黄仁勋最新断言:别再单打独斗了!未来企业必须靠“开放智能体联
字幕摘录
| 时间 | 英文 | 中文 |
|---|---|---|
| 0:11 | Excited to be here with Jensen. | 很高兴和Jensen在一起 |
| 0:13 | There has been a ton of advancements in AI and agents over the past year, but few months | 在过去的一年里,大赦国际和特工机构取得了大量进步,但几个月 |
| 0:18 | in particular, I feel. | 特别是,我觉得。 |
| 0:19 | We've seen a lot of these advancements come in the form of better performance, but at | 我们已经看到很多这些进步 表现得更好,但是... |
| 0:24 | the same time, we've also seen that openness and control and trust in a lot of these models | 同时,我们也看到 开放、控制和信任 很多这些模型 |
| 0:30 | and systems around them has become more and more important. | 周围的系统变得越来越重要。 |
| 0:34 | So the first thing I want to start with is how and why are you guys at NVIDIA investing | 首先我想说的是 你们在NVIDIA投资的方式和原因 |
| 0:40 | in an open agent ecosystem and stack? | 在开放的代理生态系统和堆积? |
| 0:44 | Well, first, before I answer, I want to congratulate you for all the work that you do. | 首先,在我回答之前,我要祝贺你所做的一切工作。 |
| 0:49 | In fact, if you look at the last six months, we could both agree that although we've been | 事实上,如果你看看过去的六个月, 我们可以都同意,虽然我们已经 |
展开字幕全文(452 条)
| 序号 | 英文 | 中文 |
|---|---|---|
| 1 | Excited to be here with Jensen. | 很高兴和Jensen在一起 |
| 2 | There has been a ton of advancements in AI and agents over the past year, but few months | 在过去的一年里,大赦国际和特工机构取得了大量进步,但几个月 |
| 3 | in particular, I feel. | 特别是,我觉得。 |
| 4 | We've seen a lot of these advancements come in the form of better performance, but at | 我们已经看到很多这些进步 表现得更好,但是... |
| 5 | the same time, we've also seen that openness and control and trust in a lot of these models | 同时,我们也看到 开放、控制和信任 很多这些模型 |
| 6 | and systems around them has become more and more important. | 周围的系统变得越来越重要。 |
| 7 | So the first thing I want to start with is how and why are you guys at NVIDIA investing | 首先我想说的是 你们在NVIDIA投资的方式和原因 |
| 8 | in an open agent ecosystem and stack? | 在开放的代理生态系统和堆积? |
| 9 | Well, first, before I answer, I want to congratulate you for all the work that you do. | 首先,在我回答之前,我要祝贺你所做的一切工作。 |
| 10 | In fact, if you look at the last six months, we could both agree that although we've been | 事实上,如果你看看过去的六个月, 我们可以都同意,虽然我们已经 |
| 11 | working in AI for 15 years, the last six months changed everything. | 在大赦国际工作了15年 过去6个月改变了一切 |
| 12 | And so all of the technology, all the large language model advances, all the scaling, | 所以所有的科技,所有的大型语言模型 进步,所有的缩放, |
| 13 | all of the breakthroughs, all the Omni models, multi-modality stuff, all that stuff is fantastic. | 所有的突破,所有的Omni模型, 多式联运的东西, 所有的东西是惊人的。 |
| 14 | But in the end, it was the last six months that where everything kind of came together. | 但最后是过去6个月, 所有的东西都走到了一起。 |
| 15 | And now finally, AI is useful. | 现在终于,AI有用了. |
| 16 | And when AI is useful, every company in the world, every enterprise in the world wants | 当AI有用时 世界上每一个公司 世界上每一个企业都想要 |
| 17 | to get their hands on it. | 以获得他们的手。 |
| 18 | And now the question is how? | 现在的问题是如何? |
| 19 | This is where Lang chain comes in. | 兰连锁店就在这里 |
| 20 | And so you always had a vision that the large language model was the essential ingredient, | 所以,你一直有一个愿景, 大语言模型是基本成分, |
| 21 | the essential enabling technology. | 关键辅助技术。 |
| 22 | But in order to turn it into a useful product, you have to surround it with what is now known | 但是为了把它变成有用的产品 你必须用现在已知的东西来包围它 |
| 23 | as a harness. | 做为绳索 |
| 24 | There's so much more. | 还有很多 |
| 25 | So much more. | 还有这么多 |
| 26 | And back in the old days, we used Lang chain to help us turn a large language model into | 回到过去,我们用Lang链 帮助我们把一个大语言模型变成 |
| 27 | a promptable API. | 一个快速的API。 |
| 28 | And we would turn, we use Lang chain to build our rags and we use Lang chain to step by | 然后我们转身,我们用Lang链 来建造我们的破布,我们用Lang链 步入 |
| 29 | step by step, which led to today's agents. | 一步一步,这导致了今天的特工。 |
| 30 | And really what happened in the last six months, the big breakthrough are these agentic systems | 其实在过去六个月里发生的事情 最大的突破是这些代理系统 |
| 31 | that are grounded on knowledge that can use tools to do search and has memory that it | 以可以使用工具进行搜索并具有内存的知识为基础 |
| 32 | manages and it has safeguards and has the ability to iterate until it gets the job done. | 管理并有保障措施,并有能力延展,直到完成任务。 |
| 33 | But it ultimately needed some models that have reached a level of capability where everything | 但它最终需要一些模型 已经达到一个水平的能力 在那里,一切 |
| 34 | comes together into that flash point. | 一起进入闪光点 |
| 35 | And that's where cloud code really kind of brought the imagination of agentic systems, | 这就是云码真正带来 代理系统想象力的地方 |
| 36 | OpenClaw, of course, was a big deal and all the work that you did, right, with deep agents | OpenClaw当然是个大问题 和你所做的所有工作 对 深层特工 |
| 37 | and we use that ourselves. | 我们用自己。 |
| 38 | And all of that kind of came together and bam, here we are with agentic systems. | 所有这种东西都聚集在一起,砰,我们有了代理系统。 |
| 39 | The reason why we do it, we've dedicated ourselves for many, many years to build open systems. | 我们这样做的原因是,我们多年来致力于建设开放的系统。 |
| 40 | And the reason for that is because ultimately, AI is a fundamental technology. | 其原因是,最终,AI是一种基本技术. |
| 41 | It can only be useful if applied in a whole bunch of different use cases. | 只有在一整批不同的使用案例中应用才能有用. |
