
## 00:03:36
**中文**
我认为这里仍需要相当多的创新。但即使只在思考这部分，看到这样的进步速度也令人兴奋。显然，对于数学、编程、科学问题，以及游戏，你都需要处理和规划，实质上进行这种思考，而不能只输出模型最先想到的东西。
**English / Original**
There's still, I think, quite a lot of innovation that's required there. But it's exciting to see the rate of progress, even in the thinking part of things. And obviously for things like maths, for coding, for scientific problems, also for gaming, you're going to need to process and plan and basically do this thinking and not just output the first thing that the model comes up with.
## 00:04:02
**中文**
那不大可能足够好。因此，你会想回过头来完善自己的思考过程，而这实际上正是思考系统所做的事。
**English / Original**
That's unlikely to be good enough. So you want to sort of go back and refine your own thought processes, which is in effect what the thinking systems do.
## 00:04:09
**中文**
是的，我此前没看过《Thinking Game》，大概一周半前看了。我一边看一边记笔记，心想：等等，德米斯和 DeepMind 团队在所有这些事情上都走在前面。片中有很多有趣的对应之处：比如你们曾试图扩展强化学习来解决以前的问题，这在今天是什么样子；还有 AlphaFold 的数据瓶颈是个很好的例子，类似于我们如今在编程等某些特定领域任务、或者科学领域之外，面对人类专家数据时的问题。
**English / Original**
Yeah, I was, I had not seen the Thinking Game and I watched it probably a week and a half ago or something like that. And I was scribbling down notes as I was as I was watching and I was like, wait, Demis and the DeepMind team were ahead of all of this stuff. I was like, there was a show and also there's just so many interesting parallels to like, as you were trying to scale up RL to solve like previous problems, like what that looks like today and like the like data bottleneck for AlphaFold is a good example and like similar to like how we have that problem today with like human expert data for some of these domain specific tasks like coding or just like outside of scientific domains.
## 00:04:45
**中文**
当我们现在用语言模型、并在其他领域扩展规模来解决这些问题时，这在多大程度上会让你有似曾相识的感觉？
**English / Original**
How much of that do you like Does it feel like deja vu as we're like now solving these problems with language models and scaling up in other domains?
## 00:04:54
**中文**
是的，我认为我们一直走在……我想如今已逐渐清楚，我们一直走在正确轨道上。基本上，我们是最早认真使用强化学习（RL）的一批。
**English / Original**
Yeah, I think we were always on the... I think it's sort of become clear we were always on the right track with You know, we were the first to sort of use RL seriously, basically.