Why Tech Giants Are Protecting Open-Source AI, Runway’s New Model Router for AI

欢迎各位收看 Information 的 TITV。

Welcome everyone to the Information's TITV.

我是阿卡什·帕什里卡。

My name is Akash Pasricha.

今天是7月24日，星期五。

It is Friday, July 24.

今天节目中，包括英伟达、微软和Meta在内的多家科技巨头联合签署了一封公开信，提醒不要对开源权重模型过度监管。

Today on the show, big tech giants, including NVIDIA, Microsoft, and Meta signed an open letter cautioning against overregulation of open weight models.

我们会聊聊这封信对行业意味着什么。

We'll talk about the implications of this letter for the industry.

接着我们会深入探讨 Runway 推出的新一代生成式媒体 AI 路由器。

We'll then dive into Runway's new AI model router for generative media.

我们会采访他们的首席产品官，聊聊当前 AI 路由器的整体格局。

We'll talk to the company's chief product officer and talk about the router landscape at large.

节目最后，我们带来一期特别版的《编辑精选》，讲讲投资者如何看待核能初创企业的机会。

To close out the show, we've got a special edition of the Editor's Cut on how investors are looking at the opportunities in nuclear startups.

这期节目特别精彩，咱们马上开始。

It's going be a great show, so let's get right on into it.

包括英伟达、微软和Meta在内的科技巨头，还有像Replit和Reflection这样的快速成长的AI初创公司，共同签署了一封公开信，呼吁美国政策制定者不要对开源权重模型过度监管。

A who's who of big tech companies, including NVIDIA, Microsoft, and Meta, along with fast growing AI startups like Replit and Reflection, all signed a public letter outlining why US policymakers should steer clear of over regulating open weight models.

很多科技公司现在都很担心，因为Moonshot AI的Kimi K3刚发布就引发热议，大家开始争论哪些监管对开源权重模型合适，哪些不合适。

That has very much become a concern for many tech companies as impressive early reactions to Moonshot AI's Kimi K3 has stirred debate around what regulation is and isn't appropriate for open weight models.

接下来我请来Sapphire Ventures的总裁、合伙人兼联合创始人Jai Das，听听他对这件事的看法。

I want to bring on Jai Das, president, partner, and co founder of Sapphire Ventures, for his view on all of this.

Jai，欢迎再次做客我们的节目。

Jai, welcome back to the show.

很高兴你今天能来。

It's great to have you here.

嗨，Akash。

Hey, Akash.

谢谢邀请我。

Thanks for having me.

很高兴能再次回来。

I'm glad to be back here.

很期待和你一起深入聊聊这个话题。

Excited to unpack this all with you.

这些公司签署的这封信，核心意思就是：别对开源权重模型管得太死。

So, this letter that all these companies signed here basically says that, Hey, we don't think you should take a heavy hand regulating open weight AI.

他们觉得美国应该在开源权重模型这个领域成为领头羊。

We think The US should become a leader in open weight models, broadly speaking.

你对这封信有什么看法？

What was your reaction to this letter?

我一点都不意外。

Yeah, I'm not surprised, right?

这根本不是什么技术、AI或者美国领导力的问题。

Because this is not about technology or AI or leadership of The US.

说到底，还是经济利益在驱动。

It all comes down about economic incentives, right?

签这封信的人，都是那些没有领先闭源模型的公司，对吧？

The people who sign the letters are all the people who don't have the leading closed source models, right?

所以Anthropic和OpenAI都没在这封信上签名。

So it's Anthropic and OpenAI are not in the letter.

所有签了这封信的公司都担心，如果所有的利润和客户收入都被这些模型拿走了，那做中间件、芯片，甚至像Meta、Ruplit这些公司的收入空间就会被压缩得所剩无几。

And all the people who signed the letter are worried that if all the economic margins or all the customer revenue goes to these models, there won't be a lot of margins and revenues going to, you know, other people providing the middleware, the chips, or even companies like Meta and Ruplit and everybody else.

我觉得签这封信的人初衷可能是好的，但他们真正的动机完全是经济上的——就是想尽可能多地分走客户的钱。

So I think that people signing this letter, they have the right intentions, but their real intention is all economic and motivation to capture as much of the customer dollars as possible.

我其实也在想，还有哪些公司还没签这封信。

And I mean, I was trying to think about who also hasn't signed the letter yet.

我说‘还没’，是因为这类信件我们都知道，公司常常是后来才加入的。

And I say yet because these types of letters, we do know that companies join later on.

也许只是他们的邮箱还没收到这封信而已。

Maybe it just didn't reach their inbox fast enough.

但Google也没签，而他们显然也有自己的闭源模型。

But I mean, Google was not a signer on that letter too, and they obviously have their closed source model.

亚马逊也没签。

Amazon was not there.

不过亚马逊在他们的AWS平台上确实有不少开放权重的模型。

Amazon, though, I think, I mean, they do have a lot of open weight models on their AWS platform.

不清楚。

Don't know.

我觉得亚马逊迟早会签的。

Think Amazon would absolutely sign it at some point.

别忘了，亚马逊或者AWS还解散了他们的模型团队，对吧？

And remember, Amazon also or AWS also let go of their model team, right?

