Sriram Krishnan 谈开源 AI 迄今为止最重大的一周

你不能把事情再回到最基础的商业原理上。

You cannot bring it back to very business first principles.

只要你提供的产品有价值，资本主义自然会找到办法让供应链为你服务。

If you're providing a product of value, capitalism will find a way to make the supply chain work for you.

所以，如果你有一个开源权重模型在创造价值，那么底层的每一个环节——无论是新兴云服务商、芯片供应商，还是提供燃气轮机或消防系统的公司——都会围绕着如何创造价值来调整自己。

So if you have an open-weight model that is providing value, that means that every part of the stack underneath, whether it is a neo cloud, the chip provider, somebody who provides gas turbines or fire suppression, is going to orient itself to provide value.

只要你提供的产品有价值，资本主义会帮你解决所有其他问题。

If you're providing a product of value, capitalism will take care of all the rest.

如果你去看看整个生态系统的现状，增长势头非常强劲，而且令人惊叹，我觉得

If you go look at how the rest of the ecosystem is doing, the growth is pretty strong and spectacular, and I think

这种趋势还会持续下去。

you're gonna see that continue.

开源人工智能的发展速度前所未有，整个行业的权力格局可能正在发生变化。

Open source AI is moving faster than ever, and the balance of power in the industry may be shifting.

在这期节目中，西奥·贾菲和索菲亚·普奇尼邀请了前白宫人工智能政策顾问斯里拉姆·克里希南，深入探讨最新一波开源模型对前沿实验室、人工智能政策、定价、网络安全，以及美国在全球人工智能竞赛中地位的影响。

In this episode, Theo Jaffee and Sofia Puccini are joined by former White House AI policy adviser, Sriram Krishnan, to unpack what the latest wave of open models means for Frontier Labs, AI policy, pricing, cybersecurity, and America's position in the global AI race.

我们回来了。

We are back.

我们现在连线的是Sriram Krishnan，他刚结束担任白宫人工智能政策高级顾问的职务，之前曾是安德森霍洛维茨的普通合伙人，并在微软、Meta、Snap和Twitter担任过高级职位。

We are live with Sriram Krishnan who just finished his tenure as the senior White House policy adviser on artificial intelligence, previously as a general partner at Andreessen Horowitz, and held senior roles at Microsoft, Meta, Snap, and Twitter.

Sriram，非常高兴你能来。

So, Sriram, we're so glad to have you on.

欢迎来到MTS。

Welcome to MTS.

谢谢。

Thank you.

过去几个月，我一直关注你们做的所有事情，非常高兴能参与这次对话。

I've been a fan of everything you folks have been doing for the, you know, last few months and excited to be here.

这大概是两年来我第一次能以这么轻松的打扮出镜，不用穿西装打领带。

I think this is the first time in about two years I've been able to do a video appearance without a suit and tie on.

能摆脱那身行头，我真是松了一大口气。

So, I am so excited to be out of that.

是的。

Yeah.

对。

Yeah.

太棒了。

Amazing.

过去一周开源领域发生了好多事。

So, there's so much going on in open source the last week.

我们刚聊到，Grok Build 开源了，接着是 Thinking Machines，然后是 Kimi K3，还有 Qwen 3.8。

We just discussed we had Grok Build was open source, and then we had Thinking Machines, and then Kimi K three, and then Qwen 3.8.

你前几天发推说，Kimi K3 对整个行业来说是个重要时刻。

So you tweeted the other day, Kimi K three is a big moment with multiple implications for the entire industry.

能再详细说说吗？

Could you go into a little more detail on that?

这些影响具体体现在哪些方面？

What are these implications?

如果你回溯到四、五个月前，那时候领先的模型基本上只有OPUS 4.6或4.7，还有GPT 5.4、5.5这些，感觉其他人都跟不上。

So if you go back maybe four, five months, I think there was a moment in time when the only leading models were, I think, OPUS four six or four seven at the time, GPT 5.4 or 5.5 or wherever we were, and it felt like there was really no one else.

那时候我们正处在自我提升的轨道上，前沿实验室的差距会越拉越大。

And we were on this curve of self improvement where the Frontier Labs were really going to draw really far away from everyone else.