| 42 | Now, of course, the first use case is just language and cognitive intelligence and that's | 现在,当然,第一个使用案例 仅仅是语言和认知智能,这就是 |
| 43 | very important, of course. | 当然很重要 |
| 44 | But we imagine a world where scientists and digital biologists and designers and roboticists | 但是,我们想象的世界 科学家和数字生物学家 设计师和机器人学家 |
| 45 | and students and researchers, enterprise IT, all of us could use agentic systems, AIs, | 和学生和研究人员,企业IT, 我们所有人都可以使用代理系统,AI, |
| 46 | to solve domain-specific problems. | 用于解决特定域的问题。 |
| 47 | And many of the problems that we want to solve, either we have specialized domain knowledge | 我们想要解决的许多问题 要么我们有专业领域的知识 |
| 48 | that is just simply not available outside, that we have to embed into, imbue into our AI, | 我们不得不嵌入 嵌入我们的AI, |
| 49 | or it's because we believe that AI becomes ultimately great, become a super agent, when | 或者是因为我们相信AI最终会变得伟大 成为超级特工 |
| 50 | we put it into a flywheel where we use it, it gets smarter, it becomes more useful, we | 我们把它放在飞轮上 用它,它变得更聪明,它变得更有用,我们 |
| 51 | get smarter, kind of like us, kind of like humans, learns over time. | 变得更聪明,有点像我们, 有点像人类, 学习随着时间的推移。 |
| 52 | And so we kind of imagine this future where AI has a foundation and the work that Anthropic | 我们想象着这个未来 AI有一个基础 还有Anthropic的作品 |
| 53 | OpenAI和Google都做得很棒 | |
| 54 | But there's specialized AIs and domain-specific AIs and proprietary AIs that people want to | 但有专门的AI和专有域的AI和专有AI是人们想要的. |
| 55 | build and we want to enable that world to happen. | 我们想让这个世界得以实现。 |
| 56 | Maybe digging into that for a second, on this topic of specialization, how exactly do you | 也许研究一下, 关于这个专业话题,你到底如何 |
| 57 | think it's best to specialize these systems? | 最好专门使用这些系统? |
| 58 | Is it going to be purely the model, is it going to be the harness as well, the context | 会不会是纯粹的模型 会不会是绳子 环境 |
| 59 | outside? | 外面吗? |
| 60 | Like what goes into the specialization? | 像什么进入专业? |
| 61 | Well, the specialization starts with you need to have intelligence that's good enough. | 好吧,专业的开始,你需要 拥有足够的智能。 |
| 62 | And this is why we worked on Nemotron and really love the fact that you're part of the | 这就是为什么我们工作 在尼莫特隆 并真正喜欢的事实 你的一部分 |
| 63 | founding team of Nemotron Coalition. | 尼莫特龙联盟创始团队. |
| 64 | 我们让Nemotron Ultra非常不可思议。 | |
| 65 | Now Nemotron Ultra is a great model as a start, but it becomes an incredible model when you | 现在Nemotron Ultra是一个伟大的模型 作为开始,但它成为一个不可思议的模型 当你 |
| 66 | put the Lang chain framework around it, the Lang chain harness around it, so that you | 把连锁框架放在周围 连锁绳套在周围 |
| 67 | ground it on information that is domain-specific and a person is an intelligent person becomes | 以特定领域的信息为基础 一个人是个聪明人 |
| 68 | super useful when we give them access to particularly important information. | 当我们让他们获得特别重要的信息时 是非常有用的 |
| 69 | And so access to information is important. | 所以获得信息很重要。 |
| 70 | Putting it into a flywheel where maybe you're even training the model, post-training the | 把它装到飞轮上 在那里你甚至正在训练模型 训练后 |
| 71 | model inside the Lang chain harness against the harness so that the model becomes good | 兰链内模型与绳索对齐 使模型变得良好 |
| 72 | at applying the harness around it. | 在它周围应用绳索。 |
| 73 | What you want it to do for that task. | 你希望它为这个任务做什么 |
| 74 | What you want it to do. | 随其所欲. |
| 75 | Yeah. | 对 |
| 76 | And so I think that this moment has now arrived, but we need an open harnessing system that | 所以我认为这一刻已经到来 但我们需要一个开放的驾驭系统 |
| 77 | we can build ourselves, that we can apply, and then of course improve against over time. | 我们可以建立自己, 我们可以应用, 然后当然改善 与时间。 |
| 78 | I love what you said about the model being good enough. | 我喜欢你说的模特够好 |
| 79 | I feel like that threshold was crossed, I don't know, maybe a year ago by some of the | 我觉得那个门槛已经越过了 我不知道 也许一年前 |
| 80 | frontier models, six months ago by some of the open weight models. | 前沿模型,半年前 一些开放重量模型。 |
| 81 | You talked about Nemotron 3 Ultra. | 你说过尼莫特龙3号超人 |
| 82 | We've done a lot of work with that to make that really good in deep agents. | 我们为此做了很多工作, 使这在深层特工中是很好的。 |
| 83 | Some of the things we did is tweak the harness to make it best for this model because we | 我们所做的一些事情是 调整绳子 使其最好的 这个模型,因为我们 |
| 84 | found that different models need different prompts, different tools. | 发现不同的模型需要不同的提示,不同的工具. |
| 85 | And with that tweaking, we managed to get Nemotron 3 Ultra in deep agents. | 有了这个变化, 我们设法让Nemotron 3 Ultra进入了深层特工。 |
| 86 | We have an internal benchmark and managed to get up to like 86% on that, Claude Opus | 我们有一个内部基准 并设法上升到86% 在那,克劳德·奥普斯 |
| 87 | for comparisons at 87. | 比较87。 |
| 88 | 里面有DeepSeq和Minimax的模型 大约82,83 | |
| 89 | So we're starting to see that some of the more recent open weight models are really | 因此,我们开始看到一些 最近的开放重量模型 真的 |
| 90 | getting to frontier performance. | 进入前沿表演。 |
| 91 | I know. | 我知道,我记得 |
| 92 | I am so proud. | 我很骄傲 |
| 93 | It is so incredible. | 太不可思议了 |
| 94 | Thank you. | 谢谢 |
| 95 | It is so incredible. | 太不可思议了 |
| 96 | One of the just as important thing is it's 10 times as cheap as Opus. | 同样重要的一件事是 它比奥普斯便宜十倍 |
