中国蒸馏美国AI算偷吗？黄仁勋：那叫竞争

发布于 2026-10-16 · 时长 9:02

字幕摘录

时间	英文	中文
0:00	I read an article over the weekend that compared you to Thanos, the supervillain from Marvel.	我周末读过一篇把你比作Thanos的文章, 来自Marvel的超级恶棍.
0:05	But when you actually sit and talk to you, it's a much more nuanced view than that.	但当你坐下来和你说话时, 它比那更细微。
0:09	You do want some of these safeties.	你确实想要一些这些安全。
0:11	I'm a responsible optimist.	我是一个负责任的乐观主义者。
0:14	You think distillation is not robbery.	你认为蒸馏不是抢劫
0:16	That's called competition.	这就是所谓的竞争。
0:18	And Jensen, welcome.	还有詹森欢迎
0:19	It is great to see you this morning.	今早见到你真好
0:20	Great to see you.	很高兴见到你。
0:21	So this is an advancement in something that we've kind of seen the national discussion	因此,这是一个进步的东西,我们已经看到全国性的讨论

展开字幕全文（149 条）

序号	英文	中文
1	I read an article over the weekend that compared you to Thanos, the supervillain from Marvel.	我周末读过一篇把你比作Thanos的文章, 来自Marvel的超级恶棍.
2	But when you actually sit and talk to you, it's a much more nuanced view than that.	但当你坐下来和你说话时, 它比那更细微。
3	You do want some of these safeties.	你确实想要一些这些安全。
4	I'm a responsible optimist.	我是一个负责任的乐观主义者。
5	You think distillation is not robbery.	你认为蒸馏不是抢劫
6	That's called competition.	这就是所谓的竞争。
7	And Jensen, welcome.	还有詹森欢迎
8	It is great to see you this morning.	今早见到你真好
9	Great to see you.	很高兴见到你。
10	So this is an advancement in something that we've kind of seen the national discussion	因此,这是一个进步的东西,我们已经看到全国性的讨论
11	take place around AI and AI safety.	围绕AI和AI安全进行.
12	And in this discussion, there have been those who have been painted as do-mers who think	在这场讨论里, 有一些被画成Do -mers的人认为
13	that the end of the world is going to come from this.	世界末日将由此而来
14	And then there have been those who have painted you as the guy who doesn't want to see anything	然后有人把你画成一个不想看到任何东西的人
15	happen on any level.	发生在任何层面。
16	I don't think that's a fair and accurate depiction of where you stand with things.	我认为这是不公平和准确的描述你站在什么事物。
17	Why don't?	为什么不呢?
18	I want to see a lot of things happen on a lot of levels.	我想看到很多事情发生在很多层面。
19	So why don't we talk about where we stand and then we're going to jump into what this	所以我们何不谈谈我们的立场然后我们跳进这个
20	is.	没错
21	Well, first of all, first of all, as you know, we're going through a major platform shift.	首先,你知道,我们正经历一个重大的平台转变。
22	AI is incredible technology, has the potential to do incredible good.	AI是不可思议的技术,有潜力做不可思议的好事.
23	But we also have to make sure that the technology is developed and deployed safely so that the	但我们也必须确保技术的开发和安全部署,以便
24	public has confidence in applying this technology.	公众有信心应用这种技术。
25	That single most important thing we can possibly do is to build it in such a way that is so	我们能做的最重要的一件事就是用这样的方式建造它
26	safe that everybody could use it in every single industry so we can transform every	保证每个人都能在每一个行业使用这样我们就能改造每一个
27	industry.	工业。
28	And so that's the hard work that we have to do.	这就是我们必须做的艰苦工作。
29	These agentic systems are obviously agentic, meaning they operate by themselves.	这些代理系统显然是代理的,意味着它们自己操作.
30	It's like a self-driving car.	如自驾车.
31	And so in order for you to deliver that agentic system in a safe way, you have to make sure	所以,为了让你安全地提供代理系统,你必须确保
32	that the sandbox around it, the force field, the playpen, if you will, all of those systems	周围的沙盒, 力场,球杆,如果你愿意,所有这些系统
33	are designed in a way that keeps the agent with minimal rights, whatever rights it has,	其设计方式使代理人拥有极小的权利,无论它拥有什么权利,
34	it needs in order to do its job.	它需要做它的工作。
35	No more rights than that.	没有更多的权利。
36	And that it can't do anything that it's not supposed to.	且其不能作其不应作之事.
37	And also, you have to monitor it.	而且,你必须监视它。
38	但是,我们已经看到的问题在这一点上,OpenAI,Anthropic,Google等
39	have had agents that have escaped and have gotten out and done mischievous things, maybe	曾经有特工逃脱了, 并做了错误的事情,也许
40	illegal things, the problems we've seen to this point you think is simply an engineering	非法的东西,我们已经看到的问题在这一点上,你认为只是一个工程
41	problem and a fault of those companies maybe not putting in the right restrictions?	问题和那些公司的过错也许没有施加正确的限制?
42	Well, I think the answer is we hope it's an engineering problem.	我想答案是我们希望这是一个工程问题。