但最近几周，如果你像我一样做令牌消耗相关业务，或者你和你的观众也在做这个，那真是个好时候。

I think the last few weeks, if you are in the token consumption business, which I am and I think many of you and your viewers are, it's been a great time.

你看，埃隆和迈克尔在Cursor那边，SpaceX的X AI团队推出了Grok 4.5，我一直在用。

Because let's see, you had Elon and Michael at Cursor, you know, the SpaceX x AI team come out with Glock 4.5, which I've been using.

这个模型太棒了。

It's a fantastic model.

我们之前居然忘了提这个。

I think we forgot to mention this.

还有Alex Wang和Meta推出的Muse Spark，也特别出色。

We had Alex Wang and Meta come out with Muse Spark, which is also awesome.

上周，Mira Thinky发布了Inkling。

Last week, we had Mira Thinky come out with Inkling.

我不清楚它的版本号，但这个模型的第一个版本，在很多基准测试上表现都几乎已经顶尖了。

I don't know their version number, but, you know, the first version of that model, which I think is near, you know, nearly sorta on many, many benchmarks.

不过过去三四天最大的新闻，显然是Kimi K3在周四或周五发布了。

But I think that the big news over the last three, four days was obviously Kimi K3 coming out, I think, on Thursday or Friday.

而过去24小时，我还没来得及用，Qwen也发布了。

And then I think the last twenty four hours, haven't really played with it yet, Qwen coming out.

现在市面上一下子冒出好多选择和替代方案。

So there's a lot of choices and alternatives coming out.

我刚才说Kimi K3，主要是想讲这两点。

And what I was referring to is with Kimi K3 is the following.

第一，生态里有更多选择真的很好，你可以自由地把你的工具或代理对接到不同的模型上。

One is that it's just great to have choice in the ecosystem and to be able to point your harness of choice or, you know, your agent of choice to multiple models.

第二，我们现在正处在一个很特别的时刻——一些美国的前沿模型开始受限了。

Second, I think we are in this really weird moment now where some of the American frontier models are constrained.

比如在网络安全和安全防护方面。

For example, on cyber and on security.

我跟一个朋友聊过，他开始用Kimi K3来做安全相关的工作，而不是用Fable，因为用Fable的时候，老是遇到拒绝和防护机制。

And I was talking to a friend of mine where this person was actually starting to do security work using Kimi K3 rather than Fable, because with Fable, he would run into these refusals and safeguards.

这种情况其实挺奇怪的，我们可以稍微聊聊这个。

So that seems like a very weird spot to be, which we can talk about for a second.

我觉得，既然你能从不同地方获取智能服务，那必然会对前沿模型形成价格压力，结果就是前沿实验室可能得降低每令牌的费用，或者想办法跟上价格，这样一来，他们的毛利恐怕会被压缩。

I think it's probably inevitable that if you are having choices from where you get your intelligence tokens from, that's going to put pricing pressure on the frontier models, which means I think you'll probably see the frontier labs have to drop token prices or find ways you know, find ways to match pricing, which means it's probably gonna erode into their gross margin.

这对新兴云服务商和整个技术栈的其他层反而是好事——比如你要是像Base10或者Fireworks这样的新兴云服务商，或者手里有大量Blackwall设备，能跑得动这些模型，就能分到一部分原本属于前沿实验室的收益。

It's probably great for the neo clouds and every other layer of the stack because, you know, if you're a neo cloud like a base 10 or a Fireworks, or if you just have a bunch of Blackwalls and you can power it in, you can run that, and you can capture some of the economics, which is probably going to go to the Frontier lab.

所以作为用户，我觉得这挺好的。

So I think it's great as a consumer.

对整个生态来说，也是好事。

I think it's great for the ecosystem.

当然，安全、模型蒸馏、这些做法到底好不好，还有很多问题值得讨论。

Lots of, obviously, questions on security, on distillation, on how this is good.

但眼下确实是个非常有意思的时刻。

But a very, very interesting moment.

完全同意。

Totally.

所以第一个问题就是，你觉得前沿实验室会怎么应对？

So I guess, like, the first question here would be, where do you predict the Frontier Labs?

你觉得前沿实验室接下来会怎么反应？

Like, how do you predict the Frontier Labs are gonna react to this?

现在整个生态已经发生了变化。

Like, now, you know, the ecosystem has been sort of changed.