| 97 | And I think open weight models are starting to really strike a good balance between performance | 我觉得开放式体重模型 正在真正在性能之间 取得良好的平衡 |
| 98 | and cost. | 和成本。 |
| 99 | So I'd be curious how you see this cost part changing the equation for builders. | 所以我想知道你是如何看待这个成本部分 改变建造者的方程式的。 |
| 100 | The benefit of cost, it comes in a couple of different ways. | 成本的好处, 它有几种不同的方式。 |
| 101 | I happen to think that when you have cost effective intelligence, people just use more | 我碰巧认为,当你有 成本效益高的智能, 人们只是使用更多 |
| 102 | of it. | 对 |
| 103 | When you have a cost effective agent, then you can iterate across a larger search space. | 当你有一个成本效益高的代理, 然后你可以跨越一个更大的搜索空间。 |
| 104 | And as a result, the answer could actually be better. | 结果,答案实际上可能更好。 |
| 105 | And in the case of Nemotron, it's cost effective because it's so fast. | 而对于尼莫特龙来说,由于速度如此之快,所以成本效率很高. |
| 106 | It's so computationally efficient. | 计算效率很高 |
| 107 | When it's computationally efficient, it could explore larger spaces. | 在计算效率时,它可以探索更大的空间. |
| 108 | And it's no different than when somebody can think fast, you could explore more space. | 和人们能快速思考的时候一样 你可以探索更多的空间 |
| 109 | When you can try things more quickly, you can find a better answer. | 当你能更快地尝试事物,你可以找到更好的答案. |
| 110 | And so this is the incredible benefit of Nemotron 3 Ultra inside the Lang chain framework and | 因此,这是尼莫特龙3号超超强星在朗链框架内 令人难以置信的好处 |
| 111 | the Lang chain harness inside DeepAgent. | 朗链系在深潜特工体内 |
| 112 | It could think so quickly, it could explore so quickly. | 它能迅速思考,能迅速探索. |
| 113 | It could iterate so quickly and efficiently that it's going to find better answers. | 它能快速高效地移动 从而找到更好的答案 |
| 114 | And so I'm just really, really excited that we created a model that was near the frontier. | 所以我非常非常非常兴奋 我们创造了一个靠近前沿的模型。 |
| 115 | But adapting the environment around Nemotron, you made it deliver frontier capabilities. | 但是适应尼莫特龙周围的环境, 你让它提供了前沿能力。 |
| 116 | Now insight, you know, for humans, it's kind of the same. | 现在的洞察力,对于人类来说,是一样的。 |
| 117 | Of course, we like to hire the smartest people in the world. | 当然,我们喜欢雇用世界上最聪明的人。 |
| 118 | But beyond that, we also give them access to tools, we give them access to information. | 但除此之外,我们还让他们获得工具,让他们获得信息。 |
| 119 | And we also create the world around them so that we, you know, we allow them, enable them | 我们还创造周围的世界 这样我们,你知道,我们允许他们,让他们 |
| 120 | to create the conditions for them to achieve their full potential. | 创造条件,使他们充分发挥潜力。 |
| 121 | And so you adjust the environment, not just the model. | 所以你要调整环境, 而不仅仅是模型。 |
| 122 | And this is where Lang chain came in. | 这是Lang连锁店的入口 |
| 123 | What you said about using more of the intelligence as it's cheaper and faster. | 你说的多用一些智能 因为它更便宜更快 |
| 124 | We see that to be so true. | 我们认为这是事实。 |
| 125 | I think one of the things that I like to think I'm AI forward, one of the things that I've | 我觉得我喜欢的东西之一 认为我是AI向前, 我曾经的东西之一 |
| 126 | estimated is just the demand for intelligence and for tokens and how big and massive that | 估计只是对情报和信物的需求 以及数量和数量 |
| 127 | market is. | 市场是。 |
| 128 | And I think that's become especially true recently with these models getting good and | 我觉得最近这些模型越来越好 |
| 129 | being really fast and really cheap. | 很快而且非常便宜。 |
| 130 | How should we think about using frontier models? | 我们该如何考虑使用前沿模型? |
| 131 | Should we just use these open source models all the time? | 我们是否应该一直使用这些开源模型? |
| 132 | Is there is there a time and place for both? | 二者有无时有处. |
| 133 | Well, the frontier models are getting better all the time. | 边疆模型越来越好 |
| 134 | And I fully expect the frontier models to be unbelievably good. | 我完全期望边疆模型会非常优秀。 |
| 135 | And they still have a long, long runway of improving the models that scaling scaling | 而且他们还有一条长长的跑道 来改进那些缩放模型 |
| 136 | laws, of course, are going to sustain their harnesses are improving all the time. | 当然,法律将维持他们的优势 不断改善。 |
| 137 | Their technology for dealing with memory, whether it's working memory or long term | 他们处理记忆的技术,无论是工作记忆还是长期记忆 |
| 138 | memory is advancing incredibly quickly. | 记忆在快速前进 |
| 139 | And the compaction technologies, they're all of the advancements in retrieval, | 而压缩技术,它们都是 检索的进步, |
| 140 | retrieval augmented generation and knowledge graphs. | 检索增加了生成和知识图表。 |
| 141 | There's still a lot of incredible advances that are being implemented into these | 还有很多令人难以置信的进步 正在落实到这些 |
| 142 | frontier model APIs. | 前沿模型 API. |
| 143 | The thing that the way I think about it is, frankly, I always start all of my work | 老实说,我总是开始所有的工作 |
| 144 | starting with the frontier. | 从边境开始 |
| 145 | And the reason for that is because because it's useful. | 其原因是因为它有用。 |
| 146 | I know what's the potential cost a little bit more money, but it's incredibly | 我知道什么是潜在的 花费多一点钱,但它令人难以置信 |
| 147 | my time to getting to work done is fast. | 我工作的时间很快 |
| 148 | However, you know, over time, I find that I want to add sub agents to them. | 然而,你知道,一段时间后,我发现我想在他们中增加副特工。 |
| 149 | I want to connect sub agents that are super agents at certain skills. | 我想联系那些具有某种技能的超级特工 |
| 150 | And so we have, you know, we have optimization problems inside our company that | 所以我们有,你知道, 我们公司内部的优化问题, |