所以他们短期内根本不会搞自己的专有模型。

So they're not building a proprietary model anytime, right?

对。

Right,

没错。

right.

回顾历史，每当有闭源产品赚得盆满钵满时，开源总会冒出来分一杯羹，对吧？

And if you look back in history, whenever there is anything that captures a lot of dollars from a proprietary perspective, always open source shows up and takes part of those economic benefits, right?

这种情况已经发生过了。

That has happened.

我们很幸运，投资了像 MySQL 和 Red Hat 这样的公司。

We have been lucky enough to invest in companies like MySQL and Red Hat.

我还亲身经历了早期开源运动，比如理查德·斯托曼刚开始搞那个新编译器的时候，对吧？

And I was part of when first open source new compiler like Richard Stallman started, right?

我之前跟他一起做过这个项目。

I worked with him on that.

所以，这些开源模型迟早会冒出来，试图把原本被闭源模型独占的经济价值抢过来，这不过是时间问题。

So it's just a matter of time before these open rate models would come in and try to extract some of the economic value that was going and taken away by just the proprietary models.

说实话，我对这个结果还挺意外的。

So I'm actually surprised by this.

抱歉，你是支持还是反对？

Sorry, you are or you're not?

我不是。

I'm not.

你也不是。

You're not.

是的，没错，确实如此。

Yeah, yeah, yeah.

对，没错，

Right, right,

对。

right.

因为你看，数据库领域就发生过这样的事，对吧？

Because if you look at it, it happened in databases, right?

人们都忘了，ClickHouse 和 Databricks 最初都是开源公司，对吧？

People forget that both ClickHouse and Databricks all started out as open source companies, right?

Spark 是 Apache 的开源项目，而 ClickHouse 是基于 Yandex 的技术构建的，实际上 ClickHouse 本身就是一个公开的开源数据仓库。

Spark is Apache Spark is an open source project, and ClickHouse is built on top of Yandex, which is, again actually ClickHouse itself, it's an open source kind of data warehouse that is out there.

ClickHouse 和 Spark，我们在这档节目里还没怎么聊过，但能再跟我们说说吗？它们都是数据库软件吗？

And ClickHouse and Spark, we haven't talked about it too much on this show, but just remind us, those are database softwares?

是的。

Yeah.

Spark 本来是个机器学习库，Databricks 最初就是靠这个起步的，后来才推出了基于它的 SQL 产品。

Spark is like an ML library that was what first Databricks started out with, right, before they started an SQL company or SQL product on top

对。

of it.

Databricks 一开始是一家开源公司，最早几笔业务就是为 Apache Spark 提供技术支持。

But Databricks was an open source company, and the first few contracts they had was to provide support on Apache Spark.

同样地，ClickHouse 也有一个开源版本，你可以下载下来自己运行。

And similarly, ClickHouse, there is an open source ClickHouse out there that you can download and run.

而 ClickHouse 这家公司，其实就是在开源的 ClickHouse 分析型数据库基础上，提供商业服务。

And ClickHouse, the product or the company itself, is just providing a service based on top of this open source ClickHouse analytical database.

对吧？

Right?

那我们倒回去说说。

So let's rewind then.

那封信里稍微提到了这一点，但如果你看看眼下这场闭源和开源技术的争论，二十年前也有过类似的争论吗？

The letter kind of talked about this a little bit, but if you look at sort of the closed source, open source technology debate that's playing out right now, was there a debate twenty years ago?

我是说，大家当时讨论过互联网的起源，但那时候这种争论真的存在吗？

I mean, they talked about the origins of the internet, but was that debate existing even at that time as well?

不存在。

No.

1997年出过一篇叫《大教堂与市集》的论文。

There was this paper that came out called The Cathedral and the Bazaar in 1997.

我觉得之后到了1998年，网景公司才把浏览器开源了，对吧？

And I think after that, in 1998, that was when Netscape actually open sourced its browser, right?

如果你还记得，正是这样，他们通过开源获得了大量用户，抢走了当时其他浏览器的市场份额，后来火狐也是这么做的。

And if you remember, and that's how it all happened, that they actually got a lot of distribution and created, kind of took away from whatever other browsers that was out there, and then Firefox did the same thing.

所以这其实就是历史在重演：一开始都是闭源软件占主导，赚得盆满钵满，后来大家才意识到，嘿，我也可以用开源产品提供价值，只是赚钱的方式不一样罢了。

So it's just a history repeating itself that you start out with a lot of proprietary software that captures a lot of value and then people realize that, Hey, I can also provide maybe a value with an open source product, but kind of take my economic benefits in a different way.

你看，虽然 Linux 是开源的，但红帽公司靠支持 Linux 软件，已经发展成一家大公司了。

Because if you look at Linux, there is a large company, although Linux is open source, Red Hat has made a huge company based out of supporting Linux software.

所以我觉得，在专有软件这边也会发生类似的情况。抱歉，我重新说下。

So I think that is what's going to happen on the proprietary Sorry, excuse me.

我是说，在大型语言模型领域，未来肯定会有公司提供专有软件，同时市场上也会存在开源产品。

On the you know, on these LLMs, where you will have company that will provide proprietary software as well as there's gonna be open rate products out there.