回不去了。

There's no going back.

Kimi已经发布了，它的能力跟Fable几乎不相上下。

Kimi has been released, it's very close to the capabilities of Fable.

那你觉得Anthropic或者OpenAI接下来会怎么走？

So what do you think is next for an Anthropic or an OpenAI?

有几点要说。

So a few things.

我觉得，前沿实验室肯定会继续在那种极不稳定但顶尖的性能上不断突破。

I think it's very clear the frontier labs are going to push at the very, very frontier of the jagged performance we get.

我过去十八九个月一直在努力确保美国能赢，所以我特别希望看到美国的模型——不管是闭源还是开源的——能始终站在前沿，我相信它们会继续保持这个地位。

As somebody who spent close to the last eighteen, nineteen months trying to make sure America wins, I really want to see the American models, whether it's closed or open weight at the frontier, which And so, think they'll continue to be that.

我觉得开源模型真正能给它们带来压力的，可能是价格方面。

What I suspect these open models could really start putting pressure on them is one on pricing.

因为真正需要顶尖智能的任务，可能只占一小部分；但很多其他任务，比如我有个助手帮我查邮件，或者快速扫一眼日程安排，其实根本用不上前沿模型。

Because it may turn out that the number of tasks that you need absolutely frontier intelligence from is, let's call it like one subset, but for a lot of other tasks, for example, like I have an agent which checks my email or I have an agent which quickly scans through my calendar, you may not need Frontier tokens.

你完全可以改用稍微差一点的模型，比如‘前沿 minus one’，或者你选的任何开源模型。

You may be able to get by with Frontier minus one or your open weight token of choice.

这样的话，前沿实验室的价格压力就会慢慢显现出来。

In which case, I think you're going to start seeing pricing pressure come on the Frontier labs.

它们很可能会想办法确保最顶尖的模型依然可用、容易获取。

You're probably going to see them try and make sure the absolute Frontier models stay available and accessible.

我觉得Anthropic已经这么做了，虽然我没法证实这是否是他们的动机，但它们确实已经扩展了Fable的功能。

I think we've already seen Anthropic, and I have no inside confirmation about this as a motivation, but you've already seen them extend Fable.

我觉得Fable原本应该只开放到大概一周前吧。

I think Fable is originally supposed to be available until, I don't know, like a week ago.

结果已经延长了。

It's already been extended.

我猜还会继续延长，不然的话，一个跟前沿模型几乎没差的开源模型就太危险了。

I predict there'll probably be more extensions just because otherwise you have an open model, which is very much near the frontier.

另一个有趣的问题是，如果你是前沿实验室，真正的护城河到底在哪？

I think the other interesting question is where is the real moat if you're a frontier lab?

是在智能本身，还是在使用的工具和接口上？

Is it in the intelligence or is it in the harness?

我觉得Claude、CoWork、Cloth Code和Codex这些产品都做得非常出色。

And I think, know, Claude, know, CoWork and, you know, Clod Code and Codex are amazing products.

我猜他们会更努力地让这些产品变得难以离开，因为真正的智能，或者至少是前沿之下的那些能力，可能越来越像大宗商品了。

And I suspect you'll probably see more effort on making them sticky because it might be that the actual intelligence itself, or at least the levels behind the actual frontier, are much more of a commodity.

那这对整个生态意味着什么？

So, what does this mean for the ecosystem?

首先，这可能会给前沿实验室带来收入压力，因为我猜很多人会开始把他们的工具转向这些开源模型，原因有很多，咱们待会儿再细说。

One, this could put some revenue pressure on the frontier labs because I suspect that people might start pointing their harnesses at some of these other open weight models for any number of reasons, which we can get into.

我觉得这对Neo Clouds以及其他拥有算力和GPU的公司来说，大概是个好事，就像我之前说的。

I think it's probably great, like I said, for the Neo Clouds and anybody else with power and GPUs.

我也说不准。

And I don't know.

接下来这段时间肯定会特别有意思。

It's gonna be a very, very interesting time.

没错。

Right.

说到开源，最近确实有些讨论。

So there's been speaking of open source, there's been some chatter.

