如果在前沿模型上落后，连参与竞争的资格都没有！Google掌
字幕摘录
| 时间 | 英文 | 中文 |
|---|---|---|
| 0:00 | We are at early stages of the most consequential technology ever. | 我们正处于有史以来最先进的技术的初级阶段。 |
| 0:03 | This is not like fiber, it's not dark, we are using it. | 这不像纤维,不是黑暗,我们正在使用它。 |
| 0:06 | AI researchers, for them, it's like the bigger the computer, the better the AI. | AI研究者,对他们来说,就像计算机越大,AI越好. |
| 0:11 | You give the researchers the biggest possible computer, | 你给研究人员最大的计算机, |
| 0:14 | even if it is to the detriment of today's profits like the cloud business. | 即便损今利如云业. |
| 0:19 | So the compute we have today is a function of decisions I made in 2024. | 因此,我们今天的计算 是一个功能 我的决定在2024年。 |
| 0:25 | Today, obviously, when you sit and look at anything, | 今天,很明显, 当你坐着看着任何东西, |
| 0:28 | everyone is shocked, not just us, other companies, and everyone wishes | 每个人都很震惊 不只是我们 还有其他公司 每个人都希望 |
| 0:34 | you had two X of what you had planned for. | 你有你计划的两个X。 |
| 0:36 | Google的Muse或其DOTS产品在哪里? |
展开字幕全文（763 条）
| 序号 | 英文 | 中文 |
|---|---|---|
| 1 | We are at early stages of the most consequential technology ever. | 我们正处于有史以来最先进的技术的初级阶段。 |
| 2 | This is not like fiber, it's not dark, we are using it. | 这不像纤维,不是黑暗,我们正在使用它。 |
| 3 | AI researchers, for them, it's like the bigger the computer, the better the AI. | AI研究者,对他们来说,就像计算机越大,AI越好. |
| 4 | You give the researchers the biggest possible computer, | 你给研究人员最大的计算机, |
| 5 | even if it is to the detriment of today's profits like the cloud business. | 即便损今利如云业. |
| 6 | So the compute we have today is a function of decisions I made in 2024. | 因此,我们今天的计算 是一个功能 我的决定在2024年。 |
| 7 | Today, obviously, when you sit and look at anything, | 今天,很明显, 当你坐着看着任何东西, |
| 8 | everyone is shocked, not just us, other companies, and everyone wishes | 每个人都很震惊 不只是我们 还有其他公司 每个人都希望 |
| 9 | you had two X of what you had planned for. | 你有你计划的两个X。 |
| 10 | Google的Muse或其DOTS产品在哪里? | |
| 11 | Users, both in consumer and enterprise, will have access to | 消费者和企业的用户将有机会使用 |
| 12 | kind of some multi-agent framework for them working all the time doing things. | 一种多代理框架,让他们 一直在做事情。 |
| 13 | And so, you know, it feels like a moment of infinite possibilities. | 所以,你知道,它感觉 像一个无限的可能性的时刻。 |
| 14 | Alphabet CEO Sundar Pichai 加入我们讨论Google的竞争计划 | |
| 15 | on the frontier of AI, the balance between cutting-edge AI research | 在AI的前沿, 平衡的前沿AI研究 |
| 16 | and the company's cloud business and what the future of the interface looks like. | 和公司的云端业务 以及界面的未来 |
| 17 | And Sundar is here with us in studio in Mountain View today. | 孙达今天和我们在山景区工作室 |
| 18 | Sundar, great to see you. | 孙达,很高兴见到你 |
| 19 | Welcome to the show. | 欢迎光临节目 |
| 20 | Thanks, Alex. | 谢谢 亚历克斯。 |
| 21 | Great to be here. | 很高兴在这里。 |
| 22 | 所以Google和双子座4 Argon重回边境了 | |
| 23 | In plain speak, that means it is a world-leading model. | 简言之,这意味着这是一个世界领先的模式。 |
| 24 | It's benchmark-breaking. | 这是打破基准。 |
| 25 | 它胜过OpenAI和Anthropix的公开模式 | |
| 26 | on a variety of benchmarks, including on coding, math, science, and knowledge work. | 关于各种基准,包括编码、数学、科学和知识工作的基准。 |
| 27 | Now, the conventional wisdom had been that Google had basically given up on the frontier, | 现在,传统的智慧是谷歌 基本上放弃了前沿, |
| 28 | that it was going to make these fast models that were good enough | 它会让这些快速的模型 足够好 |
| 29 | and use them in its products, and it wasn't pursuing frontier AI anymore. | 并用在它的产品, 它不再追求前沿AI。 |
| 30 | But that was wrong. | 但这是错误的。 |
| 31 | I mean, very wrong. | 我的意思是,非常错误。 |
| 32 | Look, I think we are so committed to being at the frontier. | 听着,我想我们非常致力于在边境。 |
| 33 | I think it's in our DNA. | 我想是在我们的DNA里 |
| 34 | We won't do it any other way. | 我们不会用其他方式 |
| 35 | We've been thinking about the AI frontier for well over a decade now. | 我们想AI边境已经有十多年了 |
| 36 | We definitely, the frontier is very dynamic. | 我们肯定,边界非常活跃。 |
| 37 | It's very competitive. | 竞争非常激烈. |
| 38 | It's very exciting. | 非常令人兴奋 |
| 39 | I couldn't be more pleased with how Argon has come together. | 我对Argon的相聚感到很满意 |
| 40 | We took our time with it. | 我们慢慢来 |
| 41 | We made sure we tested it safely and deployed it carefully. | 我们确保了安全测试 并小心部署 |
| 42 | We took a bit longer than our other models, given its capabilities. | 由于其能力,我们比其他模型要长一点。 |
| 43 | But I think across the board, be it for agentic work, knowledge work, coding, | 但我认为从整体上来说 无论是代理工作 知识工作 编码 |
| 44 | creative tasks, I think Argon is one of the really good frontier models. | 我认为阿尔贡是很好的前沿模式之一。 |
| 45 | And in the first 72 hours since we launched, it's almost up to 25 trillion tokens used, | 在发射后的头72小时里, 它几乎高达25万亿令牌, |
| 46 | which is about 3x what Gemini 3.0 had in the first 72 hours. | 大约是双子座3.0在前72小时里拥有的3x. |
| 47 | And we are capacity constrained as we are serving it. | 我们在为它服务时能力受到限制。 |
| 48 | So excited, but there are always areas, both the frontier is just last. | 如此激动,但总是有的地区, 两个边疆只是最后。 |
| 49 | It's very dynamic, and you have to keep evolving the models. | 它非常动态, 你必须不断进化模型。 |
| 50 | And there are areas which we can see we need to improve as well. | 我们可以看到,有些领域也需要改进。 |
| 51 | So we'll be at work on all of that as well. | 所以,我们将在工作 这一切。 |
| 52 | Which areas? | 哪个地区? |
| 53 | Look, on any given model, Argon will be across the board better at everything | 听着,无论哪种模式,阿贡都会在一切方面做得更好 |
| 54 | if you just look even two months back. | 如果你只看 甚至两个月前。 |
| 55 | But the space is so evolving. | 但空间正在演化 |
| 56 | But even on coding, for example, there are areas, pockets where we know we need to do better. | 但即使是在编码上,比如说,有些地方,有些地方,有些地方 我们知道我们需要做得更好。 |
| 57 | And there's clear feedback, and so we'll make it better. | 并且有清楚的反馈, 所以我们会使它更好。 |
| 58 | Let me at least make the argument for why standard models might be good enough for Google. | 让我至少提出一个论点,说明为什么标准模型可能足以适合谷歌。 |
| 59 | I'm going to channel what people say about this. | 我要传达大家对此的看法 |
| 60 | And that is that the frontier of AI today is getting so good, and the baseline of AI | 这就是今天AI的疆域 变得如此美好,AI的基线 |
| 61 | is getting so good that the frontier is really only valuable for a specific set of tasks. | 变得如此美好 以至于边境对特定的任务来说 才真正有价值 |
| 62 | And if you have a standard model, you could actually build great products with a standard | 如果你有一个标准模型, 你实际上可以建造 伟大的产品与标准 |
| 63 | model. | 型号。 |
| 64 | Take Metamuse, for example. | 以梅塔缪斯为例. |
| 65 | Meta built Muse on a good model, but not the best models. | 梅塔在良好的模型上建造了Muse,但不是最好的模型. |
| 66 | If you need to do biological research, maybe the frontier is really necessary. | 如果你需要做生物学研究, 也许前沿是真正必要的。 |
| 67 | But the people who would say Google doesn't really need the frontier, | 但那些会说谷歌的人 并不真正需要前沿, |
| 68 | they would say you just need those standard models, and you can put them in products | 他们会说你只需要那些标准模型 你可以把它们放到产品里 |
| 69 | like Maps and Gmail and Search. | 像地图和Gmail和搜索。 |
| 70 | And again, take that meta example, that would be sufficient. | 再举一个例子,就够了。 |
| 71 | What do you think about that? | 你觉得怎么样? |
| 72 | Look, it's definitely not how I have thought about it at all. | 听着,这绝对不是我怎么想的 |
| 73 | I think we are deeply committed to driving the state of the art and actually push what | 我认为我们深深致力于驱动 最新艺术,并实际推动什么 |
| 74 | this technology can go, where it can go, all the way from Transformers. | 这个技术可以去,它可以去的地方, 从变形器一路。 |
| 75 | That's a journey we've | 这是我们的旅程 |
| 76 | been on. | 一直以来。 |
| 77 | So you will see us continue to do that. | 所以你会看到我们继续这样做。 |
| 78 | I think when you're at the frontier, | 我认为当你在边境, |