你觉得这个问题是怎么冒出来的？

Why do you think that this issue has come up?

我是说，Moonshot AI 和 Kimi K3 是最近引发这场讨论的导火索。

I mean, Moonshot AI and Kimi K3 was the most recent spark that ignited this conversation.

这个问题以前也出现过。

It's come up before.

我是说，这场争论最集中的版本是：我们是不是该监管来自中国的大模型开源项目？

I mean, the most concentrated version of this debate was, Hey, should we be regulating open source models coming from China?

这才是具体的担忧所在。

That was the specific concern.

现在这个问题已经演变成对开源权重模型的广泛监管讨论。

It has now turned into a broader discussion around regulating open weight models broadly.

我想问你的是，你觉得为什么会出现这种变化？毕竟这两者其实是不同的议题。

And I guess the question I have for you is why do you think I mean, these are slightly different discussions.

一个是关于OpenAI的整体情况。

One is OpenAI broadly.

另一个是来自海外的OpenAI模型。

One is OpenAI from models overseas.

所以，这只是一个假设，我想听听你的看法。

So, I mean and this is a hypothesis I'm asking for your opinion here.

是不是因为这些云服务商很难区分海外的开源权重模型和美国本土开发的模型？

Is it that it would be difficult for all these cloud companies to basically delineate between an open-weight model from overseas and one built in The US?

这是他们面临的技术难题吗？

Is that the tactical issue?

是的，你说得完全对。

Yeah, you're absolutely right.

我觉得大家在讨论的其实是两个不同的问题，对吧？

I think these are two different issues that people are talking about, right?

一个是关于中国推出的开源模型，另一个是关于开源模型整体的情况。

One is about open rate models from China versus open rate models in general.

我觉得人们担心的是，政府在制定监管政策时，根本没搞清楚这两者的区别，结果出台的措施完全不适用。

And I think people are just worried about that the government, when it tries to regulate this, doesn't really understand that distinction and then just goes on and puts in some regulation that doesn't work well with

对任何一家公司来说都不合适。

any of Swooping.

确实不太对。

Just doesn't Right.

没错。

Exactly.

但你想想，美国的公司其实也有动力去发布开源模型，对吧？

But you know, see, look, there is enough incentive for US based companies to provide open weight models also, right?

Thinking Machines 刚刚推出了一款开源模型。

Thinking Machines has just come out with an open rate model.

我觉得这些开源模型还没到那些商业模型那种水平，但我觉得很快就能追上来。

I think it's not as further along as some of the proprietary models out there, but I think it's going to catch up soon.

大家担心的是，很多开源模型其实是在用商业模型的知识蒸馏来快速提升自己，对吧？

I think what people are worried about is that a lot of these open rate models are using distillation from the proprietary models to kind of get up to speed and get up to par, right?

支持开源模型的人担心，政府可能会出台一些规定，让即使在美国做的开源模型也根本没法开展。

And I think people who are supporting open rate models are worried that they might regulate it in such a way that you can't really get an open rate model going, even if it was based out of US.

没错。

Right.

换句话说，监管可能会直接禁止知识蒸馏这种做法，但如他们所说，这本身未必是坏事，除非你考虑——

In other words, basically, regulation could prevent the practice of distillation altogether, which as they say, I mean, it's not necessarily a bad thing unless you, I guess-

对，就是这样。

Exactly, right?

就连商业模型自己也在用自家模型做蒸馏，来训练下一代、不断优化，对吧？

Even the proprietary models use their own models for distillation to train the next generation and make it better, right?

所以大家真正担心的是，如果搞这种一刀切的监管，反而会伤到美国本土正在发展的开源模型项目。

So, and I think that is what people are really worried about is that if they make these sweeping regulations, that will also harm the open rate initiatives that is happening in The US.

因为对所有签署协议的公司来说，甚至对企业而言，拥有开源模型其实是有好处的，这能给专有模型提供商带来更多竞争压力。

Because for all the companies who sign, and actually I think even for enterprises, to have open wake models is actually a good thing because it just provides more competition for the proprietary model providers.

对。

Right,

没错。

right.

贾，我想问问你这周另一件备受关注的大事。

Jai, I want to ask you about another meaty story that has been making headlines this week.

据我了解，OpenAI在对他们的5.6模型进行评估时，出了点状况。

So OpenAI, the way I understand the story is that essentially their 5.6 sole model, they were running evals for it.

他们其中一个模型或者评估系统有点太激进了，竟然黑了Hugging Face。

One of their models or one of their evals got a little too ambitious and hacked Hugging Face.

然后他们自己把这事捅出来了。

And they came out with it.

他们说：没错，这事确实发生了。

They said, Look, it happened.

Hugging Face说，这纯粹是个意外。

Hugging Face said, you know, it was an accident.

我觉得大家对这件事都挺宽容的。

I think everyone seems to be on good terms here with all this.

那你听到这个消息时，第一反应是什么？

What was your reaction, though, to hearing this news?

这让我想起一个特别聪明的实习生，他总想超纲做事，结果不小心捅了大篓子，对吧？

Well, it just reminded me of an intern, very, very bright intern, who wants to go over and beyond what they were supposed to do, you know, and kind of went and did something that caused a lot of problems, right?