今天早上Axios刚报道，特朗普政府正在考虑限制中国开发的开源模型，要么是因为它们来自中国，可能涉及国家安全，要么是因为它们已经达到了类似Fable级别的能力，能自主编程，可能带来网络安全风险。

Axios just reported this morning that the Trump administration is considering restricting Chinese open source models, either because they are Chinese and might be a national security issue or because they have reached, like, FABLE level capabilities and agentic coding, and it could be a cybersecurity issue.

你觉得美国政府会出手打压开源模型吗？

So do you think the US government will do things that will crack down on open models?

那你觉得在这方面，他们应该做点什么吗？

And do you think that there's anything that they should be doing in this respect?

说实话，我没什么内部消息。

Well, I have, like, no inside information.

我常跟人说，我现在不再拿美国纳税人的钱了，但我那儿还有很多朋友。

As I like to tell people, I am no longer living off your American taxpayer money, but I have a lot of friends there.

依我听来的消息，我觉得……

And, you know, I I think and I think from everything I hear look.

如果回溯到一年半以前，特朗普政府刚上台那几周，就说要推出一项AI行动计划。

If you go back to a year and a half ago, in the first few weeks, the Trump administration said, Hey, we're gonna come out of this AI action plan.

而这份计划一出来，第一页就写得清清楚楚。

And when you came out of the AI action plan, it's right in the first page.

如果你翻到第一部分，就会看到里面特别强调了开源的重要性。

If you're gonna go to the first section of it, you're gonna see the action plan talk about how important open source is.

不过我觉得，现在得分开来看，这里面涉及好几个不同的问题。

Now, I think the thing to think about is there are multiple different issues.

第一，我觉得领先的一些开源模型或开放权重模型不是美国的，这并不理想。

Number one, I don't think it is great that the leading open-weight models or open source models are not American.

对吧？

Right?

Moonshot、DeepSeek 和 Qwen 确实有不少创新，但我更希望领先的模型是美国的。

I think there's some great innovation happening with Moonshot and DeepSeek and with Qwen, but I would much rather prefer that the leading models are American.

当然，现在也有很多不错的努力在进行中。

And by the way, there's lots of great efforts happening.

我相信它们会越来越好的。

I expect them to improve.

谷歌有 Gemma。

You have Gemma from Google.

英伟达有 Nemotron。

You have Nemotron from Nvidia.

还有明显的是 Thinky。

You have obviously Thinky.

现在还有一些新兴创业公司，比如Reflection，也推出了自己的模型。

You have newer startups like Reflection coming out.

我觉得它们会越来越强，但现在确实是个特殊时刻——领先的模型都是中国的，我觉得这不太理想。

I think they will continue to be better, but we are in a moment of time when the leading models are Chinese, which I think is not great.

我更希望它们是美国的。

Would much rather prefer them to be American.

我觉得第二点是，我从小就很热爱开源。

I think the second part of it is, look, I grew up loving open source.

开源在我接触计算机、塑造我职业生涯的过程中，扮演了非常重要的角色。

Open source is a big part of how I got into computers, a big part of my career.

我一直很推崇林纳斯定律，就是Linux创始人林纳斯·托瓦兹提出的那个观点。

And I was a big fan of Linus' law, as in Linus Torvalds of Linux Fame's law.

他的说法是，只要足够多的人盯着看，再隐蔽的bug都会被发现。

And his law was that given enough eyes, all bugs are shallow.

我相信这个道理用在开源模型上也一样：当你从Hugging Face下载一个模型，就意味着全世界的人都能拆开它、检查它、微调它、修改它，这些在闭源模型上根本做不到。

And what I believe with that is that open weight models are inherently secure because when you download a model of Hugging Face, it means you have the entire world being able to take it apart, inspect it, fine tune it, modify it, look at it in ways that you absolutely cannot if they are closed.

所以我相信，它们在安全方面带来了一种非常独特的积极视角。

So, I kind of believe that they bring a very, very different positive angle to security.

但我们现在所处的这个状况有点奇怪，我觉得不太妙——今天早些时候，Hugging Face 出了一件事，刚被报道出来，我记得当时还有人发了推文。

The weird, I think, or the moment we are in, which I think is not great is I think there was an incident with Hugging Face that got reported on earlier today, which, you know, I think it was an active tweet.