| 79 | you get an earlier look into the kind of emerging capabilities, earlier than others, which helps you | 你比其他人更早地研究了 新兴的能力 这有助于你 |
| 80 | better prepare for the future. | 更好地准备未来。 |
| 81 | We are fortunate to have a wide set of opportunities across | 幸运的是,我们有很多机会 |
| 82 | consumer and enterprise. | 消费者和企业。 |
| 83 | Some of our cloud customers for sure won the model of the frontier. | 我们的一些云端顾客肯定赢得了前沿的模型。 |
| 84 | I think the frontier models will be better at developing the next set of models in an | 我认为前沿模型将更好地开发 下一套模型在一个 |
| 85 | era of recursive self-improvement. | 后继自我改进的时代。 |
| 86 | So I would argue it matters there. | 所以我会说这很重要。 |
| 87 | Internally, | 内部, |
| 88 | we have definitely seen a steep point of inflection by deploying Argon to our | 我们确实看到一个尖端的裂缝 通过部署阿尔贡到我们 |
| 89 | developers. | 开发者。 |
| 90 | It definitely tackles problems which our previous models could not do. | 它确实解决了我们以前的模式无法解决的问题。 |
| 91 | And so | 这样 |
| 92 | being at that frontier puts you at the innovation frontier and helps you drive that as well. | 在那个前沿 让你进入创新前沿 也帮助你驱动它 |
| 93 | There is room for models of all sizes and capabilities. | 拥有各种大小和能力的模式。 |
| 94 | That's why we did flash, | 这就是为什么我们做闪光, |
| 95 | flashlight, flash, etc. | 手电筒,闪光灯等. |
| 96 | And I agree with you that you can develop very good product experiences | 我同意你的观点 你可以发展出非常好的产品经验 |
| 97 | with not necessarily the most frontier model because they tend to be slower and more expensive | 也不一定是最前沿的模型 因为它们往往比较慢和昂贵 |
| 98 | too. | 也一样。 |
| 99 | But we cared about the Pareto frontier and so we are committed to doing all of that well. | 但我们关心帕雷托边境 所以我们致力于做这些好事 |
| 100 | You said that developing the frontier gives you a look into what's coming. | 你说开发边境可以让你了解未来 |
| 101 | Yeah. | 对 |
| 102 | What do you see on the horizon? | 你在地平线上看到了什么? |
| 103 | You know, what makes this moment so exciting is you build something which you think is a | 你知道,是什么使这个时刻 如此激动人心的是 你建造的东西,你认为是一个 |
| 104 | culmination of months of work and you put it out, you know you already see clear headroom ahead. | 几个月的工作结束,你把它拿出来, 你知道,你已经看到清晰的头室前面。 |
| 105 | I think, I mean the frontier is going to, we've all been working on these agentic long horizon | 我想,我的意思是,边境是要去, 我们都一直在努力 这些代理长的视野 |
| 106 | capabilities. | 能力。 |
| 107 | But clearly the frontier is going to have multiple of these agents working together | 但很明显,边境上将会有多个特工合作 |
| 108 | to do things. | 做点什么 |
| 109 | So that's an example of an exciting paradigm underway. | 这是一个令人兴奋的范例。 |
| 110 | And it's also clear to me that | 我也明白这一点 |
| 111 | both in consumer and enterprise will have access to kind of some multi-agent framework for them | 消费者和企业将有机会利用某种多代理框架。 |
| 112 | working all the time doing tanks. | 总是在做坦克工作 |
| 113 | And so, you know, it feels like a moment of infinite possibilities. | 所以,你知道,它感觉 像一个无限的可能性的时刻。 |
| 114 | You mentioned recursive self-improvement. | 你提到递归自我改进。 |
| 115 | This is the models effectively making themselves better. | 这是有效让自己好起来的模型。 |
| 116 | Yeah. | 对 |
| 117 | Maybe without humans in the loop. | 也许没有人类在循环。 |
| 118 | People call it like an intelligence explosion or | 人们称之为情报爆炸 |
| 119 | this sound like boom. | 这听起来像爆破。 |
| 120 | And then all of a sudden AI is just like this superhuman capability. | 然后突然间AI就像这种超人的能力. |
| 121 | And in | 并进 |
| 122 | fact, I think Sergey was doing this round table with researchers at DeepMind and he talked about | 谢尔盖在DeepMind和研究人员举行圆桌会议 他谈论过 |
| 123 | with Argonne that the core, one of the core innovations for Argonne was relying more on | 与阿尔贡,核心, 阿贡的核心创新之一 更依赖 |
| 124 | the model to help develop the model, right? | 帮助开发模型的模型,对不对? |
| 125 | We've seen what's gone on with AI models recently. | 我们看到了AI模型最近的发展. |
| 126 | They | 他们 |
| 127 | are well aligned, but they're not perfectly aligned. | 完全对齐 但它们并不完全对齐 |
| 128 | And the imperfections are a big problem. | 和不完善是一个大问题。 |
| 129 | Obviously the tech world and now the world more broadly knows about the hugging face incident | 显然,科技界和现在的世界 更广泛地知道拥抱的脸部事件 |
| 130 | where open AIs, bots broke containment and hacked into hugging face and did some crazy stuff. | 打开的AI,bots破解 封闭和黑进拥抱的脸 做了一些疯狂的东西。 |
| 131 | 谷歌 | |
| 132 | has also had a scenario where it bots got access. | 也曾出现过它可以进入的情景。 |
| 133 | I think they were guessing passwords and they got | 我觉得他们猜到密码了 |
| 134 | access to a site they shouldn't have. | 进入他们不该拥有的网站 |
| 135 | They stopped. | 他们停下来了。 |
| 136 | But these models, there's enough evidence | 但是这些模型 有足够的证据 |
| 137 | that they're not doing exactly what we want them to be doing. | 他们并没有做我们想要他们做的事 |
| 138 | That this idea of pursuing | 这个想法是追求 |
| 139 | this recursive self-improvement or even making the models better much faster, | 这种递归的自我改进 甚至使模型变得更快 |
| 140 | it's sort of freaky. | 这有点怪异。 |
| 141 | It makes me scared. | 这让我害怕。 |
| 142 | Don't you think there's reasonable worry here and there | 难道你不认为有合理 担心在这里和那里 |
| 143 | should be something done to have the tech industry chill out about this? | 应该做点什么 让科技行业冷静下来吗? |
| 144 | Well, maybe three parts to your question. | 好吧,也许三个部分 你的问题。 |
| 145 | First, let's talk about RSI itself. | 首先,让我们谈谈RSI本身。 |
| 146 | You're right, | 你说得对 |
| 147 | Sergey mentioned RSI being critical contributor as we are developing Argonne. | 谢尔盖提到,在我们开发阿贡时,国际空间局是关键的贡献者。 |
| 148 | But to be very | 但要成为非常 |
| 149 | clear today, we are pretty far from the end state of RSI or something like that. | 今天很清楚,我们离 RSI 的结束状态 或类似的东西。 |
| 150 | Today, the right way to describe it is you have RSI loops along with AI researchers in various | 今天,描述它的正确方式是 你有RSI的循环 和AI研究人员在各种 |
| 151 | aspects of model development. | 模型开发的各个方面。 |
| 152 | I would say you almost have agents helping AI researchers | 我想说你们几乎有特工在帮人工智能研究人员 |
| 153 | in some automated way work together to build these models. | 以某种自动化方式合作建造这些模型。 |
| 154 | I think it's good to at least | 我觉得这至少对 |
| 155 | understand where we are today. | 了解我们今天的处境。 |
| 156 | It's super exciting and I expect that continuum to keep making | 超刺激的,我期望这个连续体 继续制作 |
| 157 | progress as we go into next year. | 我们明年将取得进展。 |
| 158 | The second question is around safety. | 第二个问题是关于安全。 |
| 159 | Security comes from a difference in depth. | 安全来自深度的不同。 |
| 160 | You're talking about alignment. | 你在说对齐 |
| 161 | Obviously, | 显然 |
| 162 | we all are doing a lot of work to make sure the models are as aligned as possible. | 我们都在做很多工作,以确保模式尽可能一致。 |
| 163 | But it's an active area of research and there are clear gaps. | 但它是一个活跃的研究领域,存在明显的空白. |
| 164 | But that's not the only thing | 但这不是唯一的事情 |
| 165 | we rely on. | 我们依赖。 |
| 166 | You have the work you do on alignment. | 你有你的工作 在对齐。 |
| 167 | But beyond that, we put a lot of effort into | 但除此之外,我们付出了很多努力 |
| 168 | real-time monitoring as these models are working. | 由于这些模型正在发挥作用,实时监测。 |
| 169 | Making sure you have good telemetry into what's | 确定你对什么有很好的遥测 |
| 170 | happening so that you can intervene in real-time if you see unexpected behaviors. | 如果你看到出乎意料的行为 就可以实时介入 |
| 171 | How you do that, | 或暗 |
| 172 | the architecture with which you do that, how do you make sure it's all sandboxed and isolated | 如何确保它都是沙箱和隔离的 |
| 173 | in a way that it can continue doing its job during real-time monitoring. | 在实时监测期间,它能够继续履行职责。 |
| 174 | These are all areas | 这些都是领域 |
| 175 | of investment we are doing. | 我们正在做的投资。 |