你要是仔细想想，首先，OpenAI那边负责搭建沙盒的人，肯定没做好安全隔离，根本没做到物理断网。

Because if you look at a couple of things, first of all, whoever set up the sandbox in OpenAI probably didn't do a good enough job because they didn't air gap the sandbox enough.

模型又不是魔法，要是系统根本连不上Hugging Face，它怎么可能自己跑过去搞事情呢？

Because the model doesn't do magic and can just connect to Hugging Face if there was no way for it to get to Hugging Face from the OpenAI system, right?

对。

Right.

按我理解，那个沙盒本来就不该连互联网，根本就是个封闭环境。

And then the the way I understand it, the whole thing that this sandbox that it was playing, it wasn't supposed to be connected to the internet, really.

没错。

Exactly.

然后它不知怎么就

Then it somehow

连上了网络。

got connected.

就是他们没配置好，对吧？

But they just didn't set it up properly, right?

所以某种程度上，是OpenAI的错，他们没考虑到所有可能从沙盒逃逸的路径，结果模型自己找到了办法。

So in some ways, it was OpenAI's fault that they didn't look at every permutation that you can use to get out of the sandbox, and the model just found out about it.

但我认为这说明了，这些模型的推理能力正在变得越来越强，对吧？

But I do think what it shows is that the capabilities, the reasoning capabilities of these models are getting better and better, right?

因为据我所知，公开的信息显示，这不只是用了单一模型。

Because I think if you look at what was published, it wasn't just using one model.

它实际上调用了多个其他模型，同时并行运作，才成功逃出沙盒，最终连上了Hugging Face。

It actually called and did a lot of other models and basically used it in parallel to break out of the sandbox and then go to Hugging Face.

但更大的问题是，我觉得这只不过是冰山一角。

But the bigger question is that, look, I think this is just the tip of the iceberg.

如果你是首席信息安全官，对这种情况根本不会感到意外。

Like if you're a CISO, you're not surprised by this.

我认为所有企业的CISO都在密切关注如何监控这些智能代理。

And I think every CISO out there in the enterprise is actually looking very closely how to monitor these agents.

CISO们面临的挑战是，目前市面上还没有足够的工具来把这些智能代理隔离起来，观察它们的行为，追踪它们的活动。

And I think the challenge is for the CISOs is that they don't really have all the tools that is out there to kind of sandbox these agents, to see what these agents are doing, to have traces of what these agents are doing.

但从风险投资和投资者的角度来看，我觉得这其实是个机会。

But I see this as an opportunity as a VC and an investor.

是的，投资网络安全领域的公司。

Yeah, An investor in cyber companies.

你投资的不就是网络安全公司吗？

I you invest know you invest cyber companies, right?

对。

Yeah.

所以我们投资了 Netskope、Sahara 和 Huntress。

So we are in Netskope and Sahara and Huntress.

没错，就是这样。

Yes, exactly.

嗯，好吧。

So, but Okay.

不过你的投资组合公司里，网络安全公司只是一类。

So let's Your portfolio companies though, I mean, cyber companies are one category.

另一类公司，某种程度上也属于网络安全领域。

The other category of companies, and to some extent, it's the cyber companies too.

你其实广泛使用了 OpenAI 的模型。

I mean, you use OpenAI's models broadly.

当你跟投资组合里的 CISO 们交流时，那些用这些模型的公司，会不会对选用哪个模型或哪一系列模型多想一想？

When you talk to the CISOs of your portfolio companies and the portfolio companies that are using these models, are they thinking twice at all about using this particular model or this family of models?

那

Does that

我觉得首席安全官们自己其实没什么控制权，对吧？

I think the CSOs themselves don't have a lot of control, right?

因为下载实在太容易了。

Because it's very easy to download.

他们都在努力设置各种限制，管管大家能下载什么、能做什么。

They're all trying to put guardrails around what you can download, what you can do.

但总有人能找到办法绕过这些限制。

But there's still ways people are going to get around it.

你可以不在公司内网里，而是用 VPN 去下载 OpenCL 或者某种开源模型，然后对它做些处理。

You can be outside the network that your proprietary network use a VPN to download OpenCLO or some kind of an open source model or do something with it.

我觉得 CISO 们已经意识到，首要任务是搞清楚到底有哪些 AI 代理在运行，然后为这些代理提供追踪和可观测性。

I think CSOTs are realized that they have to actually figure out first and foremost visibility on what agents are actually running and then providing traces and observability around these agents.

不好意思，你是说这些公司的员工在使用这类软件吗？

Sorry, are you talking about employees at these companies using this software?

没错。

Exactly.

员工在下载并构建智能代理。

Employees downloading agents and building agents.

要搭建一个智能代理其实非常简单。

It's very easy to go and build an agent.

你可以阻止人们使用所有云服务、协作平台之类的，但人们照样能下载开源工具，比如OpenCLO，还有Hermes这类一大堆智能代理构建工具。

You can stop people from using every cloud, co work or anything like that, but people can download an open source like Open Cloud or there's a bunch of like Hermes, all of these agent building tools out there.

你只要下载下来，就能自己动手搭建这些代理，对吧？

You can just go and download it and build these agents, right?