43	I believe it's an engineering problem.	我认为这是一个工程问题。
44	I know it's an engineering problem.	我知道这是一个工程问题。
45	And we all need to hope that it's an engineering problem.	我们都希望这是一个工程问题。
46	It's not an engineering problem.	这不是一个工程问题。
47	It's not solvable.	不可解脱.
48	And so the fact that all of these companies still are advancing the state of the art is	因此,事实上所有这些公司仍然在推进最新艺术
49	because they also believe it's solvable.	因为他们也相信它是可溶解的。
50	These are some of the brightest engineers in the world.	这些是世界上最聪明的工程师
51	I work with all of them.	我和他们一起工作
52	And this is a technically solvable problem.	这是一个技术上可以解决的问题。
53	And so the way to think about that is if you look at, go back to the days of the internet,	所以,思考的方式是如果你看看, 回到互联网时代,
54	you access a website, it downloads application, maybe a Java application.	您访问一个网站, 它下载应用程序, 也许是一个Java应用程序。
55	All of a sudden, this application comes into your PC, has access to files and resources	突然间,这个应用程序进入了您的个人电脑,可以访问文件和资源
56	and maybe the applications and tools, and it gets onto your network.	也许应用程序和工具, 它会进入你的网络。
57	Well, in the beginning, we had all kinds of viruses.	一开始,我们有各种各样的病毒。
58	But we realized that the web browser itself has to be a containment system.	但我们意识到网页浏览器本身必须是一个封存系统.
59	The web browser has to provide that application with minimal rights, only the rights it needs	网页浏览器必须提供该应用程序的最小权利,只有它需要的权利
60	in order to do its job.	以成其业.
61	And it has basically access to nothing unless the browser gives it access.	除非浏览器允许它访问否则它基本上什么都无法访问。
62	Essentially, what we're doing here, we're creating basically the modern browser.	基本上,我们在这里所做的, 我们基本上正在创建现代浏览器。
63	It's a browser for agents.	这是代理商的浏览器.
64	So somewhere inside the culture of AI researchers for years now has been this anxiety that somehow	多年来在人工智能研究者的文化中有一种焦虑
65	these models have to be clever.	这些模型必须很聪明。
66	And they have to be able to get outside of their walls because they're so clever.	他们必须能够走出他们的墙,因为他们非常聪明。
67	Where do you think that's built into the AI researcher culture?	你认为这在AI的研究者文化中是怎样构建的?
68	And so at one level, I think while there are some people who believe it's an engineering	所以从一个角度来说,我想, 虽然有些人认为这是一个工程
69	problem, there are these others even inside these companies that believe it's unsolvable.	问题是,有些公司内部的人认为这无法解决
70	I don't think so.	亦不见.
71	I don't think that's exactly wrong.	我不认为那完全是错误的。
72	First of all, you want it to be super clever so it could perform tasks and do things.	首先,你希望它非常聪明这样它就能完成任务和做事。
73	But the way you think about that is when you deploy an agent, no matter how smart, the	但是,你的想法是当你部署一个特工, 无论多么聪明,
74	first thing you do is you take away all of its rights.	你要做的第一件事就是夺走它的所有权利
75	Job number one is take away all of its rights.	一号工作剥夺了所有的权利
76	And then you provision, you give it access to files, data, tools, access to the network	然后你提供,你给它访问文件,数据,工具,访问网络
77	or even internet access only if it needs it.	甚至互联网访问只有在它需要的时候。
78	You don't put an agent into an environment, into your company and give it access to everything.	你不能让经纪人进入一个环境进入你的公司并让它接触一切
79	We don't give that to people.	我们不给人。
80	We don't give that to any level of executive, in fact.	事实上,我们不给任何行政级别。
81	Everybody has a badge.	每个人都有徽章。
82	Everybody has file access.	每个人都有文件访问。
83	Everybody has access to tools.	每个人都可以使用工具。
84	And only the things that they need in order to do their job.	只有他们需要的东西来完成他们的工作。
85	And so we have to apply the same philosophy.	所以我们必须应用同样的哲学。
86	Now the question is, how do we develop the technology to provision that policy, provision	现在的问题是,我们如何开发技术来提供这种政策,规定
87	that philosophy?	那个哲学?
88	And that's what we created.	这就是我们创造的。
89	That's what Open Shell is.	即是开示壳.
90	It's really important technology.	这是非常重要的技术。
91	It has the support of some hundred companies.	它得到了大约100家公司的支持。
92	你没提到IBM和Red Hat 还有微软和SpaceX
93	And so all of these companies, they recognize the importance of basically this force field	因此,所有这些公司, 他们认识到基本上这个力量领域的重要性
94	that you create around the agent.	你在经纪人周围创造的
95	And very importantly, it's provisioning rights and responsibilities and access.	更重要的是,它提供权利和责任和获取。