| 176 | Obviously, when we deploy these models, giving customers, enterprises | 显然,当我们部署这些模型,给客户,企业 |
| 177 | control both with real-time monitoring, but their ability to decide where they deploy | 进行实时监测,但有能力决定部署地点 |
| 178 | what type of data these models ever have access to, et cetera. | 这些模型能获取什么类型的数据等等。 |
| 179 | All that work is very, very important | 所有的工作都是非常,非常重要的 |
| 180 | 我们和Argonne做了很多 | |
| 181 | As an industry, we need to constantly up our game | 作为一个行业,我们需要不断提高我们的游戏 |
| 182 | and take the time to do all this well, including dedicating the compute required to make sure you | 并花时间做所有这些好, 包括专门 计算所需 以确保你 |
| 183 | can safely deploy these models. | 可以安全部署这些模型。 |
| 184 | The third part is, is it all enough? | 第三部分是,一切都够了吗? |
| 185 | I think this is why I think | 我想这就是为什么我想 |
| 186 | you know, there's a lot of conversations in industry and I think it's reasonable to be | 工业界有很多对话 我觉得很合理 |
| 187 | worried, but I want the right way to approach something you worry about it is early on thinking | 担心,但我想用正确的方法 处理你担心的事情 才刚刚开始思考 |
| 188 | about what more we can do. | 关于我们能做的更多。 |
| 189 | I think all of us are, I think as an industry, I think you're definitely | 我认为我们都是,我认为作为一个行业, 我认为你绝对 |
| 190 | seeing an increased levels of work on alignment, testing for safety, et cetera, as we deploy these | 我们看到,随着我们的部署,协调、安全测试等方面的工作有所增加。 |
| 191 | I think the next step would be developing better standards so that across industry standards of how | 我认为下一步是制定更好的标准,以便通过行业标准了解如何 |
| 192 | we do these things, including ability to evaluate using independent evaluators. | 包括利用独立评估员进行评估的能力。 |
| 193 | If you look at areas | 如果你看看区域 |
| 194 | like financial controls, we have like accounting standards, we have independent auditors, et cetera. | 像财务控制,我们有像会计准则一样的, 我们有独立的审计员等等。 |
| 195 | Those are all the next natural steps for us as an industry to work towards this and regulation | 这些都是接下来的自然步骤 我们作为一个行业 努力争取这个和监管 |
| 196 | end up playing a role as we go through all of that as well. | 最终在我们经历这一切的时候扮演了角色. |
| 197 | Okay. | 摆 |
| 198 | Technology question for you. | 技术问题 |
| 199 | Google is somewhat unique in that you research the language or the LLM side of the equation, | Google在研究语言或LLM方程方面有些独特, |
| 200 | which started with the transformer language, and then you also pursue world models. | 它从变压器语言开始, 然后你也追求世界模型。 |
| 201 | There are some AI research efforts out there like OpenAI that have basically determined that the | 有一些AI的研究工作,比如OpenAI, 基本上确定 |
| 202 | LLM is the path to AGI and they've actually subset part of their world model research, | LLM是通往AGI的路径,他们实际上 子集了他们的世界模型研究的一部分, |
| 203 | which is like the AI model that tries to understand the physical world, but you're pursuing both. | 这就像AI模型 试图理解物理世界, 但你追求两者。 |
| 204 | So talk a little bit about why you find the world model side to be important. | 说说为什么你觉得世界模式很重要。 |
| 205 | As humans, you obviously perceive the world around you and you're taking all those inputs, | 作为人类,你显然能体会到周围的世界 并且你正在接受所有这些投入, |
| 206 | you know, making sure these models are able to understand the laws of physics, | 你知道,确保这些模型 能够理解物理定律, |
| 207 | simulate the world in high fidelity, all that I think are key enabling blocks for intelligence. | 以高度的忠诚模拟世界 我认为一切是智力的关键要素 |
| 208 | And I think as you're working across all possible applications of this, | 我认为,当你工作 在所有可能的应用 这个, |
| 209 | take an area like robotics, it's not an accident or gone. | 带一个区域像机器人, 它不是一个意外或消失。 |
| 210 | If you look at embodied reasoning | 如果你看看内含的推理 |
| 211 | benchmarks are going to state of the art, right? | 基准会是 最新技术,对不对? |
| 212 | I think literally sawed on many of these benchmarks | 我认为,这些基准中有很多被锯了 |
| 213 | and that all comes from, you know, our experience building world models as well. | 这一切都来自,你知道,我们的经验 建设世界模型。 |
| 214 | So | 这么说 |
| 215 | inherently, I would expect, you know, there will be architectures which will | 内在的,我会期望,你知道, 会有建筑会 |
| 216 | converge all of this in the future, but I think it's for us, we view this as today, | 将来把这些都汇合在一起 但我认为这对我们来说是今天 |
| 217 | you know, in this jagged frontier that is intelligence, we want to be at the frontier | 要知道,在这个悬殊的边界,即情报, 我们想在边界 |
| 218 | of all the possible approaches and I think that will help us both drive better research, | 我想这将有助于我们推动更好的研究 |
| 219 | better insights and develop a much better, you know, AGI model, you know, is how we think about it. | 更好的洞察力,开发出更好的,你知道的, AGI模型,你知道的, 是我们如何思考的。 |
| 220 | Yeah, I looked at your Twitter after the Argonne announcement and it's all just like | 是的,我看了你的推特 之后的阿尔贡公告 这一切就像 |
| 221 | real world application and it's not like text or, you know, a model generating text. | 真实世界的应用, 它不像文本 或者,你知道,一个模型生成文本。 |
| 222 | Okay, I want to talk to you about the way that you balance allocating compute, which is your most | 好吧,我想和你谈谈 你如何平衡分配计算, 这是你最 |
| 223 | precious resource between cloud and AI research. | 云与AI研究之间的宝贵资源. |
| 224 | That's a big topic about Google. | 这是一个关于谷歌的大话题。 |
| 225 | I've had people | 我有过人 |
| 226 | suggest to me that Google should get out of the AI research game and, you know, take all of its | 我建议Google从AI研究游戏中脱身 并且,你知道的,拿走它的全部 |
| 227 | cloud computing resources and its chips and just sell that and you look at the world of business | 云计算资源及其芯片,只要卖掉它,你就看看商业世界 |
| 228 | right now and Nvidia, which is like just doing that effectively, five trillion dollar company, | 现在和恩维迪亚, 这就像只是有效做, 五万亿美元的公司, |
| 229 | there's a push and pull between cloud that wants to, you know, sell this computing resource | 云层之间有一种推力和拉力 想要卖掉这个计算资源 |
| 230 | at a high markup and AI research that needs a very big computer to make the best AI. | 在高标价和AI的研究, 需要一个非常大的计算机 做最好的AI。 |
| 231 | Now, you've said a big part of your job is figuring out how to allocate compute. | 现在,你说你工作的很大一部分 是想出如何分配计算。 |
| 232 | How do you do this and how do you balance these two? | 你怎么做到的 如何平衡这两个? |
| 233 | Look, I think we are, | 听着,我想我们是, |
| 234 | look, the ground zero for us is making sure we are investing in the compute that's needed | 听着,我们的地面0 正在确保我们投资 计算需要 |
| 235 | to be at the AI frontier. | 在AI边境。 |
| 236 | So that is, you know, in 2015, I said the company to be AI first. | 所以,在2015年,我说公司是AI。 |
| 237 | So AI is | 所以说,AI是 |
| 238 | that key horizontal ingredient across everything we do, all the way from search to cloud to Waymo | 从搜索到云端到Waymo |
| 239 | to everything we do. | 我们所做的一切。 |
| 240 | So I think that is ground zero for us, making sure both for research | 所以我觉得这对我们来说是零地, 确保两个研究 |
| 241 | and developing the models that we have enough compute needed. | 并开发我们有足够的计算所需的模型。 |
| 242 | Beyond that, look, we have these | 除此之外,你看,我们有这些 |
| 243 | independent businesses which each have their own long-range plans and so they're using those | 独立企业,每个人都有自己的长期计划 所以他们利用这些计划 |
| 244 | long-range plans to project forward and we plan for compute for all of them. | 远期计划 向前投影 我们计划 计算所有它们。 |
| 245 | And so the cloud team, | 故云团队, |
| 246 | you know, they have a business and they have a projection and they have demand signals, | 你知道,他们有一个业务, 他们有一个投影, 他们有需求信号, |
| 247 | they've committed contracts. | 他们签了合同 |
| 248 | So with that help, we can plan on a long-term compute trajectory for | 因此,有了这个帮助,我们可以规划一个长期的计算轨迹 |
| 249 | them. | 他们。 |
| 250 | I think we are fortunate to have this wide set of opportunities. | 我认为,我们很幸运有这样广泛的机会。 |