| 151 | that relates to supply chain. | 这与供应链有关。 |
| 152 | Maybe it's related to chip design optimization, floor planning optimization. | 可能与芯片设计优化,地板规划优化有关. |
| 153 | And these problems, these optimization problems are insanely hard. | 这些问题,这些优化问题都非常困难。 |
| 154 | And so you're not going to just have an a general AI go off and crunch on it and | 所以,你不会让 一个将军AI去 并压缩它, |
| 155 | think that you're going to find a great answer. | 认为你会找到一个伟大的答案。 |
| 156 | We create super sub agents and these super up sub agents we would create with | 我们创建超级子代理 和这些超级高级子代理 我们会与 |
| 157 | deep agents, Lang chain deep agents with Neymotron 3 inside. | 深层探员 朗把深层探员和内森三号连在一起 |
| 158 | And we'll even connect them to specialized tools. | 我们甚至会把他们和专用工具联系起来。 |
| 159 | And that thing is built for one job. | 那东西是为一份工作而建造的。 |
| 160 | That super agent is not trying to, you know, book me travel, travel appointments. | 那个超级经纪人并不想,你知道, 订我的旅行,旅行预约。 |
| 161 | It's just trying to optimize our supply chain. | 它只是想优化我们的供应链。 |
| 162 | And in that case, I really do need to have Lang chain. | 这样的话 我真的需要Lang的锁链 |
| 163 | 我真的需要有Neymotron 3 Ultra。 | |
| 164 | And I connected to a lot of proprietary knowledge and proprietary skills. | 我和很多专有知识和专有技能联系在一起。 |
| 165 | I've got a whole team who's just dedicated to refining that. | 我有一队人 致力于完善它。 |
| 166 | Now, I think that that defines a company. | 现在,我认为这定义了公司。 |
| 167 | A company is really about a collection of a whole bunch of these super, super | 公司其实是一大堆超级超人 |
| 168 | proprietary, super important workflows. | 专有,超重要的工作流程. |
| 169 | And now we can have Lang chain with deep agent and ultra, you know, Neymotron 3 | 现在我们可以让Lang 带着深层特工和超超能力特工 |
| 170 | inside and it gives them all of the control they need. | 里面有他们需要的所有控制 |
| 171 | Super efficient access to incredible tools. | 超级高效的进入 不可思议的工具。 |
| 172 | You know, that's the future. | 你知道,这就是未来。 |
| 173 | Do you have any advice for enterprises if they're following your practice of | 你对企业有什么建议吗 如果他们遵循你的做法 |
| 174 | starting with the frontier and then starting to specialize? | 从边境开始,然后开始专攻? |
| 175 | When should they think about specializing? | 他们什么时候该考虑专业? |
| 176 | What are some triggers that you look for that? | 你在找什么触发器? |
| 177 | As soon as it gets good enough, you know? | 一旦它变得足够好,你知道吗? |
| 178 | So I would start with cloud code and codecs and use it for as long as I can. | 所以我从云码和解码器开始 尽可能长时间使用 |
| 179 | And frankly, for a lot of things, you never have to replace, you know, | 坦白地说,对于很多事情, 你永远不用替换,你知道, |
| 180 | because they're getting better all the time. | 因为他们越来越好 所有的时间。 |
| 181 | And they're on a trajectory that's going to bring capabilities insanely fast. | 并且它们正在一个轨迹上, 将带来能力 疯狂的快速。 |
| 182 | And so I think that in the future, just like companies are today, | 所以我认为,在未来, 就像今天的公司, |
| 183 | we have employees that we hire because their domain specialization | 我们有员工,我们雇用他们 因为他们的领域专业化 |
| 184 | and the refinement of the work and the work process and all of their life | 改进工作、工作过程和一生 |
| 185 | learnings here in the company is just too valuable. | 公司里的学习太有价值了 |
| 186 | But we also hire consultants and we license external tools and we outsource | 但我们也雇了顾问 给外部工具发许可证 然后外包 |
| 187 | work to other people and so on and so forth. | 对其他人的工作等等。 |
| 188 | I think this is the future for AI, you know? | 我认为这是AI的未来,你知道吗? |
| 189 | And so are we going to continue to use frontier models? | 那么我们要继续使用前沿模型吗? |
| 190 | Absolutely, and tons of it. | 当然,还有很多 |
| 191 | But are we also going to create specialized, you know, super agents | 但我们也要创造出专业的特工吗? |
| 192 | with Lang chain and Nemotron 3 Ultra that, in fact, arguably could be your crown jewels? | 兰连锁店和尼莫特龙3号超市 可以说是你的皇冠珠宝? |
| 193 | And the answer is absolutely true. | 答案是绝对正确的。 |
| 194 | I think even for the consultants that you bring in, | 我想即使你请来的顾问 |
| 195 | just like when you bring in a consultant, you need to get them up to speed | 就像你请来一个顾问的时候, 你得让他们加快速度 |
| 196 | in your organization and give them context on the organization, how things work, | 在你的组织, 并给他们的背景 组织,事情如何工作, |
| 197 | what tools do they need that have access to data that's only inside your organization. | 他们需要哪些工具 能够获取你组织内部的数据 |
| 198 | And so I think like one of the things we've seen is as enterprises start to adopt AI, | 所以我想,就像我们所看到的一样 企业开始采用AI, |
| 199 | there's all of these kind of like systems that they have to build around them in order to make | 他们必须围绕这些系统来制造 |
| 200 | the agentic systems as a whole trustworthy and safe and proper kind of like governance. | 整个代理系统都是值得信赖的 安全而恰当的治理 |
| 201 | I'm curious, how do you see? | 我很好奇 你怎么看出来的? |
| 202 | And just to add on that, just, you know, today, | 顺便说一句 今天 |
| 203 | most companies are built on business processes. | 大多数公司都以业务流程为基础。 |
| 204 | In the future, most companies will be built on harnesses. | 未来,大多数公司都将建立在利用的基础之上。 |
| 205 | And so the idea, you know, Lang chain would just become the tool that creates the operating | 所以这个想法,你知道,兰链 将只是成为工具 创建操作 |
| 206 | system for the company. | 公司的系统。 |
