本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
欢迎各位收看《信息时报》TITV节目。
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22日，星期三。
It is Wednesday, July 22.
本期节目，我们独家披露：XAI——SpaceX的人工智能部门，正在德州筹建一座大型数据中心。
Today on the show, we have exclusive reporting that XAI, SpaceXAI's AI division, is working on a big new Texas data center.
接下来，我们将连线我们的埃隆·马斯克特派记者，带来更多内幕细节。
We'll bring on our Elon Musk reporter with more details on that effort.
我们还独家报道了甲骨文公司在新建数据中心过程中遇到的阻力：意外开支和繁复的审批流程让项目举步维艰。
We also have exclusive reporting on how Oracle's push to build new data centers is hitting roadblocks facing surprise costs and heavy red tape.
此外，我们也会看看Neo Cloud通用计算业务如何开始摆脱对英伟达的依赖，全力推进自己的Neo Cloud平台。
We'll also take a look at how Neo Cloud General Compute is looking beyond NVIDIA as it seeks to build out its Neo Cloud offering.
节目最后，我们将专访一部可能成为史上首部院线上映的AI生成电影的导演与制片人。
And to close out the show, we'll talk to the director and producer behind what could become the very first AI generated movie to hit theaters.
这期节目特别精彩，咱们马上开始。
It's going to be a great show, so let's get right on into it.
SpaceXAI正在德克萨斯州建设一个新的数据中心。
SpaceXAI is working on a new data center in Texas.
这是我的同事格蕾丝·凯提供的独家报道，她一直跟进埃隆·马斯克的动态。
That is according to exclusive reporting from my colleague Grace Kay, who covers all things Elon Musk.
我请格蕾丝来跟大家详细说说我们了解到的情况。
I want to bring on Grace to share more about what we know.
格蕾丝，欢迎再次做客我们的节目。
Grace, welcome back to the show.
很高兴你今天能来。
It's great to have you here.
谢谢。
Thanks.
好的。
Okay.
那么，关于埃隆·马斯克在德克萨斯州建设的这个数据中心，我们现在知道些什么？
So what do we know about this data center that Elon Musk is building in Texas?
是的。
Yeah.
几个月来，公司内部一直在讨论在德克萨斯州扩建数据中心的事。
So they've been talking about a data center expansion Texas for a few months internally.
他们考察了几个地点，正在考虑在该州建造多个数据中心，可能是从零开始新建，也可能是改造仓库——就像他们在孟菲斯建第一个数据中心时那样。
They've been exploring a few sites and are considering potentially building multiple data centers in the state, including building it from scratch as a greenfield project and repurposing a warehouse, which is what they did for their first data center in Memphis.
我们有没有关于这个数据中心规模的具体信息？
Do we know how big this data center is supposed to be at all?
是的。
Yeah.
现在还太早，所以得说明一下，这些信息可能会有变动。
So it's very early, so there's that caveat that this could change.
但根据我从消息源那里听到的情况，他们希望这个项目的规模能和孟菲斯的一样大，甚至更大。
But based on what I've heard from sources, they want to build this on the same scale as Memphis or potentially larger.
我们这档节目里已经聊过很多次孟菲斯的设施了。
So we've talked about the Memphis facilities a lot on this show.
我觉得有一个已经建好了，还有两个正在建。
I think there's one that's built and two more that's being built.
是这样吗？
Is that right?
对。
Yeah.
它们都处在不同的建设阶段。
They're all in various stages of being built.
有一个已经完全投入运营了，只用了一百二十二天就建好了，这在业内是破纪录的。
There's one that's fully operational, was built in about one hundred and twenty two days, which is a record for the industry.
不过没错，这些设施分布在孟菲斯和附近的南哈文地区。
But yes, they're spread out across Memphis and South Haven nearby.
所以这将是SpaceXAI正在推进的第四个设施了。
So this would be, I guess, the fourth facility that SpaceXAI would be working on.
咱们来聊聊 SpaceXAI 的 AI 产品需求吧。
I mean, let's talk about the demand for SpaceXAI's AI products.
Grok 这块，我觉得你也提到过，需求目前有点不稳定，那他们怎么还敢在这时候大力建设这么多设施呢？
I mean, Grok, we've I I think you've reported as well that demand has been a little iffy on that front, so talk a little bit about how they're considering building all these facilities when it's looking a little iffy for Grok right now.