96	It's an open source program.	开源程序.
97	Which is very, very important.	这很重要
98	And the reason for that is this.	原因就是这个。
99	We want this sandbox to be visible to the whole world so that if there are any vulnerabilities,	我们希望全世界都能看到这个沙盒如果有弱点
100	somebody will find it.	有人会找到的
101	You've got all the...	你已经得到了所有... ...
102	Look at these hundred companies.	看看这百家公司
103	There's actually a lot more companies.	实际上还有更多的公司。
104	Those are the only we can list.	这是我们唯一可以列出的。
105	The industry will all be jumping in looking at any possible vulnerabilities and patch it up.	业界将全心全意地研究任何可能的弱点,并弥补这些弱点。
106	你有Anthropic和SpaceX 都在这百家公司的名单上
107	Open AI is not there.	打开AI不在那里。
108	Are they not cooperating with this?	他们不合作吗?
109	They didn't think it was important?	他们不认为这很重要吗?
110	No, this is an open standard.	不,这是一个开放的标准。
111	It's an open platform.	开源平台.
112	And however they would like to participate is going to be super welcome.	然而,他们愿意参加,将是非常受欢迎的。
113	They could look at the code.	他们可以查看密码
114	They could use whatever part of it they would like.	他们可以使用任何他们想要的部分。
115	They could build their own inspired by this.	他们可以建立自己的灵感。
116	Or they could use it.	或者他们可以使用它。
117	I think over time, we really want people to use the software.	我想随着时间的推移,我们真的希望人们使用软件。
118	And the reason for that is because the more people use it, the more vibrant it's going to be,	原因是人们越多使用它,它就越充满活力,
119	the more resilient and more robust it's going to be.	越坚韧,它会变得更强壮。
120	你说过Open Shell会阻止来自Open AI的拥抱面部突围.
121	Clem和Hugging Face是袭击的受害者之一,也是Open Shell的巨大支持者.
122	Their computer sciences are very deep on Open Shell and they're a big part of it.	他们的计算机科学非常深入Open Shell,并且是其中的一大部分.
123	What happened that you all can do this and Open AI, Anthropic and others haven't been able to do this to this point?	你们都能做到什么开放AI、Anthropic和其他人还没有做到这一点?
124	Well, we started building this about a year ago.	我们大约一年前就开始造了
125	Has it been a year?	有一年吗?
126	Maybe slightly longer.	也许再长一点
127	Open Claw came out.	打开的克劳出来了。
128	And when Open Claw came out, you guys might have heard me say this is the operating system of AI,	当Open Claw出来时你们可能听到我说这是AI的操作系统
129	meaning that it's the operating system of agents.	意思是它是代理的操作系统.
130	And when we looked at Open Claw, we said, hey, this is incredible.	当我们看着Open Claw, 我们说,嘿,这是不可思议的。
131	But you can't deploy this into companies.	但你不能把这个部署到公司
132	You can't have agents roam around and drift around the company.	你不能让特工在公司里到处游荡
133	And so you have to find a way to container it.	所以,你必须找到一个方法把它装进去。
134	You have to contain it is number one.	你必须控制它,第一。
135	Rights provisioning is number two.	权利供给第二.
136	Being able to monitor its adherence and conformance to all those policies, number three.	能够监测其遵守和遵守所有这些政策的情况,第3项。
137	And so we created Containment, a policy provisioning supervisor, and we created a monitoring system.	于是我们创建了控制系统, 一个提供政策的监督, 我们建立了一个监测系统。
138	And the monitoring system is called Sentry and it monitors it in silicon.	而监控系统则被称为哨兵,并以硅形式进行监控.
139	Not to get too technical about this.	别太技术了
140	But the concept of containment and monitoring is very simple.	但遏制和监测的概念非常简单。
141	For those who know about, for example, AWS Bedrock has Nitro and other things built into it.	例如,对于知道AWS Bedrock的人来说,它包含了硝酸盐和其他东西。
142	This seems to me a much broader open source version, frankly, of that.	坦率地说,这似乎是一个更广泛的开放源码版本。
143	Is that a terrible way to think about it?	这是不明智的思考方式吗?
144	Yeah, it's like Nitro in the sense that it's a rights provisioning system.	是啊,这就像硝酸盐的意义上它是一个权利供给系统。
145	It's like Nitro in the sense that it's also hardware isolation.	这就像硝酸盐,它也是硬件隔离。
146	Right, which is really which Nitro, as you know, really revolutionized multi-tenant secure cloud.	没错,这真的是Nitro,你知道的, 真正革命性的多租户安全云。
147	In a lot of ways, this is multi-agent secured, you know, world.	在很多方面,这是多代理安全,你知道,世界。
148	And so a very similar idea.	一个非常相似的想法。
149	The difference, the differences are going to be around the speed at which we have to.	差别,差别将围绕我们必须的速度。

该视频共有字幕 149 条。解锁更多字幕为会员功能，请移动到价格

下载安装