| 207 | And everybody will be using Lang chain to create their specialized harness, | 每个人都会利用Lang连锁公司 来创造他们的专用绳索 |
| 208 | which represents a workflow of the past. | 这代表了过去的工作流程。 |
| 209 | And now this harness inside that workflow becomes autonomous, agentic, you know, | 而现在,这个拉带 在工作流程中 变得自主, 代理,你知道, |
| 210 | much more efficient. | 效率要高得多。 |
| 211 | I think we see that these things are, there's the harness, there's the model, | 我认为我们看到这些东西是, 有绳子,有模型, |
| 212 | and then there's all the context around it. | 然后是周围的所有背景。 |
| 213 | And all of these can be optimized at different points in time. | 而所有这些都可以在不同的时间点优化. |
| 214 | That's right. | 没错 |
| 215 | And so the work that we did with Nemotron 3, I think was a great example of doing | 因此我们与尼莫特龙3的工作 我认为是一个很好的例子 |
| 216 | some pretty high ROI things around the harness, changing the prompt, changing the tools. | 一些相当高的ROI 事物在拉带周围, 改变迅速, 改变工具。 |
| 217 | One of the things we're looking forward to is experimenting with post-training | 我们期待的一件事是试验 后训练 |
| 218 | Nemotron to, it takes a little bit more time, but I think it really raises the ceiling | 尼莫特隆需要多一点时间 但我觉得这真的能提高天花板 |
| 219 | of what this overall system can do. | 这整个系统能做什么。 |
| 220 | This is incredible. | 这是不可思议的。 |
| 221 | This is the big breakthrough. | 这是大突破 |
| 222 | And so what you just described is a future where, once you get the harness built, | 所以你刚才描述的是一个未来 一旦你得到了这个带子, |
| 223 | and it's doing the work, and it's now part of the business process, | 而它正在做工作, 现在它的一部分 业务流程, |
| 224 | and it's very successful. | 它非常成功。 和它非常成功。 |
| 225 | Now the question is, how do we get it even better than that? | 现在的问题是,我们如何得到它甚至更好? |
| 226 | Of course, you can keep improving the information that you provided. | 当然,你可以不断改进你提供的信息。 |
| 227 | You can tune the harness. | 你可以调谐绳索。 |
| 228 | But you can now also improve the AI model, the large language model, | 但你现在也可以改进AI模型,大型语言模型, |
| 229 | Nemotron 3 Ultra, inside the harness. | 尼莫特龙3号超极,在绳索内 |
| 230 | I think that's a complete breakthrough, a capability that's never existed before. | 我认为这是一个完全的突破,一种从未有过的能力. |
| 231 | And I'm super excited about that. | 我对此非常兴奋 |
| 232 | And it's really going to take all of these enterprise-specific business processes | 并且它真的要采取 所有这些企业特有的业务流程 |
| 233 | and really start to tune this flywheel. | 并开始调整这个飞轮。 |
| 234 | And I think one of the things that we've heard when talking to enterprises is | 我想当我们跟企业交谈时听到的一件事就是 |
| 235 | the demand or need for this to be built on an open ecosystem. | 建立开放生态系统的要求或需要。 |
| 236 | This is all of this enterprise's knowledge and processes that they're putting in there. | 这是企业所有的知识 和流程 他们在那里。 |
| 237 | Having full control over that seems paramount to a lot of them. | 完全控制着这件事对很多人来说都至高无上 |
| 238 | So I'm curious if you can touch on how you see open stacks | 所以,我很想知道,如果你可以接触 如何看到打开的堆 |
| 239 | really empowering enterprises going further with AI. | 真正赋予企业更大的权力 |
| 240 | Every company is built fundamentally on domain-specific | 每个公司都以特定领域为基础 |
| 241 | or some specialized intellectual property. | 或某些专门的知识产权。 |
| 242 | The reason why we call it intellectual property, intellectual, it's intelligence. | 我们之所以称之为知识产权,知识,就是智力. |
| 243 | Every single company is built on intelligence, | 每个公司都靠情报工作 |
| 244 | some foundation of intelligence that's specialized. | 某种专门的情报基础 |
| 245 | Our company is specialized in something. | 我们的公司是专门从事某事的 |
| 246 | We're not good at everything, but we're very, very good at one thing. | 我们并不擅长一切,但我们非常,非常擅长一件事. |
| 247 | And every company is built that way, you know? | 每个公司都是这样建的,你知道吗? |
| 248 | And so that specialization, your company's intelligence is who you are. | 所以,专业,你的公司智能就是你是谁。 |
| 249 | You can't possibly not continue to control it, improve it, make it better, right? | 你不能不继续控制它, 改进它,使它更好,对不对? |
| 250 | And somehow outsourcing that intelligence, whether you're a person, company, | 不知何故外包了情报 不管你是个人还是公司 |
| 251 | country, makes no sense to me. | 国家,对我来说毫无意义。 |
| 252 | And so, of course, there's general intelligence. | 因此,当然还有一般的智能。 |
| 253 | And there are general things that we all do. | 也有一些一般的事情 我们都做。 |
| 254 | And it's a lot of the economy, you know? | 和它很多的经济,你知道吗? |
| 255 | And for example, software coding is actually a general thing. | 举例来说,软件编码其实是一般的事情. |
| 256 | 我们都在Python计划。 | |
| 257 | We all program in C++. | 我们都用C++程序 |
| 258 | We all program, right? | 我们都在计划 对吧? |
| 259 | And so you're applying it to different things, | 所以你把它应用到不同的事物上, |
| 260 | but the skill of coding is largely the same. | 但编码技巧基本相同. |
| 261 | And that's a general skill. | 这是一般的技巧 |
| 262 | Writing is a general skill. | 写作是一种一般的技能. |
| 263 | But those are foundational skills that we then apply for our specialized domain intelligence. | 但是,这些是基础技能, 然后我们申请 我们的专业域情报。 |
| 264 | And that's where Langchain and Neemotron comes in. | 兰辛和尼莫特隆就从这里进来了 |
| 265 | I think the foundation of society is going to have these foundational models, | 我认为社会的基础 将会有这些基础模式, |
| 266 | and they're going to be general, and they're going to be available in the cloud, | 且将作大将军,云中可用. |
| 267 | and it's going to be incredible. | 这将是不可思议的。 |