Hugging Face 发现，有人在用一个明显的 AI 大模型代理，从多个方向持续攻击他们，试图找出漏洞。要应对这种问题，关键是要让美国、西方或盟友这边的安全团队能接触到最先进的模型，这样我们才能把软件做得更安全。

And what Hugging Face seems to have found is that somebody was using essentially definitely an AI LLM agent to hammer them at multiple places and to find ways to break And I think the way to counter that is to make sure American or Western or allies defenders have access to the best models to make one, our software, like just more secure.

所以，正如我之前说的，我觉得现在用 Fable 这样的工具检查自己代码里的漏洞，反而比用中国模型更难，这局面真有点别扭。

So, is why, like I said, I don't love that we are in a moment where it may be harder to look at your own code for exploits with say Fable than it is when you use a Chinese model.

这种处境确实挺奇怪的。

That's just a weird spot to be.

我觉得我们得想办法改变这个状况。

And I think we should fix that.

现在有个特别有意思的话题，在 X 上讨论得特别多，就是模型蒸馏到底意味着什么。

Now, one very interesting discussion, which has been happening a lot on X is about distillation and what that means.

这话题其实挺复杂的，因为里面牵扯到好几层意思。

So, it's kind of a complex topic because there's a few things in there.

首先，Ida，我们今天所有的模型，本质上都是从人类的全部知识中提炼出来的。

So, first of all, Ida, like every model we have today distilled off of all human knowledge.

对吧？

Right?

如果你回看最初的GPT或者Claude，它们都得从互联网上爬取数据，抓取我们的博客、推文和各种内容。

Like, if you go back to the original GPT or the original Claude, they all had to derive crawling off the internet, crawling off all of our blogs or tweets and content.

它们实际上是吸收了人类的知识，来完成自身的初始化。

They kind of consume human knowledge and to bootstrap that.

所以，知识蒸馏一直以来都是这些模型训练的核心部分。

So, distillation has always been a core part of how these models have been trained.

其次，如果你——

Second, if you Okay.

我问你一个问题。

Let me ask you this.

现在我发推文的时候，总是担心一不小心就用了好几个连字符，对吧。

When I write a tweet these days, I am terrified of accidentally using, like, you know, multiple hyphens Mhmm.

或者不小心说了什么，让Pangram判断这是AI生成的内容。

Or accidentally saying something which will cause Pangram to say this is AI generated.

我最近发过一条推文，明明每个字都是我自己写的，还是拿去给Pangram检测，怕它觉得不够像真人。

I once wrote a tweet recently, which I put it through Pangram to make sure it felt human, even though I knew I had written every human generated token.

我们大家都注意到，网上现在到处都是AI写的垃圾内容。

And so, what we've all seen is, I would say, the rise of AI slop on the internet.

想想看，过去几年互联网上的内容增长得特别快。

So, if you think about that, the internet has grown a lot over the last couple of years.

其中很大一部分都是AI生成的。

A lot of that has been AI generated.

你我每天刷到的内容，我一眼就能看出来，这里面肯定有AI的痕迹。

You and I see content every single day where I go, there's definitely some tokens in there.

这些内容又反过来被拿去训练模型了。

And that also goes into the training of these models.

所以，现在AI模型的训练，核心环节之一就是对AI生成内容的提炼。

So, distillation of AI models is a core part of how these models are trained today.

这事儿根本躲不掉。

There's just no escaping that.

这是第一点。

That's number one.

第二点，我觉得大家讨论的问题是，有些模型正在大规模、工业化地进行提炼，比如用一堆虚假账号，或者有人转售云服务订阅并转发，各种方式都在违反服务条款，而且很糟糕。

Number two, I think the issue, I think people are talking about is that some of these models may be distilling at scale in kind of these industrialized ways where you have maybe a bunch of fake accounts, or maybe you have people reselling, say, cloud subscriptions to somebody else and routing it, any number of ways which are breaking terms of service and which are bad.

我的看法是这样的。

So, here's what I think.

第一，我认为所有实验室都应该在政府协助下，尽一切努力确保用户完成实名认证，追踪访问来源的IP地址等等。

Number one is I think all the labs, with the help of everybody from the government, should be doing everything they can to make sure whether they're doing KYC, whether they are doing things to check the IP addresses of where they're coming from.

他们其实还做了很多更复杂的事。

There are a lot more sophisticated things they do.