是的。
Yeah.
我觉得他们现在跟 Anthropic 和谷歌达成了算力合作，这也让他们意识到，自己的核心优势其实是能超快地建成数据中心。
I think we're seeing these compute deals with Anthropic and Google, and I think that's something that SpaceXAI has kind of realized as their secret sauce is they're really good at building these data centers really quickly.
就像我说的，第一个数据中心只用了一百二十二天就建好了，这在业内简直是闻所未闻。
You know, like I said, the hundred and twenty two days for the first one was pretty unheard of in the industry.
所以他们看到很多公司愿意花大价钱租用这些算力资源。
So I think they're seeing the capacity from other companies that are willing to pay, large amounts of money to rent this compute.
Grok 现在的需求可能确实不如其他 AI 公司那么旺，但马斯克已经表示，他们会继续推进这个项目。
So yes, Grok maybe doesn't have as much demand as some of the other AI companies right now, but Elon Musk has indicated that they're gonna continue to work on that.
不过，他们还有一个细分业务方向，就是靠快速建造数据中心来赚大钱。
But also, there's this subset of the business where they can make a lot of money on their ability to build these data centers quickly.
没错。
Right.
我的意思是，确实能赚不少钱，但说起来有点讽刺。
I mean, can make a lot of money, but it's kind of funny.
作为一个云服务公司，这恐怕不是SpaceXAI最初的目标。
Being a Neo Cloud, I don't think it was the vision for SpaceXAI to begin with.
所以某种程度上，我也好奇他们为什么还要继续下去。
And so in some ways, wonder why keep going.
我不太清楚。
I don't know.
我想知道，一线员工是不是也在问：这真的是公司的新方向吗？
I wonder if people on the ground, if they're telling you, like, this the new vision for the company, really?
还是说，只是因为埃隆说了算，他们才硬着头皮往前推？
Is it just something that they are pushing ahead with because Elon is telling them to?
到底他们的战略是什么？
Like, what what's what's the strategy here?
是的。
Yeah.
这确实是马斯克对公司愿景的一部分。
I mean, it's definitely part of, like, the Elon Musk vision for the company.
他多次告诉员工，赢得人工智能竞争的关键就是拥有最多的计算资源。
He has repeatedly told staff that, like, how they're gonna win the AI wars is by having the most compute.
我们已经看到像Anthropic这样的公司，因为计算资源受限而转向其他方向。
And we've seen with other companies like Anthropic, compute constraints have led them to look elsewhere.
所以这确实是一个优势，也是马斯克正在推进的方向。
So it is an advantage and it's something that Elon Musk is pursuing.
但没错，建造这些数据中心也非常昂贵，这也是另一个因素。
But yes, it's very expensive to build these data centers also, so that's another factor.
我们来谈谈人们对数据中心的反应。
Let's talk about the reaction to data centers.
在孟菲斯，当地对这些设施的建设已经出现了一些反对声音。
So in Memphis, there's been some backlash to these facilities getting built.
我们在德克萨斯州也会遇到类似的情况吗？
Are we expecting some of that in Texas as well?
这很难说。
It's hard to say.
德克萨斯现在简直就是数据中心的法外之地，看起来是这样。
I mean, Texas is kind of the Wild West for data centers right now, seems like.
很多公司都在那里建厂。
A lot of companies are building there.
现在下结论还为时过早。
So it's a little early to tell.
在孟菲斯，我觉得反对声主要来自他们用燃气轮机为数据中心供电这件事。
In Memphis, I think a lot of the backlash has been around their use of turbines, gas turbines to power the data center.
至于德克萨斯会不会也这么干，我们还得再看看。
And, you know, we'll have to see if that's something they're going to pursue in Texas as well.
对。
Right.
这确实是个值得关注的故事。
Well, it's certainly a story to watch.
格蕾丝，谢谢你今天来做客。
Grace, I want to thank you for coming on.
这位是格蕾丝·凯，我们《信息》杂志负责报道埃隆·马斯克的记者。
That is Grace Kay, our Elon Musk reporter here at The Information.
甲骨文的数据中心建设仍在继续，而《信息》最新披露了项目过程中出现的数十亿美元意外开支。
Oracle's data center build continues, and The Information has new reporting on multibillion dollar surprise costs that have come up through the process.