| 268 | But on top of that platform, we're going to have to build our own specialized capabilities, | 但是在这个平台上 我们必须建立我们自己的专业能力 |
| 269 | and you need open tools for that. | 你需要打开的工具。 |
| 270 | And you can't outsource it. | 则不能外包. |
| 271 | You know, I can't imagine calling a third party when I need to enhance my intelligence. | 你知道,我不能想象 打电话给第三方 当我需要增强我的智力。 |
| 272 | You know, I need to enhance it right here inside the company. | 我需要在公司内部加强它 |
| 273 | And so I think that that future is not one or the other. | 所以我认为未来不是一对一。 |
| 274 | It's a completely complementary vision. | 这是一个完全互补的愿景. |
| 275 | And really what we're doing is just making sure that we automated intelligence is integrated | 我们要做的就是确保我们自动智能集成 |
| 276 | into all aspects of everything that we do. | 我们所做的一切 |
| 277 | And as a result, you know, we're all going to be better. | 结果我们都会好起来的 |
| 278 | Completely agree. | 完全赞成。 |
| 279 | And I think it's still hard to get that integration up and running. | 我认为仍然很难实现整合。 |
| 280 | And so one of the things that we're announcing today is a blueprint with deep agents | 所以我们今天宣布的一件事 是一份有深层特工的蓝图 |
| 281 | and open shell inside of the Neemoclaw blueprints out there. | 并打开壳 在尼莫茨瓦夫蓝图 在那里。 |
| 282 | And so this will let enterprises run deep agents with Neemotron 3 Ultra | 这样企业就能和Neemotron 3 Ultra 一起经营深层代理 |
| 283 | inside of open shell, which is a secure and open runtime and take advantage of that. | 打开的外壳里面,这是一个安全开放的运行时间 并利用它。 |
| 284 | This is one of just such a huge deal. | 这是一个如此巨大的交易。 |
| 285 | Hopefully it makes it way easier for enterprises to get up and running. | 希望这能让企业更容易起来和运行。 |
| 286 | Yeah. | 对 |
| 287 | All of the key ingredients necessary for you to build your personal domain specific | 为您构建个人领域所需的所有关键要素 |
| 288 | proprietary, you know, your super agent, all of the technologies, all the components, | 独家,你知道,你的超级代理, 所有的技术,所有的组件, |
| 289 | all the tooling, all of the harnessing, everything and the blueprint, a great example, | 所有的工具,所有的操纵, 一切和蓝图,一个伟大的例子, |
| 290 | all put together for you. | 都为你准备好了 |
| 291 | How do you guys think about blueprints? | 你们觉得蓝图怎么样? |
| 292 | You have many of them. | 你有很多这些 |
| 293 | This is obviously the best one. | 这显然是最好的。 |
| 294 | I won't make you say that, but I'll say that this is the best one out there, | 我不会逼你这么说的 但我会说这是最好的 |
| 295 | but you have a ton of blueprints. | 但是你有一堆蓝图 |
| 296 | Why invest so heavily in them? | 为什么投资这么大? |
| 297 | Well, because the tools are arcane still, you know, and there are a lot of pieces to it, | 嗯,因为工具仍然是 arcane 仍然,你知道, 有很多块它, |
| 298 | you know, building an agentic system, building AI is not simple. | 你知道,建立一个代理系统, 建立AI不是简单的。 |
| 299 | And there's a lot of different pieces of technology. | 并有许多不同的技术. |
| 300 | And we already talked about some of them. | 我们已经谈过其中一些了。 |
| 301 | There's the large language model. | 有大语言模型. |
| 302 | There's the tools it uses and, you know, the knowledge graph that it has to deal with. | 有它使用的工具 还有它必须处理的知识图 |
| 303 | It's memory system and it's guard railing system and it's fine tuning system. | 它是记忆系统,也是警卫栏杆系统,也是精细的调制系统. |
| 304 | Now, you know, the technology, you're going to create the post-training against the harness. | 现在,你知道,技术, 你要去创建 后训练与牵引。 |
| 305 | And then, of course, there's the harness itself. | 当然,还有绳子本身。 |
| 306 | But what about the runtime? | 跑步时间呢? |
| 307 | When you're done, you still have to run time. | 完成后,仍须运行时间. |
| 308 | You have to keep it in a sandbox so it's secure, it's private. | 你必须把它藏在沙盒里 这样它才安全,是私人的 |
| 309 | That access control is something that IT organizations can control. | 这种访问控制是IT组织可以控制的. |
| 310 | Is that the hardest thing about the runtime? | 跑步时间最难吗? |
| 311 | You think inside of enterprises, all the security things that go alongside it? | 你想想企业内部, 所有安全的东西 一起走? |
| 312 | Without solving the security, the access control, it's impossible to deploy. | 如果不解决安全,出入控制问题,就无法部署. |
| 313 | It's no different than it's impossible to hire a new employee into the company. | 这与公司聘请新员工是不可能的一样. |
| 314 | If you don't onboard them, give them access control. | 如果你不在船上, 给他们进入控制。 |
| 315 | We don't give every employee access to every file and every network, right? | 我们不给每个雇员 访问每一个文件和每个网络,对不对? |
| 316 | And so you have every single employee based on their job and their responsibility | 因此,你每个员工 基于他们的工作和责任 |
| 317 | and what they need to have access to. | 以及他们需要获得的东西。 |
| 318 | We give them access to tools, the laptops and, you know, design tools and, you know, | 我们给他们提供工具 笔记本电脑 设计工具 |
| 319 | programming tools and whatnot. | 编程工具什么的。 |
| 320 | We give them access to certain parts of the network. | 我们让他们进入网络的某些部分 |
| 321 | We give them access to information. | 我们让他们获得信息。 |
| 322 | We connect them to other agents. | 我们把他们和其他特工联系起来 |
| 323 | We connect them to other colleagues that they work within. | 我们把他们与他们在其中工作的其他同事联系起来。 |
| 324 | And we provide them a skills file, you know? | 我们给他们一个技能档案 你知道吗? |
| 325 | We essentially, you know, give them a document about this is your mission. | 我们基本上,你知道, 给他们一个文件 关于这是你的任务。 |