| 251 | I think there's no company | 我觉得没有伴 |
| 252 | in the world which isn't making trade offsets at any given moment in time. | 在世界上, 贸易不会在任何特定的时刻抵消。 |
| 253 | But the first principles | 但第一个原则 |
| 254 | we operate on is to making sure we have enough compute to invest in the frontier. | 我们的行动就是确保我们有足够的计算 投资边境。 |
| 255 | And that's | 这是 |
| 256 | non-negotiable and everything else works on top of it. | 不可谈判,其他一切在它之上。 |
| 257 | I think what you're seeing in these | 我想你看到的这些 |
| 258 | conversations is if you sit now, remember, we are building our own compute, our own data centers. | 对话是,如果你现在坐下,记住, 我们正在建立我们自己的计算, 我们自己的数据中心。 |
| 259 | So there's a long lead times. | 所以,有一个很长的准备时间。 |
| 260 | So the compute we have today is a function of decisions I made in | 所以,我们今天的计算 是一个函数 我的决定在 |
| 261 | 2024. | 2024 (英语). |
| 262 | Even if you think back to a year from today, last year, the agent recording takeoff hadn't | 即使你想起从今天起的一年 去年 记录起飞的特工没有 |
| 263 | happened. | 发生了 |
| 264 | So today, obviously, when you sit and look at anything, everyone is short, not just us, | 所以,今天,很明显,当你坐着看任何东西, 每个人都很矮,不只是我们, |
| 265 | other companies, and everyone wishes you had two X of what you had planned for. | 所有人都希望你有你计划的两个X |
| 266 | I think that comes | 我想会来的 |
| 267 | with the territory, right? | 与领地,对不对? |
| 268 | And like, you know, at any given moment in time, we are planning for what | 就像,你知道,在任何特定的时间, 我们计划什么 |
| 269 | we would need two to three years out in the future. | 我们以后需要两到三年的时间 |
| 270 | So I think you're going to find some | 所以,我想你会找到一些 |
| 271 | decisions you make may look suboptimal two years out, right, depending on where the state of the | 你的决定可能看起来不够理想 两年后,对不对,取决于 状态 |
| 272 | art is. | 艺术是。 |
| 273 | So I don't spend too much time thinking about it. | 所以我不会花太多时间去考虑 |
| 274 | The way I think about computer allocation | 我对电脑分配的看法 |
| 275 | is making sure Google DeepMind, both for research and frontier model development, | 正在确保Google DeepMind, 既用于研究和前沿模型开发, |
| 276 | is cutting edge in terms of compute access, right? | 就计算访问量而言 是前沿的 对吧? |
| 277 | And after that, we plan for all our businesses. | 之后,我们计划我们所有的生意。 |
| 278 | I think it's great we can provide our compute to other people too, like we have done. | 我觉得我们也可以像以前一样 向其他人提供我们的计算数据 |
| 279 | It allows us to scale that up, get real test for it in the widest application set possible, | 它允许我们扩大范围, 得到真正的测试 在最广泛的应用设置, |
| 280 | test it in the real world and make sure we are at the cutting edge of technology at every layer of | 在现实世界中测试它 确保我们处于技术的前沿 |
| 281 | our AI stack as well. | 我们的人工智能堆栈也一样。 |
| 282 | Briefly, there was a podcast where Ben Horowitz relayed that Andy Jassy had | 简而言之,有一个播客 Ben Horowitz 传递安迪・贾西 |
| 283 | said something like, computer sold out till 2028. | 说到2028年电脑卖光了 |
| 284 | Are you seeing the same thing? | 你看到同样的东西吗? |
| 285 | You know, I think it's a reasonable, like, you know, if today a customer is coming out and | 你知道,我认为这是一个合理的, 像,你知道,如果一个顾客今天要出来, |
| 286 | asking us for compute in 27, that is incredibly difficult to do. | 要求我们计算27, 这很难做。 |
| 287 | You know, there's part of the | 你知道,还有一部分 |
| 288 | reason even last year we went out and leased some short-term compute was to accommodate some | 连去年我们出去租了几笔短期的计算费 是为了适应一些 |
| 289 | customers and help them out. | 顾客和帮助他们。 |
| 290 | But yes, yeah, I think we are all dealing with similar timelines. | 但是,是的,是的,我想我们都在处理类似的时间表。 |
| 291 | One bubble question. | 一个泡泡问题 |
| 292 | There's talk of how every technology shift has seen overbuilding | 人们在谈论每个技术转变 如何看到过度建设 |
| 293 | railroads, fiber. | 铁路 纤维 |
| 294 | Do you think there's overbuilding here? | 你觉得这里有过度建设吗? |
| 295 | It depends on the timeframes you're looking at. | 这取决于你所看到的时间范围。 |
| 296 | Are we, you know, this level of build, | 难道我们,你知道, 这个水平的建筑, |
| 297 | you know, if I take a long enough timeframe, it makes complete sense to me. | 如果我花足够长的时间 这对我来说是完全合理的 |
| 298 | We are at early | 我们早到了 |
| 299 | stages of the most consequential technology ever. | 这是史上最先进的技术 |
| 300 | And this is not like fiber, it's not dark, | 这不是像纤维, 它不是黑暗, |
| 301 | we are using it, we are capacity constrained on a daily basis. | 我们正在使用它,我们的能力每天都受到限制。 |
| 302 | The applications, the underlying | 应用、基础 |
| 303 | technology is improving rapidly, right? | 科技进步很快,对吧? |
| 304 | But running a company, the way I work through the moment is | 但经营一家公司,我的工作方式是 |
| 305 | you look at your demand signals, you look at your committed revenue opportunities that you're getting | 你看你的需求信号,你看你的承诺 收入机会,你得到 |
| 306 | in the forms of contracts, you balance that with like the underlying technical capability, | 在合同的形式,你平衡 与基础技术能力, |
| 307 | our internal evals, how we are using it internally, the model trajectories. | 我们内部的电压,我们如何在内部使用, 模型轨迹。 |
| 308 | And based on all that, | 根据这些 |
| 309 | you take a disciplined approach, a confident, aggressive, but, you know, a disciplined approach | 你采取纪律性的方法, 自信的,积极的, 但是,你知道,纪律性的方法 |
| 310 | to plan out compute. | 计划计算。 |
| 311 | But when you talk about the broader economy as a whole, you're having people | 但当你谈论整个经济的时候, 你有一个人 |
| 312 | operate with various levels of risk. | 操作时有不同程度的风险。 |
| 313 | And so, you know, it's tough for me to, I focus on what I can | 所以,你知道,这对我来说很难, 我专注于我能 |
| 314 | control as a company and plan through it. | 控制公司,并计划通过它。 |
| 315 | If your question is, can there be an overbuild? | 如果你的问题是, 会不会有过度建设? |
| 316 | Yes, | 对 |
| 317 | because, you know, you can have some constrained components. | 因为,你知道, 你可以有一些限制的组件。 |
| 318 | And remember, what matters is a square | 记住,重要的是一个方块 |
| 319 | set of everything you need to build out that compute. | 编集所有你需要建立 计算。 |
| 320 | Memory, electricity, electricians. | 记忆 电力 电工 |
| 321 | Yeah, like 2000 components or more, right? | 是的,像2000个组件或更多,对不对? |
| 322 | And so, you could get constrained at some point. | 所以,你可以在某个时候受到约束。 |
| 323 | There could be things which cause bottlenecks and, you know, and then you could have overbuilds. | 可能有些事情 造成瓶颈,你知道的, 然后,你可以过度建设。 |
| 324 | There could be external shocks, right, to the economy, et cetera. | 可能会有外部冲击,对经济等等。 |
| 325 | So there are macro risks, | 所以有宏观风险, |
| 326 | all that is possible at any given time. | 任何时间都有可能 |
| 327 | But, you know, am I confident that the compute we are | 但是,你知道,我相信 我们的计算 |
| 328 | given enough time? | 足够时间吗? |
| 329 | Absolutely, yes. | 当然,是的。 |
| 330 | Okay, one more on this. | 好吧,再来一个这个。 |
| 331 | You know, if you're going to do both, there's been a line of thinking out there that AI researchers, | 你知道,如果你要做两者, 有一条线的想法 在那里,AI研究人员, |
| 332 | for them, it's like the bigger the computer, the better the AI. | 对他们来说,这就像计算机越大,AI越好。 |
| 333 | And if, you know, you're someone, | 如果,你知道,你是某个人, |
| 334 | I was in the room when you said that AI would be as profound as fire and electricity, and we all | 我当时在房间里 你说AI会像火和电一样深刻 |
| 335 | laughed at the time, and I kind of regret that laugh now. | 当时我笑了,现在我有点后悔 |
| 336 | But if you're someone that actually | 但如果你是一个真正的人 |
| 337 | believes that, then the thinking would be you give the researchers the biggest possible computer, | 相信,那么你的想法是 你会给研究者最大的可能计算机, |
| 338 | even if it is to the detriment of today's profits, like the cloud business. | 即便损今日之利,如云业. |
| 339 | Your thoughts? | 你的想法吗? |
| 340 | We're all doing some form of that, right? | 我们都在做某种形式,对吧? |