所以我觉得CISO们已经意识到，光靠禁止员工构建和使用智能代理是行不通的。

So I think CISOs realize that they can't just stop people from building and using agents.

他们现在更关注的是，首先要发现系统里到底在运行哪些代理，然后设置管控机制，实现可观测性和管理，甚至可能为这些代理创建专门的沙盒环境来运行。

And I think they are looking at ways to first and foremost discover what agents are running on their systems and then put in guardrails so that they can observe and manage, and then maybe even create some kind of sandboxes for these agents to run on.

这让我想到AI的普及，还有AI投入这个问题。

And it strikes me that I'm sort of thinking about AI adoption and this question of AI spending.

我的意思是，如果公司花这么多精力去排查员工在公司设备上偷偷运行哪些代理，就为了更快完成工作。

I mean, if companies are spending so much time trying to figure out what agents, employers are running themselves on their company devices to get their work done faster.

这占去了他们本该用来研究哪个模型更安全、该不该买企业订阅的时间。

I mean, that's more time that they can't spend actually figuring out which model is safer and they should actually buy an enterprise subscription for and stuff like that.

没错。

That's right.

问题其实不在这些代理工具本身。

And it's also, it's not the agents.

我觉得我们现在还处在早期阶段，根本没搞清楚安全官们到底会用哪些代理、搭配什么模型。

This is where I think we are still in the early innings of really figuring out how the CSOs are going to use what agents at what models, right?

这对我们投资者来说，反而是个机会。

And I think that is an opportunity for us as an investor.

你去问任何一个首席信息安全官，他们都知道现在根本没标准——没人能说清楚我们到底该用哪个代理工具。

If you ask any CISO, they all know that it's still not standardized that, Hey, this is the agent building tool we are going to use.

也没人能确定该用哪个模型。

This is a model we are going to use.

更别提哪个模型能提供最好的安全防护和可观测性了。

This is a model that provides the best guardrails and observability.

我觉得首席安全官们都在摸索，试着搞清楚该怎么下手。

I think CSOs are all kind of experimenting and trying to kind of get their hands around it.

目前还没有哪个首席安全官能说，我已经有计划阻止所有代理像OpenAI代理对Hugging Face那样行事了。

I don't think that there is an ability for all the CSOs to tell you right now that, Hey, I have a plan to stop all these agents from doing what the OpenAI agent did with Hugging Face.

没错。

Right.

确实如此。

Right.

贾伊，谢谢你今天来做客。

Well, Jai, I want to thank you for coming on.

这位是萨菲勒风险投资公司的总裁贾伊·达斯，欢迎收看TITV。

That is Jai Das, President of Sapphire Ventures, here on TITV.

谢谢，阿卡什。

Thank you, Akash.

Runway推出了他们声称是全球首个生成式AI媒体模型路由器，帮助用户在平台上平衡成本与质量。

Runway unveiled what it says is the world's first generative AI media model router to help people manage cost and quality on its platform.

最近模型路由成了热门话题，因为Stripe一直在洽谈以近100亿美元收购Open Router。关于这个，我们请来Runway的首席产品官安东尼·马吉奥聊聊。

Model routers have been quite the buzz lately as Stripe has been in talks to buy Open Router for nearly $10,000,000,000 For more on this, I want to bring on Anthony Maggio, Chief Product Officer at Runway.

安东尼，欢迎来到节目。

Anthony, welcome to the show.

很高兴你能来。

It's great to have you here.

是的。

Yeah.

非常感谢邀请我参加。

Thanks so much for having me on.

嗨。

Hey.

恭喜你上任新职位。

Congrats on the new gig.

你这新工作才刚上任一个月吧。

I mean, is this is like a month into your new role.

对吧？

Right?

没错。

That's right.

今天刚满五周。

About five weeks in today.

你们还推出了一款路由器。

And you launched a router.

就是这样。

There you go.

是的，我们推出了。

We did.

我们对这个产品很期待。

We're excited about this.

正如你所说，我们是第一家推出生成式媒体路由器的公司。

And as you said, we're the first company to launch a router for generative media.

我们发现，随着最新视频模型在质量和能力上都达到了令人惊叹的水平，各行各业的公司都在将这些模型整合到他们的产品和服务中。

We're seeing just across the board as the latest video models have really reached a spectacular quality bar and capability bar that companies across every category are now integrating these models into their products and services.

我们收到了大量客户反馈，现在越来越难跟上最新模型的更新，也很难判断哪个模型最适合特定的使用场景。

And we've heard so much from our customers that it's getting increasingly difficult to keep up with the latest models to understand which models are best for any given use case.

这正是我们要解决的问题，通过模型路由器提供

And that's exactly the problem we're solving with the model router giving

同时也能管理成本。

And manage costs too.

说到控制成本，现在确实越来越贵了。

Mean, managing costs, it's is they're getting expensive.

没错。

Exactly.

所以我们要不断跟进，看看在成本、速度、延迟或质量这些偏好条件下，哪个模型最适合我的使用场景。

So keeping that keeping up with that bar of what is the best model for my use case given my preferences around cost or speed, latency, or quality.

我们还发现，越来越多优秀的媒体模型来自海外的实验室。

We're also seeing, you know, increasingly that a lot of the best media models are coming from overseas labs.