| 326 | This is how it's previously been done. | 之前就是这样做的 |
| 327 | And now, you know, help do it even better than that. | 现在,你知道,帮助做它甚至更好。 |
| 328 | And so in a lot of ways, we are creating an HR system, if you will, you know, for AI | 因此,在许多方面,我们正在创建一个HR系统, 如果你愿意,你知道,为AI |
| 329 | that allows the IT organizations and all of the different business units inside the companies | 信息技术组织和公司内部所有不同的业务单位 |
| 330 | to be able to build, improve and deploy these agents inside companies. | 在公司内部建立、改进和部署这些代理人。 |
| 331 | This is more of a philosophical question, but you're talking and I think a lot of people | 这更是一个哲学问题 但你在说话 我认为很多人 |
| 332 | talk about these agents and anthropomorphize them a lot, bring them into human systems. | 说到这些特工 并把他们变成人类 带入人类系统 |
| 333 | But agents aren't human and they have some things that are better than what humans are. | 但特工不是人类 他们有些比人类更好的东西 |
| 334 | And they have other places where they are, you know, very different | 他们有其他地方 他们在那里,你知道,非常不同 |
| 335 | and maybe not as good as what humans are good at. | 也许不如人类擅长的好 |
| 336 | What is the right level to anthropomorphize these agents? | 将这些人化为人类的正确层次是什么? |
| 337 | Well, it's electrons, you know, it's electrons, not atoms. | 是电子,是电子,不是原子 |
| 338 | And it's not biological, has no consciousness, you know, it's not awake. | 而且它不是生物的,没有意识的, 你知道,它没有清醒。 |
| 339 | And so it's not any of that. | 所以它不是任何这样的。 |
| 340 | It's a tool that it's kind of like my vacuum cleaner that's roaming around the house. | 这是一个工具,它有点像我的吸尘器 在房子周围游荡。 |
| 341 | And it's, of course, roaming around the house, cleaning up the house, | 当然,它在房子周围游荡, 清理房子, |
| 342 | doing something that I used to do. | 做我以前做过的事 |
| 343 | And, you know, have autonomous lawnmowers and you have, you know, and so I, | 而且,你知道,有 自主的草坪和你有,你知道, 所以我, |
| 344 | you could just imagine a hundred years ago when the first dishwasher came along. | 你可以想象一百年前 第一个洗碗机来的时候 |
| 345 | And now it's doing dishes by itself. | 现在它自己做菜 |
| 346 | It must have been magical to watch it and we call it a dishwasher, | 看着它一定很神奇 我们称之为洗碗机 |
| 347 | which is a little bit like a human. | 有点像人类 |
| 348 | And so we have dishwashers. | 所以我们有洗碗机。 |
| 349 | When my first job, I was a dishwasher. | 当我第一次工作的时候,我是一个洗碗机。 |
| 350 | And so, you know, and so in a lot of ways, I mean, you know, we'll get used to it. | 所以,你知道, 所以在许多方面, 我的意思是,你知道,我们会习惯它。 |
| 351 | I think it's right now, we tend to imbue too much human properties to it. | 我认为现在,我们倾向于把 太多的人性化为它。 |
| 352 | It's nothing close to that. | 何等为近. |
| 353 | It's software, it's computers. | 是软件 是电脑 |
| 354 | We know exactly how it's working because obviously we created the harnesses around it. | 我们非常清楚它是如何工作的, 因为很明显我们创造了周围的绳子。 |
| 355 | We obviously know how it works because it's getting better all the time. | 我们显然知道它是如何工作的,因为它越来越好。 |
| 356 | If we don't understand how something works, how do we make it better every time? | 如果我们不明白 东西是如何工作的, 我们如何使它更好? |
| 357 | And so if we don't understand how something works, how do we improve it? | 所以,如果我们不明白 东西是如何工作的, 我们如何改进它? |
| 358 | You know, how do we fix it? | 你知道,我们如何修复它? |
| 359 | And so obviously we understand how these things work. | 很明显,我们理解这些事情是如何运作的。 |
| 360 | And I think that we ought to keep it there, you know. | 我认为我们应该把它放在那里,你知道的。 |
| 361 | And meanwhile, one of the things that we know is that the more AI we use, | 与此同时,我们所知道的一件事 就是我们使用的AI越多 |
| 362 | somehow the more people we have to hire. | 不知何故,我们不得不雇用更多的人。 |
| 363 | And the reason for that is because these agentic systems are new skills. | 这是因为这些代理系统是新的技能。 |
| 364 | And now we have a lot of software engineers building agents. | 现在,我们有许多软件工程师 建筑代理。 |
| 365 | They used to code software, but now they're building agents. | 他们以前用代码软件,但现在他们正在建造代理. |
| 366 | If you ask me, every one of my software engineers prefer to be building agents | 如果你问我,我的软件工程师都喜欢做建筑代理 |
| 367 | 而不是写Python代码 | |
| 368 | Coding is like typing. | 编码如打字. |
| 369 | And so they're going to do less typing. | 所以他们会少打字。 |
| 370 | They're going to be more systems engineers, more building engineers, you know, building | 他们将会是更多的系统工程师, 更多的建筑工程师,你知道,建筑 |
| 371 | and creating these autonomous systems that are super cool. | 并创建这些超酷的自主系统。 |
| 372 | They're creating evals. | 他们正在创造Evals。 |
| 373 | They're creating benchmarks. | 他们创造了基准。 |
| 374 | They're creating guardrails, isn't that right? | 他们在制造护栏 不是吗? |
| 375 | And so the amount of work that we have to do to bring AI into the world | 因此,我们必须做大量的工作 让AI进入世界 |
| 376 | is really quite incredible. | 真是不可思议 |
| 377 | And so it's creating a whole bunch of jobs. | 所以它创造了一整批的工作。 |
| 378 | And my software engineers love this. | 我的软件工程师很喜欢这个。 |
| 379 | I think we've seen, you mentioned evals briefly. | 我想我们已经看到了, 你简短地提到了Evals。 |
| 380 | I think we see that being a key part to unlocking a lot of agentic usage | 我认为,我们看到,这是一个关键 部分解锁 很多代理用法 |
| 381 | inside an enterprise. | 企业内部 |
| 382 | You need to have some sense of how it's doing. | 你需要有一点感觉 它是如何做的。 |
| 383 | And quantifying whether it's good or not is oftentimes best done by subject matter | 并量化它是否好 通常是最好的时间 通过主题 |