他们正在努力阻止这种大规模的推理过程掠夺行为。

They are making sure this heavy industrial extraction of reasoning traces isn't happening.

不过，我要借用红杉的迪恩·梅耶的一句话，他发过一篇特别棒的帖子，请大家在转发时记得注明他的贡献。

But, and I'm going to steal this from Dean Mayer of Sequoia, who had a fantastic post and please give him credit for this in his tweet.

他昨天刚写了这篇文章。

He wrote this yesterday.

我觉得现在的问题是，有些国外的模型可以拿美国的模型来训练。

I think the situation which is bad today is that some of these models from other countries can train off American models.

但如果你是美国的开源模型，反而可能搞不清楚能不能用其他美国模型的数据来做蒸馏。

Whereas if you are an American open weight model, it may be really confusing or challenging on whether you can distill off of other American models.

对吧？

Right?

所以现在这个生态很不平衡——如果你是中国的模型，大概能轻松拿到大量推理轨迹。

So, if you kind of have a really uneven ecosystem here where if you're a Chinese model, you could probably get a bunch of reasoning traces.

但如果你是个硅谷的新创公司，想用一些推理轨迹，又根本不知道法律上能不能这么做。

But if you are, say, a new Valley startup and you want to use some reasoning traces, you don't know what the legal situation is.

有个很棒的主意，我觉得是他提的，今天本·汤普森也说了类似的想法：我们能不能找到办法，让模型蒸馏在各种情况下都变得合法合规，不管是直接用其他模型的输出，还是更隐蔽的方式。

So, one great idea, which I think came from him, I think Ben Thompson of Strategy had a similar idea today, was to basically say, how do we find a way to make distillation acceptable in any number of ways, whether you are getting outputs of other models, or sometimes it's more subtle.

你看看现在任何一个美国的开源模型，其实都在用中国模型当老师，或者至少作为微调的一部分。

If you look at any American open source model today, they are using Chinese models as a teacher or in a way as a part of the fine tuning process.

那我们怎么确保这一点得到保障和确立呢？

And how do we make sure that is protected and enshrined?

所以，如果你是美国的模型公司，就能和中国模型站在同一起跑线上。

So, if you're an American model company, you have the same level playing field as the Chinese models.

不过，我们确实正处在一个非常独特而关键的时刻。

Now, again, I think we understand this very interesting, unique moment in time.

但总的来说，我觉得这些开源权重模型对整个生态非常有利。

But overall, I think these open-weight models are great for the ecosystem.

这带来了更多的选择。

It's more choice.

大家在价格上展开竞争。

It's it's people competing on price.

这正是创新本该带来的所有好处。

It's all the great things that innovation is supposed to bring.

对。

Right.

完全正确。

Totally.

另一个目前正在考虑的问题是，让人工智能先自动化地提升自身，然后再自动化地进行人工智能研究。

So another consideration that is going on right now is the idea of AI being able to automate to improve first and then automate AI research.

你知道，Anthropic 已经写过这方面的内容。

So, you know, Anthropic has written about this.

OpenAI 也提到过，先打造一个自动化的研究实习生，再逐步实现人工智能研究的自动化。

OpenAI has talked about, you know, building in, first, an automated research intern and then working towards automating AI research.

很多人认为，如果真能做到这一点，人工智能能力的进步速度会变得极其、极其、极其迅速。

And many people believe that if that were to happen, the pace of AI capabilities progress would go very, very, very fast.

那么，你认为政府是否应该为此制定什么样的政策？

So what kind of policies do you think the government, if any, should have around this?

如果未来真的出现人工智能能力的极速提升，你认为政府会采取什么措施？

What do you think the government is going to do around a potential very rapid increase in capabilities of AI if that were to happen in the future?

再说一遍，我已经不代表政府说话了。

Well, again, I don't speak for the government anymore.

说实话，我手里没什么内部消息。

You know, I don't have any inside information.

我不确定政府在这个问题上该不该插手。

I don't know whether the government has a real role in this.

比如RSI到底存不存在，你认为我们处在指数增长的哪个阶段，这争议可大了。

Like, whether RSI is real or not, where you are on the exponent is a much debated topic.

我听过不少观点，说再过几年就能出现能自主搞研究的AI了。

I've heard many, many schools of thought where they believe you're going to have automated AI researchers in a couple of years.