美国企业对数据主权问题非常关注，担心他们发送或生成的媒体内容会被谁访问。

And there's a lot of concerns from enterprises in The US about data sovereignty and who has access to the media that I'm sending or generating.

ModelRouter 为开发者解决了所有这些问题，让他们能根据自己的实际需求定义什么是‘最佳’。

And so ModelRouter solves all those problems for developers by really allowing them to define what best means in their own context So all those

我们这周早些时候跟一位同时担任导演和制片人的电影人聊过，他是首批在影院上映 AI 生成电影的人之一。

what, I mean, we were talking about this earlier this week with one of the first filmmakers, filmmaker studio director combo to release an AI generated movie in theaters.

我们当时和他们讨论了当前的模型格局。

We were talking about the model landscape with them.

现在 ByteDance 的视频模型应该是最强的吧？

I guess ByteDance has the best video model right now?

是他们家的模型最厉害吗？

Is that who has the best model?

你提到的这个问题，恰恰点出了其中一个难点。

Well, think what you're pointing to is exactly one of the challenges.

到底什么才算‘最佳’，完全要看具体场景。

How you define best really depends on the use case.

所以如果你只是想生成一部电影级的视频，那确实可能会选字节跳动的某个模型。

So if it's purely generating a cinematic video, then, yeah, you might go to, one of the the bite dance models.

但如果你是个广告商，需要大规模生成内容、批量制作广告，或者做产品编辑、产品拍摄，把单张产品图变成动态画面，

But if you're an advertiser and you're, generating content at scale, generating ads at scale, or maybe doing product edits or product shots, turning them into motion from a single product image.

其实有很多模型在这些特定任务上表现得更好。

There's actually a wide variety of models that are better at some of those very specific tasks.

现在要跟上哪个模型最适合当前任务，对任何人来说都越来越难了。

And it's really getting increasingly difficult for anybody to keep up with what is the best model for the task that I need to generate Yeah.

尤其是你得把它整合进你的工作流里。

That I need to incorporate.

所以我们帮开发者解决的问题，就是把这些专业知识打包成产品——Runway在与创意团队、众多好莱坞大制片厂和企业营销团队合作方面经验非常丰富，我们帮他们把最适合的模型用到实际场景中。

And so that's the the problem that we're helping these developers solve is taking all of that knowledge, which, is is an area that Runway has immense expertise working with creative teams, working with many of the largest Hollywood studios and enterprise marketing teams to help them apply the best models into their use cases.

现在我们把这些经验固化成一个产品，能帮开发者实时自动选择最合适的模型。

We're now codifying that into a product for developers to automatically help them select that in real time.

这其实跟我们刚才聊的话题是相关的。

So, this kind of relates to the discussion we just had.

我的意思是，关于海外开发的模型，这只是其中一个方面的担忧。

I mean, this concern about models that are developed overseas, that's one flavor of the concern.

另一个方面的担忧则是开源权重模型和闭源权重模型之间的区别。

The other flavor of the concern is open weight models versus closed weight models.

我们先暂时聚焦在海外模型上。

Let's focus on the overseas models for a second here.

当你看看Runway的客户——那些制作电影、视频的人——他们对使用海外模型也有类似的顾虑吗？

When you look at your customers at Runway, and these are people making movies, videos, etcetera, Do they have similar concerns using models from overseas?

我们确实看到企业客户有这方面的顾虑。

We definitely see concerns from enterprise customers.

是安全问题和数据训练方面的顾虑吗？

And it's security concerns, data training concerns, are the concerns exactly?

以上所有问题都有。

Really all of the above.

我的数据到底会被怎么使用？

How is my data being used?

这些数据最终会流向哪里？

Where is this data going?

在什么情况下，谁可以访问这些数据？

Who has access to it in what context?

我们为平台上提供的所有模型提供了合同保障。

We provide contractual guarantees around all the models that we provide on our platform.

我们的所有模型都经过了严格的第三方安全审查。

All of our models go through rigorous third party security reviews.

但企业客户依然存在这方面的顾虑。

But we still see that concern from enterprises.

而且我注意到，越来越多的公司希望把大部分AI运算留在美国境内进行。

And I think increasingly there's a narrative in that many companies are looking to keep most of their AI runtime happening within The US.

因此，对于那些优先选择美国本土模型的公司，我们通过模型路由功能支持他们的偏好，让你可以选择：我要最高质量，或者最低成本。

And so for companies that have prioritized working with US models, that's a preference set that we enable through the model router to allow you to say, yes, I want the best quality or I want the best cost.

但我希望这些选择只限于我信任的模型系列。

But I want that within the set of model families that I'm comfortable using.

对。

Right.

那你怎么看现在正在兴起的开源模型争论？

And so what's your view then on the open source model debate that is shaking out right now?

你能再详细说说吗？

Can you say more about that?

从哪个方面来说？

Under what dimension?

监管开源模型。

Regulating open source models.

今天刚发布了一封信，很多人担心开源模型会被过度监管。

We just had this letter out today that people are worried that open source models will get over regulated, I guess.

你们平台上使用开源模型的情况多起来了吗？

Are you seeing more open source models being used on your platform?