| 384 | experts who already live inside the enterprise and can easily give feedback | 已经住在企业里并且可以轻易提供反馈的专家 |
| 385 | and work with these systems to automate a lot of the tedious parts of their job | 和与这些系统合作 自动化 很多乏味的工作部分 |
| 386 | and then spend time on the really intellectually stimulating parts | 然后把时间花在 真正的智力刺激部分 |
| 387 | and the creative parts. | 和创作部分。 |
| 388 | And so that's right. | 所以这是对的。 |
| 389 | In a lot of ways, whether you're a doctor or a designer or a software engineer, | 在很多方面,无论是医生,设计师还是软件工程师, |
| 390 | you are creating an agent and you're taking all the mundane work | 你正在创造一个代理 而你要承担所有的工作 |
| 391 | and you're trying to get this agent to do it. | 你试图让这个特工来做 |
| 392 | But meanwhile, we're all trying to get our agents elevated | 但与此同时 我们都在试图提升我们的特工 |
| 393 | to do things with us that we couldn't do before. | 做我们以前不能做的事 |
| 394 | And so that requires imagination. | 而这需要想象力。 |
| 395 | That requires creativity, you know, a lot of technology. | 这需要创造力,你知道, 很多技术。 |
| 396 | I think that's spot on. | 我想这是位置。 |
| 397 | I think currently, a lot of the best usages that we see of agents | 我认为,目前,很多最好的用途 我们看到的代理 |
| 398 | are giving ourselves more leverage to do more things. | 给自己更多的影响力 做更多的事情。 |
| 399 | But I think a lot of that approach is thinking about what did we do previously | 但我认为很多方法 是在思考我们以前做了什么 |
| 400 | and can we automate that? | 我们可以自动化吗? |
| 401 | But I think a lot of the unlock will come in the future | 但我觉得将来会有很多解锁的 |
| 402 | of what couldn't we do before. | 我们以前不能做的 |
| 403 | That's right. | 没错 |
| 404 | That now we can do. | 现在我们可以做到这一点。 |
| 405 | And so maybe- | 所以也许... |
| 406 | Ambition helps. | 野心有帮助。 |
| 407 | A hundred percent. | 百分百都一样 |
| 408 | Ambition, agency. | 野心,机关。 |
| 409 | Ambitions help. | 野心有帮助。 |
| 410 | Maybe on that vein, wrapping up, as you think about how to help drive towards this future, | 也许在那个脉络, 结束, 当你想怎样 帮助驱动未来, |
| 411 | what are some of the missing pieces of this agentic stack? | 这个代理堆的残骸是什么? |
| 412 | Well, today we're announcing a very big deal. | 今天我们要宣布一件大事 |
| 413 | I mean, this is a very big deal thing that we're doing today. | 我的意思是,这是一个非常重大的事情 我们今天做的。 |
| 414 | We are providing the basic building blocks, the foundation, | 我们正在提供基本的构件,基础, |
| 415 | all of the key ingredients. | 所有关键成分 |
| 416 | All of the key ingredients to build super agents. | 建造超级特工的关键成分 |
| 417 | When I say super agents, I mean, they're domain specific. | 当我说超级特工时,我是说,他们是特定领域。 |
| 418 | They belong to you. | 他们属于你。 |
| 419 | You build them, you improve them, you refine them over time. | 你建造它们,改进它们,并随着时间的推移加以完善。 |
| 420 | You give them access to proprietary information, knowledge. | 你让他们获得专有信息 知识 |
| 421 | Maybe it's super private to you. | 也许对你来说是超级隐私 |
| 422 | And as a result, this super agent will be able to do things that you can imagine. | 由此,这个超级特工将能够做你能想象的事情. |
| 423 | And it will be extremely, extremely good. | 这将是极好的,极好的。 |
| 424 | We've created all of the key parts. | 我们创造了所有的关键部分。 |
| 425 | A world-class language model. | 世界一流语言模式. |
| 426 | A framework called Langchain Deep Agents that has now been also fine-tuned | 一个叫做"Langchain深层特工"的框架 现在已经做了微调 |
| 427 | to expose the full potential of the Nemotron 3 Ultra. | 揭开尼莫特龙3号超极星的全部潜力. |
| 428 | A blueprint that helps everybody do that. | 帮助每个人这样做的蓝图。 |
| 429 | And of course, the runtime, the open shell runtime that keeps it secure. | 当然,运行时间, 开放的壳运行时间 保持它的安全。 |
| 430 | The acceleration stacks that are all integrated into it. | 全部融入其中的加速度堆栈. |
| 431 | And so every company in the world should be able to, | 所以世界上的每一个公司都应该能够 |
| 432 | every developer in the world should be able to now create these super agents, | 世界上每个开发商 都应该能创造出这些超级代理 |
| 433 | deploy it anywhere in the cloud, on-prem. | 在云中任何地方部署它,在预告。 |
| 434 | A good friend of mine just built one for DGX Spark. | 我的一个好朋友刚为DGX Spark造了一台 |
| 435 | And so now you have these agents running on DGX Spark right next to your laptop. | 现在你让这些特工 运行在DGX Spark 在你的笔记本电脑旁边。 |
| 436 | You could have it running on a DGX station. | 你可以让它运行在一个DGX站。 |
| 437 | You could build your own supercomputer inside your company, | 你可以在你的公司里建造自己的超级计算机 |
| 438 | if you like, or do it in the cloud. | 如果你喜欢,或者在云中做。 |
| 439 | We now have agentic capabilities that you can now build for yourself everywhere. | 我们现在有你现在可以为自己建立的能力。 |
| 440 | All the pieces are now here. | 所有碎片都在这里 |
| 441 | There are no excuses not to engage it. | 没有借口不参与。 |
| 442 | I think that's a perfect way to end it. | 我想这是结束这一切的绝佳办法 |
| 443 | You got me so pumped up when you were speaking. | 你说话的时候让我很兴奋 |
| 444 | A great motivational speach. | 伟大的动力弹簧。 |
| 445 | So I'm going to go out and build some. | 所以我要出去建造一些。 |
| 446 | I'm going to go down and build some agents. | 我要下楼盖些探员 |
| 447 | Thank you, Johnson. | 谢谢你,约翰逊。 |
| 448 | Congratulations. | 祝贺你 |
| 449 | Good job. | 干得不错 |
| 450 | Proud of you guys. | 为你们骄傲 |
| 451 | Job, proud of you guys. | Job,为你们骄傲。 |
| 452 | Thank you. | 谢谢 |
该视频共有字幕 453 条。解锁更多字幕为会员功能,请移动到 价格