但也有人觉得，有些根本性的突破是没法轻易破解的。

But then there are others who think, well, there are fundamental improvements that you cannot crack.

没错，模型能训练其他模型，确实会带来进步，但增长曲线不会那么陡峭，只会慢慢往上走。

And yes, you're going to see improvements from models being able to train other models, but the exponent is going to be a lot more gradual rather than a very, very steep curve.

我当年在政府工作的时候，不管是对我还是对大卫来说，重点都是打造一个充满活力的生态，让大家竞相开发产品，把创新推到极致。

So, think when I was in government for me or for David, the entire focus was how do we have a fantastic ecosystem where there are people competing to build products, you are pushing innovation as fast as possible.

等到真有靠谱的风险出现时，再及时应对。

And then when there are credible risks, making sure when they appear, making sure you tackle them.

比如网络安全，一旦你遇到切实的威胁，知道这些模型能生成最新固件或操作系统的漏洞利用，你该怎么应对？

So, example, with cyber, when you have a credible threat, when you know these models are capable of, for example, generating exploits in the latest firmware or the latest operating system, how do you then go tackle it?

当有人问我这个问题时，我觉得它太抽象了，我更倾向于拆解成几个实际问题：第一，我们怎么确保有多种竞争性的解决方案？

So, when somebody gives me this question, I sometimes find it very theoretical and I often find it a lot more useful to break it down into, okay, one, how do we make sure we have a bunch of competing choices?

第二，哪些威胁是真正可信的？

Two is what do we know to be credible?

我们又该怎么具体应对这些威胁？

And how do we actually particularly tackle that?

我觉得网络安全、生物技术进步，还有其他几个领域，确实存在很现实的威胁。

I think there are very credible threats on cyber, on biological advances, a couple of other topics.

而且我也看到，已经有各种努力在针对这些问题了。

And I think there are definitely efforts to try and tackle just those.

最后我想说，我非常支持用人工智能来帮忙解决这些问题。

And I think the final part I would say is that in a lot of ways, I'm a big believer in using AI to help.

比如在网络安全方面，更好的办法往往是用更多AI来扫描你的代码库，提升代码的安全性。

So often, for example, the answer to helping with cyber is if you have, you know, using more AI to go scan your code base and making your code base more secure.

所以，我们通常得想，怎么在实际推进中应对这些威胁呢？

So, very often, I think like, how do we then deploy to actually meet these threats as we go along?

完全对。

Totally.

没错。

Totally.

还有，迪恩·鲍尔提到过一个观点，就是开源模型会降低资本支出，削弱前沿实验室靠技术领先盈利的能力。

So another point that Dean Ball raised actually was sort of this sentiment that open-weight models deter CapEx, and they kind of destroy the ability for Frontier Labs to monetize their capability leads.

那我们该怎么推动一个健康的开源生态，同时又让前沿实验室能继续发挥他们的优势呢？

So how do we, like, promote a very healthy, like, open source ecosystem while, you know, ensuring that the Frontier labs are still able to, like, do what they're best at?

我觉得归根结底，只要从商业本质出发，如果你提供的产品真有价值，市场自然会帮你打通整个供应链。

Well, I think at the end of the day, if you kind of bring it back to very business first principles, if you're providing a product of value, capitalism will find a way to make the supply chain work for you.

所以，如果你的开源模型确实有用，那下面每一层——不管是云服务商、数据中心、芯片厂商，还是给数据中心提供燃气轮机或消防系统的公司——都会主动围绕你来提供价值。因为你作为开源模型的提供方，可能正和银行合作，而银行不愿意用前沿实验室的模型。

So, if you have an open weight model that is providing value, that means that every part of the stack underneath, whether it is a Neo Cloud, whether it is a data center, whether it is a chip provider, whether it is somebody who provides gas turbines or fire suppression to a data center, every other part of that stack is going to orient itself to provide value because you, as an open weight model provider, you might be working with a bank because they don't wanna work with a frontier model.

他们更愿意用能自己部署的方案。

They want to work with somebody that they can run-in house.

或者你去看看Thinky这样的公司，他们正在用客户内部的数据对模型进行微调，而这些数据他们并不对外公开。

Or if you go look at what somebody like Thinky is doing, they are fine tuning models for specific clients using internal data that they don't expose.