| 341 | I think our cloud business is, at any given moment in time, | 我认为我们的云层生意是, 在任何特定的时间时刻, |
| 342 | forgoing short-term revenues because we are making compute trade-offs to support frontier AI | 放弃短期收入,因为我们正在计算权衡,以支持边境AI |
| 343 | development. | 发展。 |
| 344 | So I think what you're saying is true already, right? | 所以我认为你说的是真的,对吧? |
| 345 | But, you know, you want to be | 但是,你知道,你想成为 |
| 346 | thoughtful, right? | 体贴,对不对? |
| 347 | Even our AI researchers, you know, some constraints are also good. | 甚至我们的人工智能研究人员, 你知道,一些限制也是好的。 |
| 348 | That's what | 就这样 |
| 349 | inspires creativity. | 激发创造力。 |
| 350 | You know, that's what helps you come with a more efficient algorithmic | 你知道,这就是帮助你 来一个更有效的算法 |
| 351 | breakthrough, maybe to use less memory, etc. | 突破,也许用来减少记忆等等。 |
| 352 | So, you know, so what we are committed is | 所以,你知道,所以我们的承诺是 |
| 353 | to invest what's needed to stay at the frontier, right? | 投资需要留在边境,对不对? |
| 354 | That's how I think about it. | 亦复如是念处. |
| 355 | I wonder what you think the future of the interface looks like, the way that we use | 我想知道你认为界面的未来是什么样子的 我们使用的方式 |
| 356 | computing in the future as AI starts to roll out. | 随着AI开始推出,计算未来. |
| 357 | I was speaking with Greg Brockman, the president | 我和总统格雷格·布鲁克曼谈过 |
| 358 | of OpenAI, about this a couple months ago. | OpenAI公司 几个月前的事 |
| 359 | He had a very interesting quote. | 他有一个非常有趣的引言。 |
| 360 | He said, | 他说, |
| 361 | what is the interface that you want? | 您想要的接口是什么 ? |
| 362 | What is the product that you want? | 你想要的产品是什么? |
| 363 | The answer is that | 答案是, |
| 364 | you want no interface. | 您不需要接口。 |
| 365 | You want no product. | 你不想要产品。 |
| 366 | You want this to be what's the interface between | 你想让这成为我们之间的接口 |
| 367 | you and me, basically saying, just talking. | 你和我,基本上说,只是说话。 |
| 368 | That if AI is that smart, there shouldn't be like, | 如果AI那么聪明,那就不应该像... |
| 369 | you know, maybe a screen or just so many buttons and things that you have to touch. | 你知道,也许一个屏幕 或只是这么多按钮和东西 你必须触摸。 |
| 370 | What is your view of what the interface looks like in the age of AI? | 你对界面在AI时代是什么样子的看法是什么? |
| 371 | The way I think about it is, I do feel, you know, computing historically has been humans | 我的想法是,我觉得, 你知道,计算历史上是人类 |
| 372 | adapting to computing. | 适应计算。 |
| 373 | With AI, it'll be the other way about. | 与AI,这将是另一条路。 |
| 374 | The technology will adapt seamlessly | 这种技术会无缝地适应 |
| 375 | to how humans live and go about their day-to-day lives. | 人类如何过着日常生活 |
| 376 | So there's an aspect of it. | 所以有其一义. |
| 377 | You know, | 你知道吗, |
| 378 | you want that AI to be able to speak to it when you want it to. | 你想让人工智能在你想说话的时候能和它说话 |
| 379 | It has context. | 它有内涵。 |
| 380 | It can | 这可以 |
| 381 | be persistently running somewhere, give you continuity. | 坚持跑的地方,给你连续性。 |
| 382 | You're not dealing with a lot of | 你不处理很多 |
| 383 | complexity. | 复杂。 |
| 384 | It's natural. | 自然无碍. |
| 385 | All that makes sense to me. | 所有这一切对我来说是有意义的。 |
| 386 | That doesn't mean it won't have an interface, | 这并不意味着它不会有接口 |
| 387 | however, to me. | 然而,对我来说。 |
| 388 | Look, today, for example, I mean, nowadays, I just even in my phone, like, you know, | 比如说,今天,我是说,现在,我甚至刚在我的手机里,喜欢,你知道, |
| 389 | Gemini voice models are so good, I can just press the microphone button. | 双子座的音效很好 我可以按下麦克风按钮 |
| 390 | I can just ramble. | 我可以轻举妄动。 |
| 391 | That's | 这是 |
| 392 | 叫做drambler,对不对? | |
| 393 | I can ramble. | 我可以摇摆。 |
| 394 | It kind of intelligently corrects with my personal context. | 我的个人背景 有点明智地纠正了 |
| 395 | So even my friend's names, weird names, it kind of gets it right, even if it first guesses it wrong. | 因此,即使是我的朋友的名字,奇怪的名字, 它有点得到它是正确的, 即使它首先猜错了。 |
| 396 | It's amazing. | 不可思议. |
| 397 | I'm not going in backspace editing once I finish saying it, right? | 我一说完就不去后太空编辑室了 对吧? |
| 398 | It's remarkable | 真不错 |
| 399 | how good it's gotten. | 感觉真好 |
| 400 | But there are going to be moments you will need an interface. | 但有时你需要接口 |
| 401 | You know, | 你知道吗, |
| 402 | you may want to entertain yourself. | 你可能想娱乐自己。 |
| 403 | You may want to watch a video, like, you know, so you may ask | 你也许想看一个视频, 比如,你知道,所以你可以问 |
| 404 | AI to create something and you want to watch it, consume it, right? | AI来创造一些东西 你想看它,消耗它,对不对? |
| 405 | So I think the way I think | 所以,我认为的方式 |
| 406 | about it is you want it to be as seamless and as natural as possible and that you are not doing the | 你希望它尽可能的无缝和自然 而你没有做 |
| 407 | hard work adapting to whatever the interface is. | 努力适应任何接口 |
| 408 | The technology is adapting to how you're working | 技术是适应你的工作方式 |
| 409 | at any given moment in time. | 在任何特定的时间。 |
| 410 | Yeah, I'm just going to play the open AI hits because, you know, last | 是啊,我只是要播放 公开AI点击,因为,你知道,最后 |
| 411 | year you sent them into a code red. | 你把他们送到一个红色代码。 |
| 412 | I'm sure you know. | 我肯定你们知道 |
| 413 | Gemini was doing very well in open AI, | 双子座表现很好 在公开AI, |
| 414 | batten down the hatches. | 关上舱门 |
| 415 | And I was speaking with Sam Altman about the challenge from Google. | 我和萨姆·阿尔特曼谈到谷歌的挑战。 |
| 416 | What he said was he thought open AI would fend you off because they're building AI native. | 他说他觉得开放的AI会挡住你 因为他们在建造AI本地人 |
| 417 | And I don't know if he said it specifically like this, but he implied Google is bolting on AI to | 我不知道他是不是这么说的 但他暗示Google正在对AI进行搜索 |
| 418 | existing products. | 现有产品。 |
| 419 | What is your view of whether, you know, a bolt on, I wonder if you would accept | 你的看法是什么,你知道, 是否一个螺栓, 我想知道你是否会接受 |
| 420 | framing can compete with an AI native product? | 框架可以和AI本地产品竞争? |
| 421 | First of all, I mean, it's a very dynamic moment. | 首先,我的意思是,这是一个非常活跃的时刻。 |
| 422 | We are all pushing each other. | 我们都相互推挤。 |
| 423 | I think open AI has pushed us hard as a company too. | 我认为开放的AI也把我们作为公司大力推倒了. |
| 424 | So this is | 这就是 |
| 425 | not just, you know, we are doing something to them alone, et cetera, far from it. | 不仅仅是,你知道,我们正在做一些事情 他们单独,等等,远离它。 |
| 426 | Yeah, you went into | 是啊,你进入了 |
| 427 | code red too. | 红色代码也。 |
| 428 | I mean, I know it's disputed about whether it was, but a version of this. | 我的意思是,我知道这是不是有争议的, 但这个版本。 |
| 429 | I mean, | 我是说... |
| 430 | it's been the AI moment all of us had to respond with varying levels of urgency at any given moment | 在任何特定时刻,我们都不得不以不同程度的紧迫感作出反应。 |
| 431 | in time. | 在时间。 |
| 432 | Look, I think, you know, we are constantly building new products as well. | 你看,我想,你知道, 我们正在不断建设新的产品。 |
| 433 | I mean, Gemini was | 我的意思是,双子座是 |
| 434 | built from the ground up as a product, the Gemini app I'm talking about. | 我说的是双子星的应用软件 |
| 435 | Or, you know, tomorrow, | 或者,你知道吗,明天, |
| 436 | as all of us are adapting to agents, are you bolting agents onto ChatGPD or are they building | 因为我们都是在适应特工, 你们是把特工放进ChatGPD, 还是他们正在建造 |
| 437 | it natively? | 是本地来的吗? |
| 438 | How would you think about that? | 你会怎么想? |
| 439 | Right. | 对 |
| 440 | So you can always flip these questions back. | 所以你总能把这些问题翻回来 |
| 441 | Right. | 对 |
| 442 | So I don't think of it that way. | 所以我不这么认为 |
| 443 | I mean, I think of it as you want to build world-class product | 我的意思是,我想它,因为你想 建造世界级的产品 |
| 444 | experiences that are AI native and they are winning. | 经验是AI本土的 他们正在赢得。 |
| 445 | And some of it is adapting products, | 有些是适应产品 |
| 446 | some of it is rethinking new experiences. | 其中一些正在重新考虑新的经验。 |