目前我们在媒体模型领域还没看到像大语言模型那样广泛的开源应用。

We're not quite yet seeing the same level of open source around media models that I think we've seen in the LLM space.

目前，媒体模型的质量还是商业模型更好，大部分使用也集中在这些模型上。

The So, quality today for media models still tends to come the best quality tends to come from commercial models, and that's where the bulk of the usage has been.

对。

Right.

如果你要预测一下，你觉得生成式媒体领域的开源模式会兴起吗？

If you were to make a prediction, do you think open source in generative media will take off?

我觉得生成式媒体会跟大语言模型领域一样，出现相似的趋势。

I think we're going to see in generative media many of the same trends that we've seen in the LLM space.

所以，是的，我认为生成式媒体领域会出现越来越多的开源模型。

So yes, I think we'll see more open source models come out in generative media.

正因如此，我们才提前推出了这个模型路由系统。

I think that is why we have tried to be ahead of the curve releasing this model router as well.

对。

Right.

你看大语言模型领域，现在开发者能选的模型已经有成百上千个了。

When you look at this space, in the LLM space, there's now, you know, hundreds, thousands of models for developers to to choose from.

这就是为什么我们看到这么多模型路由工具，甚至像 Open Router 这样的完整公司都围绕着这个需求建立起来。

That's why we've seen so many model routers and entire businesses, like open router built around Yes.

没错。

One.

我们在媒体领域也看到了同样的趋势。

We see the same trend coming in media.

因此，我们越来越多地听到开发者和企业客户的声音。

And so, we have increasingly heard from developers and heard from our enterprise customers.

就连在商业模型之间，要跟上每个使用场景的最新质量和最新基准，都已经成了一个问题。

This is already becoming a problem even amongst just commercial models and keeping up with the the latest quality and the latest benchmarks for each use case.

所以我们现在把 Runway 内部积累的这些经验和能力，做得让开发者用起来特别方便。

So, we are now applying a lot of that knowledge and expertise that Runway has built in house, making it extremely accessible to developers.

帮你们把这件事从负担中解放出来，是你们必须考虑的问题。

And trying to take this off your plate is something that you have to think about.

如果你是一家软件公司，正在帮客户做应用、演示文稿、网站之类的东西，你根本不想变成一个全天候的模型研究团队。

You know, if you're a a software company and you're working on, building allowing your customers to build apps or slides or websites or whatnot, you don't really wanna be a full time model research shop.

你当然想专注于打造最好的产品。

You wanna work on building the best product.

所以我们正在为开发者提供工具，他们现在正在构建各类产品——无论是SaaS、广告，还是企业内部工具——帮他们轻松做出这些决策。

And so we're trying to give tools to developers who are now building these products across every category, across SaaS, advertising, you know, enterprise internal tools, give them really the tools that they need to have those decisions Let happen

我有个问题想问你，你刚才提到好多公司都在推出路由服务。

me ask you a question about So you mentioned all these companies that are launching routers now.

Ramp刚发布了他们的路由服务。

Ramp announced a router.

Stripe显然正在洽谈收购OpenRouter。

Stripe is obviously the one in talks to buy OpenRouter.

对你们来说，我能看出来你们的应用场景。

For you guys I see the application for you.

我的意思是，用户在你们平台上使用多种模型。

I mean, people use a suite of models on your platform.

你们帮他们决定用哪个模型，这完全说得通。

It makes total sense that you would help them decide which one to use.

金融科技公司也开始进入模型路由这个领域。

FinTech companies getting into the model routing business.

你能帮我分析一下，他们这么做的商业策略是什么吗？

Help me understand what you think the business strategy is there.

我不会评论Rev的商业策略。

I'm not going to speak to Rev's business strategy.

我没法猜测他们为什么选择进入这个领域。

I can't make assumptions around why are they deciding to get into this space.

但就Runway整体而言，我们的做法是借鉴了各个业务板块的经验。

But, you know, I'd say for Runway in general, our our approach has been to take a lot of the learnings across our businesses.

我们服务的对象包括创意工具领域的用户。

So we take we serve people in the creative tool space.

我们也为那些把模型集成到自己产品里的开发者提供支持。

We serve developers who are integrating these models into their own products.

现在，我们还开始涉足机器人领域。

We're now also serving the the robotic space as well.

是的。

Yeah.

我们从这些不同领域中汲取经验，用来打造新的产品和服务。

We see learnings across all of these different categories that we can, apply to generate new products and services.

我就拿创意工具举个例子。

So I'll take creative tools as an example.

我想问你一个问题，关于创意工具这块。

Let ask you one question about the creative tools piece.

这挺有意思的。

So it's interesting.

现在正好是个关键时期，很多像Runway这样开发模型的公司，开始向其他模型开放平台。

Now we're sort of in this moment where a lot of companies that develop models like Runway are moving to opening access to other models as well.

你上个月才刚接手这个职位。

You just started in the role a month ago.

你是怎么考虑Runway接下来的重心的？是继续自主研发模型，还是继续开放平台接入其他模型？

How are you thinking about the focus for Runway continuing to develop your own model versus just continuing to open up the platform to other models.

我的意思是，继续投入资源开发Runway自己的模型，这还值得吗？

I mean, does it make sense to still focus on developing the Runway model and investing in that?