所以，只要你提供的产品有价值，市场自然会解决其他所有问题。

So, kind of think of like, if you're providing a product of value, the capitalism will take care of all the rest.

顺便说一句，我觉得我的观点正在被现实验证。

And by the way, I think I'm being born out.

如果你看看整个生态链的其他部分，比如推理云的表现，再看看整个生态系统的整体发展，增长势头非常强劲，简直惊人。

If you go look at the fraction of every other part of the stack, if you go look at how the inference clouds are doing, if you go look at how the rest of the ecosystem is going, the growth is pretty strong and spectacular.

我相信这种趋势还会持续下去。

And I think you're gonna continue seeing that continue.

那么，既然你已经离开了政府，能跟我们说说你现在在忙些什么吗？

So, to the extent that you can tell us, what are you working on now that you're out of government?

我接下来会多做一些现场直播，可能还会做主播。

Well, I'm gonna be doing more, you know, live drop ins, I think, and I'm going to be Streamer.

不对。

No.

你看。

Look.

我觉得，你看。

I think well, look.

我有过一段非凡的人生经历，独一无二，实在是一种莫大的荣幸。

I I had this amazing life experience, which is unparalleled and it was such a unique honor.

这段经历让我深刻理解了国家和企业之间该如何协作，确保各方都能获得人工智能技术——无论是来自前沿实验室，还是开源模型。

And it's kind of really given me a sense of what countries and companies need to do to work with each other and to make sure everybody gets intelligence, whether it's from the frontier labs and open weight models or how that happens.

我想尽我所能，推动这件事实现。

And I want to try and make that happen in some shape or form.

所以，我可能还是会神神秘秘的，但我坚信，让美国及其盟友在政府与企业协同下，大规模获得人工智能技术，这个使命至关重要。

So, I'm going to be annoyingly elusive, but I think that mission of making sure America, our allies get access to AI at scale with having governments and these companies work together is a very important one.

多年来，我一直以各种方式在推动这件事。

I've been working on it in many different ways, if you think about it for many years now.

所以我打算继续做下去。

So, I wanna continue to do just that.

而且呢，多发点爆款帖子，时不时也去直播间互动一下。

And, you know, write some banger posts and, you know, jump in on, you know, livestreams from time to time.

是的。

Yeah.

我们真的很期待。

Well, we're excited.

这真是太棒了。

It's been so great.

非常感谢你，Sriram，来参加MTS。

Thank you so much, Sriram, for coming on MTS.

对。

Yes.

我们肯定还会再请你来的。

I'm sure we'll have you on again.

谢谢。

Thank you.

谢谢邀请我来。

Thanks for having me.

后面还有更多精彩内容。

Much more to come.

当然。

Absolutely.

感谢您收听这期 a16z 播客。

Thanks for listening to this episode of the a 16 z podcast.

如果喜欢这期节目，别忘了点赞、评论、订阅，给我们打分或写评价，也欢迎分享给亲朋好友。

If you like this episode, be sure to like, comment, subscribe, leave us a rating or a review, and share it with your friends and family.

想收听更多节目，请前往 YouTube、Apple Podcast 和 Spotify。

For more episodes, go to YouTube, Apple Podcast, and Spotify.

关注我们在 X 上的账号 a16z，并订阅我们的 Substack：a16z.substack.com。

Follow us on x at a sixteen z, and subscribe to our Substack at a16z.substack.com.

再次感谢收听，我们下期节目再见。

Thanks again for listening, and I'll see you in the next episode.

本信息仅用于教育目的，不构成任何投资或金融产品的买卖或持有建议。

This information is for educational purposes only and is not a recommendation to buy, hold, or sell any investment or financial product.

本播客由第三方制作，可能包含付费推广广告、其他公司提及以及与a16z无关的个人。

This podcast has been produced by a third and may include paid promotional advertisements, other company references, and individuals unaffiliated with a 16.

这些广告、公司和个人均未经Ah Capital Management LLC、a16z或其任何关联方认可。

Such advertisements, companies, and individuals are not endorsed by Ah Capital Management LLC, a sixteen z, or any of its affiliates.

所载信息基于发布时可信赖的来源，但a16z不保证其准确性。

Information is from sources deemed reliable on date of publication, but a16z does not guarantee its accuracy.