| 447 | So they will all be part of how you approach this moment. | 因此,他们都将是 你如何接近这一刻的一部分。 |
| 448 | Talk a little bit about the way that you see the business of this AI interface in the future. | 略谈你未来看待这个AI接口业务的方式. |
| 449 | So | 这么说 |
| 450 | for instance, if I'm asking my agent to go out and do something for me and | 比如,如果我要求我的经纪人出去 帮我做点什么 |
| 451 | it suggests a product, should it, I mean, maybe even if disclosed, does it get a cut of the sale? | 它暗示一种产品, 是否,我的意思是,也许,即使披露, 它得到一个削减的销售? |
| 452 | If I make the sale or when you think about the way that you're going to make money from these products, | 如果我做这个买卖 或者当你想到 你从这些产品赚钱的方式, |
| 453 | what's the plan? | 有什么计划? |
| 454 | I think, you know, obviously I'm going to put enterprise aside. | 我想,你知道, 显然,我要把企业放在一边。 |
| 455 | I think consumer | 我觉得消费者 |
| 456 | question. | 问题。 |
| 457 | So, you know, if it's consumer, look, I think there are consumers are going to increasingly | 所以,你知道,如果它是消费者, 你看,我认为有消费者会越来越 |
| 458 | value this technology. | 重视这项技术。 |
| 459 | They're going to meaningfully make take software developers today. | 他们今天要让软件开发者 真正参与其中 |
| 460 | I think | 我觉得 |
| 461 | developers are willing to pay for these coding models. | 开发者愿意为这些编码模型付费. |
| 462 | Right. | 对 |
| 463 | And because it really helps make | 因为它真的 帮助制作 |
| 464 | them more productive. | 他们更有成效。 |
| 465 | Nobody would want to code the way they were coding three years ago, or even | 没有人会想 编码的方式 他们的编码三年前,甚至 |
| 466 | like same time last year. | 就像去年一样 |
| 467 | So I think when you're delivering value, there's value to be had. | 所以,我认为当你提供价值, 有价值。 |
| 468 | I think history of consumer product shows, if you build a consumer product, | 我认为消费产品的历史显示, 如果你建立一个消费产品, |
| 469 | I've never seen examples where consumer products really scale to billions of users, | 我从未见过 消费者产品 真正规模到数十亿用户, |
| 470 | where people are using it intensely and hasn't had monetization to go along with it. | 人们正在紧张地使用它 也没有货币化来配合它。 |
| 471 | If you're providing value to people, there is inherent value to be had. | 如果你在为人们提供价值, 就有内在的价值。 |
| 472 | Totally. | 当然 |
| 473 | And so | 这样 |
| 474 | the nature of it may change. | 它的性质可能会改变。 |
| 475 | Right. | 对 |
| 476 | I think you're asking a scenario in which agents are doing stuff | 我想你是在问一个 特工们在干嘛的情景 |
| 477 | for you. | 给你的 |
| 478 | At the end of the day, you will have commercial needs. | 到最后,你会有商业需求。 |
| 479 | The agents need to fulfill | 特工们需要完成 |
| 480 | those commercial needs. | 这些商业需求。 |
| 481 | The providers on the other side are going to see value in that. | 另一边的供应商会从中看到价值。 |
| 482 | And there's going to be some mechanism by which it gets fleshed out. | 并会有一些机制,通过它得到充实。 |
| 483 | That's what an efficient | 这么有效率 |
| 484 | system does. | 系统有。 |
| 485 | But I think what matters is that you're building those world-class consumer | 但我认为重要的是 你正在建造这些世界级的消费者 |
| 486 | experiences. | 经历。 |
| 487 | Right. | 对 |
| 488 | And so I'm answering your question. | 所以我回答你的问题 |
| 489 | I just want to make sure. | 我只想确认一下 |
| 490 | Definitely. | 当然 |
| 491 | Yeah. | 对 |
| 492 | I promised myself one product feature request and one only. | 我答应过自己一个产品特性 只有一个 |
| 493 | Yes. | 对 |
| 494 | Here it goes. | 开始了 |
| 495 | Is it possible to make search and Gmail powered by Gemini as good as | 能否让双子座的搜索和Gmail充电 |
| 496 | search and Gmail that you access when you plug Gmail into your chat GPT interface? | 搜索和Gmail 当你插入Gmail到您的聊天 GPT 界面时您访问的邮件 ? |
| 497 | What's going | 发生什么事了 |
| 498 | on here? | 在这里? |
| 499 | I think we have one of the exciting things we announced recently is in a unified | 我觉得我们有一个令人兴奋的事情 我们最近宣布是一个统一的 |
| 500 | Gemini agent for enterprise, which we have announced and we're bringing that to consumers. | 企业的双子星代理人,我们已经宣布了, 并且我们把它带给消费者。 |
| 501 | Right. | 对 |
| 502 | So these are persistent agents, which run 24 seven in the cloud. | 因此这些是持久性的毒剂,它们在云中运行了24 7. |
| 503 | Like you're spinning up a | 就像你在旋转 |
| 504 | VM effectively for every query. | 每个查询都有效。 |
| 505 | And from the runtime, the harness, the model, it's all a | 而从运行时间, 绳子,模型, 这一切都是一个 |
| 506 | headless service. | 无头礼拜. |
| 507 | Right. | 对 |
| 508 | And it'll work everywhere. | 而能随处行. |
| 509 | And it has your context with you. | 它有你的背景与你。 |
| 510 | You can address | 您可以联系 |
| 511 | this agent, whether you're in workspace, in search, in Gemini, all of it. | 这个探员,无论你是否在工作空间, 寻找,双子座,所有的一切。 |
| 512 | So I think agents | 所以我觉得探员们 |
| 513 | are going to drive this unification from a customer experience, which people are looking for. | 将推动这个统一 从客户的经验, 人们正在寻找。 |
| 514 | Yeah. | 对 |
| 515 | And you are right. | 你说得对。 |
| 516 | When you have a broad product portfolio and you're moving forward fast, | 当你有广泛的产品组合 并且你正在向前发展, |
| 517 | there are always areas where you're moving forward faster than some other areas. | 总是有的地区 你前进的速度 比其他一些领域。 |
| 518 | That's a valid | 这是有效的 |
| 519 | feedback. | 反馈。 |
| 520 | Right. | 对 |
| 521 | But we are putting a lot of time and effort into making sure as we build out | 但是,我们正在投入大量的时间和努力 以确保我们建设 |
| 522 | our agent experience, that unification and that capability works well across all these surfaces. | 我们的特工经验,这种统一和这种能力在所有这些表面都行之有效。 |
| 523 | That's a good answer to the interface question too. | 这对界面问题也是很好的答案. |
| 524 | Yes, it is. | 没错 |
| 525 | Yeah. | 对 |
| 526 | When people have thought | 当人们想到 |
| 527 | about where Google plays next, a question comes up, you know, where's Google's Muse or its DOTS | 关于Google的下一场比赛,一个问题出现了,你知道,Google的Muse或者它的DOTS在哪里? |
| 528 | product? | 产品? |
| 529 | Those have kind of come out of nowhere recently and they've taken off. | 这些东西最近突然冒出来 他们就走了 |
| 530 | You have a version | 你有一个版本 |
| 531 | of that. | 就是这样 |
| 532 | Do you anticipate creating a new product that competes with, let's just say Muse, for | 你预期会创造出一种新产品 与之竞争吗? |
| 533 | instance? | 比如说? |
| 534 | Look, already today, so the way we have envisioned this, in fact, we describe Gemini, | 你看,已经今天, 所以我们设想的方式, 事实上,我们描述双子座, |
| 535 | our goal with Gemini app to be a personal, proactive and powerful assistant. | 我们的目标 双子座应用 是一个个人,主动和强大的助手。 |
| 536 | That's what | 就这样 |
| 537 | we are working towards. | 我们正在努力。 |
| 538 | 你在Google I. | |
| 539 | this year | 今年 |
| 540 | 和Spark一起来 | |
| 541 | But effectively, already, you know, if you open the Gemini app as a pro or an ultra user, | 但是,实际上,已经,你知道, 如果你打开双子座应用程序 作为一个职业或超用户, |
| 542 | you see chat and task and they start giving you that agent experience. | 你可以看到聊天和任务, 他们开始给你的代理经验。 |
| 543 | With Argon now, | 现在跟阿贡在一起 |
| 544 | that capability is dramatically accelerated. | 能力大大加快。 |
| 545 | And what we announced recently in our enterprise | 我们最近宣布的事业 |
| 546 | event, we are bringing that to consumers as well. | 也给消费者带来了这些。 |
| 547 | But over time, so you will soon in Gemini, | 但随着时间的推移,你很快就会在双子座, |
| 548 | you will be able to ask something. | 你会问一些问题 |
| 549 | And that will, it's a continuum. | 这将是一个连续体。 |
| 550 | You don't need to think about | 你不需要考虑 |
| 551 | whether you're asking a chat or an agent task. | 无论你要求聊天还是代理任务 |
| 552 | It'll just kind of automatically do the right thing | 它会自动做正确的事情 |
| 553 | and it can work persistently on the background securely and get things done for you. | 并且它可以坚持不懈地工作 在背景安全 并为你做事情。 |
| 554 | Already empirically, when I look at what's coming into Gemini in our app, about last year, | 从经验上看,当我看到 双子座进入我们的应用, 关于去年, |