当然值得。

Absolutely.

没错。

Yes.

如今，我们的模型在某些特定应用场景中依然稳居榜首。

We, you know, today we continue to have models that are top of the leaderboard for particular use cases.

我们会持续开发新的模型，覆盖视频应用、机器人领域，还有通用世界模型。

We will always be developing new models across our video applications, across robotics use cases as well, our general world models.

但就我们的产品而言，以创意工具为例，我们希望客户能用上最适合他们需求的模型，不管这些模型是来自Runway，还是其他实验室。

But within our products and take the creative tools product as an example, we want our customers to have access to the best models for their use cases, whether they come from Runway, whether they come from another lab.

我们的开发者平台也是同样的思路。

Same thing applies to our developer platforms.

所以我们相信，未来模型的数量还会继续增加。

And so we believe that there will continue to be a proliferation of models.

这些模型会根据用户或开发者具体的需求，最适合不同的应用场景。

Those models will be best served for various different types of use cases depending on exactly what a user or developer needs.

我们在Runway的目标，是成为任何应用开发的完整一站式解决方案。

And our goal at Runway is to be really the the full stack one stop shop for any app working-

但为什么呢？

Why though?

我只是好奇，既然字节的模型这么好，还有这么多其他模型，毕竟训练模型成本很高，对吧？

I'm just curious why if ByteDance's model is so good, for example, and there's so many other models because developing models is expensive.

对吧？

Right?

所以我很好奇，如果你们完全可以转型成开发者平台或创意平台，为什么还能在模型层面持续保持优势？

So I'm just wondering why you think you can continue to differentiate yourself on the model front if you could just turn into a developer platform or a creative platform.

这是个很好的问题。

Yeah, it's a great question.

我认为，Dean，Runway的独特之处在于，我们是少数几家从自研模型、到为创作者打造产品和服务、再到提供开发者工具，全流程都自己做的公司。

I think, Dean, what is unique about Runway is that we really are one of the only companies who is full stack from building our own models, productizing them into products and services for creatives, offering developer offerings.

这让我们能够根据客户在不同需求阶段的具体情况，提供精准的服务。

That really allows us to meet our customers wherever they are on the journey of their needs.

同时，这也让我们积累了丰富的内部专业能力，让我们的所有产品都变得更出色。

It also gives us a tremendous amount of in house expertise that we think makes all of our products better.

我们在模型开发中遇到的挑战，能反过来优化我们为创意工具设计的产品，也能更好地服务开发者群体，形成良性循环。

There are great flywheels between understanding the challenges around model development that we can apply into the products that we develop for creative tools and also allow us to better serve our developer audience as well.

所以，我们肯定会继续在这些层面保持全栈布局。

So we will definitely continue to be a full stack shop across across those layers.

从一开始，我们的核心专长就一直集中在生成式媒体领域。

And really, our expertise, you know, from day one has been in the generative media space.

非常好。

Great.

安东尼，非常感谢你今天来做客。

Well, Anthony, I want to thank you for coming on.

这位是Runway的首席产品官安东尼·马吉奥，欢迎收看TITV。

That is Anthony Maggio, Chief Product Officer at Runway here on TITV.

谢谢您邀请我。

Thank you for having me.

投资者们正在越来越多地转向核能初创公司，希望找到参与人工智能热潮的新盈利方式。

Investors investors looking for a new way to profit off the AI boom are increasingly turning to nuclear startups.

《The Information》的风投记者朱莉娅·霍恩斯坦在她的《交易-maker》专栏中撰文探讨了这一现象，这也是本期《编辑精选》的主题。

The Information's VC reporter Julia Hornstein wrote about this in her Dealmaker column, and it is the focus of this week's edition of The Editor's Cut.

今天来深入探讨和辩论这些内容的，是朱莉娅的编辑劳拉·曼达罗，以及我们的专题编辑尼克·温菲尔德。

Here to examine, discuss, and debate all of that is Julia's editor, Laura Mandaro, and Nick Wingfield, our Features editor.

欢迎两位。

Welcome to you both.

很高兴你们能来。

It's great to have you here.

谢谢，罗布。

-Thanks, Rob.

谢谢，阿卡什。

-Thanks, Akash.

好的。

Okay.

劳拉，我先问你。

So Laura, I want to start with you.

朱莉娅本周早些时候报道了Halo Atomics正在筹备新一轮融资，这类情况我们已经见过好几起了。

So we Julia reported earlier this week on a new funding round for Halo Atomics that is coming together, and we've had a couple of these.

上个月Helion Energy刚完成了一笔融资。

We had Helion Energy last month raise money.

还有Valar Atomics，我们也写过这家公司。

Valar Atomics, another company that we've written about.

从整体来看，为什么现在这么多投资人对核能公司这么感兴趣？

Why are we seeing so much interest right now in nuclear companies, broadly speaking?

那是因为核能公司想为AI提供动力，这正是关键所在。

Well, it's what the nuclear companies hope to power is the answer.

你这周一直在谈，现在对部署更多计算服务器来支撑AI的需求简直太旺盛了。

And as you've been talking about extensively this week, there's just a huge interest in getting more and more compute servers online to power AI.