| 555 | we would have tagged them and most of them would have been chats. | 我们本来会给它们贴上标签的 大部分都是聊天 |
| 556 | But a steadily growing | 但不断增长 |
| 557 | percentage is already agentic tasks in Gemini. | 百分比在双子座中已经是代理任务 。 |
| 558 | And so, yes, we are going to bring personal agents | 所以,是的,我们要去 带个人特工 |
| 559 | to our consumers. | 给我们的消费者。 |
| 560 | It'll work in the Gemini app. | 会在双子座应用中起作用. |
| 561 | It'll work in Google search. | 将在Google搜索中工作. |
| 562 | 将在Gmail工作。 | |
| 563 | You can address it, call it, etc. | 你可以处理,调用,等等。 |
| 564 | But we want to make sure we're getting it right. | 但我们要确定我们做得对 |
| 565 | It'll be | 这将是 |
| 566 | by Argonne. | 由阿贡. |
| 567 | And so, we are very excited to bring it to our users. | 因此,我们非常兴奋地把它带给我们的用户。 |
| 568 | I've got that pro subscription. | 我有那个专业订阅。 |
| 569 | I'm in tasks, so I'm looking forward to using it more. | 我在执行任务,所以很期待更多使用. |
| 570 | 好了,我们走 All right, | |
| 571 | a couple of random questions. | 几个随机的问题 |
| 572 | Google's invested and has committed tens of billions of dollars | Google已经投入了数百亿美元 |
| 573 | to Anthropic. | 去安特罗皮克 |
| 574 | What do you see in Dario that makes you want to have such a deep | 你在达里奥看到什么 让你想有这么深 |
| 575 | commitment and investment to him? | 对他的承诺和投资? |
| 576 | Very early on, when we looked at the | 很久以前,当我们看着 |
| 577 | Anthropic的团队, 我们很清楚。 | |
| 578 | This was a world-class team which was involved in | 这是一个世界级的球队,参与 |
| 579 | some of the earliest GPD models. | 一些最早的GPD模型. |
| 580 | And so, we were quickly able to assess that they would be at the | 因此,我们很快能够评估 他们将会在 |
| 581 | frontier. | 边疆. |
| 582 | And so, we've always tried to identify and partner with companies which would be at the | 因此,我们一直试图 识别和合作伙伴 公司将会在 |
| 583 | cutting edge of technology. | 技术的前沿。 |
| 584 | 我们用SpaceX做了 | |
| 585 | 我们已经用Anthropic做了。 | |
| 586 | That's what we saw, and we had a chance to partner. | 这就是我们所看到的, 我们有机会合作。 |
| 587 | Some of the partnerships turned in | 一些伙伴关系已经投入使用 |
| 588 | the construct of our TPUs as well. | 建造我们的TPU。 |
| 589 | So, I think that's how we approached it. | 所以,我认为这就是我们如何接近它。 |
| 590 | With Apple, you had a partnership with them to license Gemini to them to help improve Siri. | 你和苹果公司合作 给双子座许可 帮助他们改进Siri |
| 591 | Is that partnership expanding, contracting, standing still? | 这种伙伴关系是否在扩大、缩小、维持不变? |
| 592 | I mean, for both us and Apple, that partnership was about translating the capabilities of AI into | 我的意思是,对我们和苹果公司来说, 伙伴关系是将AI的能力转化为 |
| 593 | better experiences for users. | 为用户提供更好的经验。 |
| 594 | I thought Apple has done a phenomenal job of incorporating. | 我以为苹果公司已经做了一个惊人的整合工作。 |
| 595 | They customized these models and built it into Siri, and it's been great to see the reception. | 他们定制了这些模型,并将其建在Siri中,看到这个接待会真好. |
| 596 | And so, we are committed to doing more. | 因此,我们致力于做更多的工作。 |
| 597 | You're going to do more with them? | 你要对他们做更多吗? |
| 598 | There's nothing specific to comment, but it's a partnership that's working well. | 虽然没有具体的评论,但合作效果很好. |
| 599 | Datacenters in space, you've announced that you're doing more there. | 数据中心在太空中,你已经宣布 你在做更多的。 |
| 600 | What have you learned so far | 你迄今学到了什么 |
| 601 | about the feasibility of that project? | 那项目的可行性呢? |
| 602 | It's a very exciting project. | 这是一个非常激动人心的项目。 |
| 603 | Projects | 项目 |
| 604 | like this remind me. | 像这样提醒我 |
| 605 | I still recall that moment in the desert for the DARPA Grand Challenge, | 我仍然记得 在沙漠中为DARPA大挑战, |
| 606 | watching the Stanley car compete in it, which ended up being the basis for Waymo. | 看着斯坦利汽车在其中竞争,最终成为了Waymo的基础. |
| 607 | You're always, with all these technologies, you're at a moment in time where it looks a | 你总是,用所有这些技术, 你在一个时刻 它看起来像一个 |
| 608 | bit like sci-fi, but 10 to 20 years down the line, it's actually working in a meaningful way. | 有点像科幻, 但10到20年的线下, 它实际上以一种有意义的方式工作。 |
| 609 | I feel the same at space. | 我在太空里也一样 |
| 610 | We've always believed in betting big on these technologies. | 我们一直相信这些技术是大赌注 |
| 611 | Quantum has | 量子已经 |
| 612 | been a great example of it. | 是一个很好的例子。 |
| 613 | We're probably 10 years into the journey. | 我们大概10年的旅程。 |
| 614 | Waymo, we are about | 韦莫,我们是 |
| 615 | almost 20 years into the journey. | 将近20年的旅程。 |
| 616 | Space, we are two to three years into the journey. | 太空,我们两到三年的旅程。 |
| 617 | It's clear to me that it's an approach which will make a lot of sense in the future, | 很明显,这是一个方法 将来会有很多意义, |
| 618 | being in sun-synchronous orbits with 24-7 access to the most abundant form of energy we have access | 处于太阳同步轨道,24-7 获得我们获得的最丰富的能量 |
| 619 | to, will make a lot of sense for this exponential increase in compute that we are going to need. | 对于我们所需要的计算成倍增加,这将非常合理。 |
| 620 | Having said that, this is a challenging project. | 尽管如此,这是一个具有挑战性的项目。 |
| 621 | The team has disciplined milestones. | 团队有纪律性的里程碑。 |
| 622 | It was a | 这是一个 |
| 623 | super exciting milestone to send the TPUs up to space. | 超令人兴奋的里程碑 将TPU送上太空。 |
| 624 | Sometime as we are speaking, we are | 当我们说话的时候,我们是 |
| 625 | expecting a hello world moment from them. | 等待着他们世界的问候 |
| 626 | I look forward to seeing that happen. | 我期待着看到这种情况发生。 |
| 627 | We'll continue | 我们继续 |
| 628 | investing here too. | 也在这里投资。 |
| 629 | You recently, maybe last year, you said that 25% of code written within | 你最近,也许去年, 你说25%的代码写在 |
| 630 | Google was written by AI. | 谷歌由AI编写. |
| 631 | What's the number today? | 今天号码是多少? |
| 632 | I think we have almost. | 我想我们几乎有。 |
| 633 | That number | 那个号码 |
| 634 | in our world today is meaningless in the sense that every part of code we now touch is influenced | 在当今世界上,我们所触及的代码的每一部分都受到了影响,这毫无意义。 |
| 635 | by AI deeply. | 由大赦国际进行深入调查。 |
| 636 | What we are tracking with Argonne, for example, we are now able to see. | 例如,我们正在追踪Argonne的情况,我们现在可以看到。 |
| 637 | It's been | 已经是 |
| 638 | point of inflection. | 点的不适。 |
| 639 | People are using Argonne to really go tackle complex projects, including | 人们利用Argonne 去解决复杂的项目,包括 |
| 640 | projects which affects millions of lines of code. | 影响数百万行代码的项目。 |
| 641 | We have definitely, in the four weeks since we have | 我们肯定,在四个星期后 |
| 642 | launched Argonne, we have seen a 30% increase in CLs and, more importantly, almost a similar increase, | 阿尔贡发射,我们已经看到 30%的CL增加,更重要的是, 几乎类似的增加, |
| 643 | one third decrease in rollbacks. | 翻转减少三分之一。 |
| 644 | I think people are writing code more productively and better | 我觉得人们写代码 更有成果和更好 |
| 645 | code with it. | 密码 |
| 646 | We are tracking it with newer metrics now. | 我们正在用新的度量衡来追踪它。 |
| 647 | Talking about the opposition to | 谈论反对 |
| 648 | data centers, how do you feel that? | 数据中心 你觉得怎么样? |
| 649 | If you can't build data centers in the U.S. the way that you | 如果你不能像你们那样在美国建立数据中心 |
| 650 | want to, where else do you go? | 想,你还能去哪? |
| 651 | We are very committed to building our data centers very | 我们非常致力于建设我们的数据中心 |
| 652 | responsibly. | 负责任的。 |
| 653 | When I look at our largest data centers in some of the communities where we have | 当我看看我们最大的数据中心 在一些社区 我们拥有 |
| 654 | been the longest, I visit them. | 时间最长,我拜访他们。 |
| 655 | I talk to people. | 我和别人说话。 |
| 656 | There are people who would come and tell us their | 有人会来告诉我们他们 |
| 657 | kids have gone to college because of our investments in these communities. | 孩子们上大学 因为我们在这些社区的投资。 |
| 658 | We have grown in those | 我们已经成长了 |
| 659 | communities. | 社区。 |
| 660 | I think, to me, this is about doing it right, being engaging with the communities | 我认为,对我来说,这是 做正确,与社区接触 |
| 661 | early, doing it transparently. | 早点做,透明。 |
| 662 | We are committed to making sure we are bringing incremental power, | 我们致力于确保我们带来增量的力量, |
| 663 | and we cover 100% the cost of our power needs for these projects. | 我们百分之百地支付 我们的电力需求 这些项目。 |
| 664 | It's understanding the impact on | 它明白影响 |
| 665 | neighborhoods and making sure you're working early to address them, etc. | 社区,确保你工作 早期解决他们,等等。 |
| 666 | I think done | 说完了 |
| 667 | rightly. | 对 |
| 668 | I think you can continue making progress. | 我想你可以继续取得进展。 |
| 669 | Obviously, we are constantly taking | 很明显,我们一直在服用 |
| 670 | feedback into account and so on. | 反馈入计等. |
| 671 | I do worry not everyone operates the same way. | 我担心的不是每个人都一样 |
| 672 | People have been | 人们已经 |
| 673 | surprised by data centers showing up in their backyard. | 数据中心在后院露面 令人惊讶 |
| 674 | I think those are all important to, | 我认为这些都很重要, |
| 675 | as an industry, we should continue to do better. | 作为一个行业,我们应当继续做得更好。 |
| 676 | But part of the reason, in a 20-year time frame, | 但部分原因,在一个20年的时间范围内, |
| 677 | thinking about things like data centers and space, is making sure we are able to put these | 对数据中心和空间的思考 正在确保我们能够把这些 |
| 678 | data centers where maybe the tradeoffs will be different. | 数据中心,也许交易会有所不同。 |
| 679 | Is there an unproven discipline | 是否有没有证明的纪律 |
| 680 | in AI right now that Google is working hard to crack? | 现在Google正在努力破解? |
| 681 | People have talked about continuous | 人们谈论的是连续 |
| 682 | learning, getting these AIs to continue to update after they get released as opposed to | 学习,让这些AI在发布后继续更新,而不是 |
| 683 | forget through every session. | 忘记每场比赛 |
| 684 | Look, there are multiple parallel bets. | 你看,有多个平行的赌注。 |
| 685 | You saw with Argonne, | 你看到阿尔贡, |
| 686 | a million output tokens. | 100万个输出符号。 |
| 687 | With our Gemini models early on, we had a one million context window | 我们的双子座模型很早就有了100万个背景窗口 |
| 688 | on the input side. | 在输入侧。 |
| 689 | These are all examples of breakthroughs we are constantly working on. | 这些都是我们不断在努力取得突破的例子。 |
| 690 | How do you think about synthetic data? | 你觉得合成数据怎么样? |
| 691 | How do you think about continual learning? | 你觉得继续学习怎么样? |
| 692 | There are more efficient architectures. | 还有效率更高的建筑. |
| 693 | We do dedicate a meaningful percentage of our | 我们确实投入了有意义的百分比 |
| 694 | compute to these research bets too. | 计算这些研究赌注。 |
| 695 | It always goes hand in hand, including our world models, | 它总是携手并进, 包括我们的世界模式, |
| 696 | etc. | 页:1 |
| 697 | Those are all bets we are placing on. | 这些都是我们打的赌 |
| 698 | Sometimes as part of these bets, there are | 有时候,作为这些赌注的一部分, |
| 699 | ideas that emerge that just make their way into the mainline models. | 想法的出现,只是让他们进入主线模型。 |
| 700 | Maybe you don't | 也许你不会 |
| 701 | fully talk about it, but I think there are multiple bets in these areas. | 完全谈论它, 但我认为这些方面有多重赌注。 |
| 702 | I want to end with this. | 我想结束这个。 |
| 703 | To be involved in the tech industry the way that you are, | 要像你一样参与技术产业 |
| 704 | there is a certain inherent optimism to it. | 它有着某种固有的乐观。 |
| 705 | A belief that if you build, you will improve the | 一个信念,如果你建造, 你会改善 |
| 706 | industry. | 工业。 |
| 707 | Now, there are sides to that. | 现在,有的方面。 |
| 708 | It often requires betting a lot and hoping that the thing | 这往往需要赌注 并希望的东西 |
| 709 | that you are working on is going to come to fruition in the way that you want it to be, | 你正在做的 将会以你想要的方式取得成果, |
| 710 | but it doesn't always work that way. | 但并不总是这样 |
| 711 | I'll give you an example. | 举个例子 |
| 712 | I was just visiting El Salvador. | 我刚去萨尔瓦多 |
| 713 | I went in a five-year bracket and I saw in this one town just an amazing amount of enthusiasm for | 我在一个镇上看到一个令人惊叹的热情 |
| 714 | Bitcoin and the blockchain. | 比特币和块链。 |
| 715 | There was construction happening and it was booming, and then it didn't | 建筑正在兴旺,后来没有 |
| 716 | happen the way that folks expected. | 和人们预期的一样 |
| 717 | This visit was basically a ghost town with a lot of unfinished | 这次访问基本上是一个鬼城 有很多未完成的工作 |
| 718 | construction projects. | 建设项目. |
| 719 | I think that to do what you do, it requires a type of optimism that says it's | 我认为,为了做你做的事, 它需要一种乐观类型, 说它是 |
| 720 | worth going for even if the bets don't pan out. | 即使赌注没完也值得去 |
| 721 | There could be consequences there. | 可能会有后果 |
| 722 | I'd just love | 我很喜欢 |
| 723 | hear your perspective. | 听听你的看法 |
| 724 | You spoke about this at your Stanford commencement address about the need | 你在斯坦福大学的演讲稿上 谈到过这个问题 |
| 725 | for optimism, but it's a nuanced form of optimism. | 但这是一种细微的乐观 |
| 726 | I'd love to hear your perspective on how you | 我很想听听你的看法 |
| 727 | maintain that. | 保持这一点。 |
| 728 | Is it pure optimism or is it a couched optimism in a way? | 这是纯粹的乐观还是某种意义上的乐观? |
| 729 | I genuinely have | 我确实有 |
| 730 | a deep conviction that improving technology and improving access to technology makes | 坚信改进技术和增加获得技术的机会, |
| 731 | people's lives better. | 人们的生活更好。 |
| 732 | It's been my personal experience in my life, out of which comes that | 这是我一生中的个人经历,从中而来 |
| 733 | deep conviction optimism. | 坚定的乐观 |
| 734 | With that, I've always felt humanity's journey is always about working | 我总是觉得人类的旅程总是工作 |
| 735 | hard to harness the technology to make it work well for society. | 很难利用技术 使它为社会良好工作。 |
| 736 | The optimism is not an optimism | 乐观不是乐观 |
| 737 | that you just build whatever you want. | 你只要创造任何你想要的。 |
| 738 | You work hard to make sure you're building technology | 你努力工作,确保你建设技术 |
| 739 | in the service of humanity to make the human condition better. | 帮助人类改善人类状况。 |
| 740 | That's got to be the defining | 这一定是定义 |
| 741 | goal as we are approaching it. | 我们正接近目标 |
| 742 | With that framework, I'm very, very optimistic. | 有了这个框架,我很乐观。 |
| 743 | I think we are in very early stages of the kind of progress that we are going to see | 我认为我们正处于我们将要看到的那种进展的初期阶段。 |
| 744 | on many areas. | 在许多领域。 |
| 745 | We're all talking about areas like healthcare and stuff, but I think a lot of the | 我们都在谈论诸如医疗保健之类的领域, 但我认为很多 |
| 746 | make things in a fundamentally cheaper way on many, many things where things have been getting | 在很多事情上 以更便宜的方式做一些事情 |
| 747 | expensive for a long, long time, discovering new materials, et cetera, which really bend the curve. | 价格昂贵,很长一段时间, 发现了新的材料,等等, 这些材料真的弯曲。 |
| 748 | I think it's tough to step today and understand the impact of all that, but they're going to be | 我认为今天很难一步步理解这一切的影响,但他们会是 |
| 749 | very profound. | 相当深刻。 |
| 750 | I'm very optimistic, but we have to be rightfully concerned about all the ways this | 我很乐观,但我们必须正确 关心所有的方式 |
| 751 | can go wrong. | 会出错的 |
| 752 | That's how you make sure you solve those concerns. | 这就是你如何确保 你解决这些担忧。 |
| 753 | I think they both go hand in hand. | 我想他们俩是携手并进的 |
| 754 | Sundar, I know you don't do many of these. | 孙达,我知道你不做很多这些。 |
| 755 | I really appreciate you spending the time with | 我真的很感激你花时间 |
| 756 | me today. | 我今为我. |
| 757 | It was great to get a chance to see you and speak with you again. | 很高兴有机会再见到你 和你说话 |
| 758 | Thanks, Alex. | 谢谢 亚历克斯。 |
| 759 | I appreciate it. | 谢谢 |
| 760 | Thanks. | 谢谢 |
| 761 | Hi, everybody. | 大家好,你们好 |
| 762 | Thanks so much for listening and watching. | 非常感谢你的聆听和观看。 |
| 763 | We'll see you next time on Big Technology Podcast. | 下次在大科技节目上再见 |
该视频共有字幕 763 条。解锁更多字幕为会员功能，请移动到 价格