我请来《信息》AI基础设施通讯的作者安·戴维斯·沃恩，带大家看看她发现了什么。
I want to bring on Ann Davis Vaughan, author of the Information's AI infrastructure newsletter, to walk us through what she found.
安，欢迎再次做客我们的节目。
Ann, welcome back to the show.
很高兴你能来。
It's great to have you here.
谢谢，阿卡什。
Thanks, Akash.
很高兴能来。
Great to be here.
好，我们来聊聊甲骨文在数据中心建设方面的进展。
Okay, so let's talk about where Oracle is at with its data center build.
你之前报道过他们在新墨西哥州的具体建设情况。
You had some reporting on how that build specifically is going in New Mexico.
你发现了什么？
What did you find?
阿卡什，这个新墨西哥项目对甲骨文来说非常重要。
So Akash, the New Mexico project is very important for Oracle.
它是甲骨文五大‘星门’项目之一。
It's one of its big five stargate projects.
关键问题是，这个已经动工、投入了大量资金的地点，能不能顺利拿到电力供应。
And the key question here is whether it's going to get power to this site, where construction is already underway and a lot of money is already being spent.
为了确保电力供应，他们已经采取了一系列非同寻常的措施。
And it has gone through some pretty extraordinary steps to try to get a clear path to power.
可突然间，之前做的那些似乎还不够。
And all of a sudden, it doesn't look like what it's done is going to be enough.
那它打算怎么解决电力问题呢？
So how does it plan or how is it trying to fix the power issue then?
它原本计划建一个离网的天然气发电厂。
So it was planning to do an off the grid natural gas power plant.
大家担心天然气发电会造成污染，还有温室气体排放的问题。
There were concerns both about the pollution from the natural gas power as well as greenhouse gas emissions.
结果四月份它突然改用布卢姆能源的天然气燃料电池方案。
And it switched abruptly in April to a natural gas fuel cell option with Bloom Energy.
它还花了大力气迁移了一条管道。
It also took some extraordinary efforts to move a pipeline.
但最近，新墨西哥州的环保部门表示，他们还不确定是否能批准空气排放许可。
But in very recent days, the environment department in New Mexico has said, We're not sure we're ready to give an air permit.
他们会推迟相关听证会。
We're going to push out some hearings on that.
新墨西哥州的土地专员第二次否决了这条管线的路线。
And the land commissioner in New Mexico has, for a second time, rejected a pipeline route.
这意味着，要找到能让这个项目顺利启动的解决方案，可能需要更长的时间。
And this means that it could take longer to come up with solutions that could get this project up and running.
那这对开发成本意味着什么？
What does all this mean for the costs of development then?
这种转向Bloom Energy的新方案，再加上还得满足地方政府的要求，我猜成本肯定在上涨。
Mean, this new approach that embarked on with Bloom Energy and then whatever they need to do to satisfy the local governments, I presume this means that costs are going up here.
到底怎么回事？
What's going on?
成本确实在上涨，而且我们发现，所有这些AI大型园区项目都是如此——原本以为的预算，最后总是超得离谱。
Costs are going up, and basically what we're finding with all of these AI mega campuses is that what you thought it was going to cost just tends to go up, it's more than you really bargained for to begin with.
燃料电池本身成本就更高。
Fuel cells are a more expensive solution just on their face.
相比原来的天然气方案，现在多花的钱至少有十亿，甚至可能更多。
It's probably in at least a billion, maybe more of expense compared with the old natural gas design.
更有趣的是，这些电池必须持续运行，否则性能会下降。
And more interestingly, cells need to run continuously or they degrade.
比如，就算某天阳光充足，想切换到便宜的太阳能发电也没那么容易。
It's not as easy to switch to cheap solar power, for example, on a day when it's sunny and that is a better way to get your power.
所以对Bloom来说，不仅延误和能源形式推高了成本，每周、每月等电力上线的过程本身也是一笔不小的开销。
So both the delays and the form of energy are a more expensive thing for Bloom, and so is just waiting every week, every month to get power online to their site.
这一点，投资者们都在密切关注。
And it is something that investors are watching carefully.
你在专栏里提到，这些数据中心公司都在追求电力供应的灵活性。
So in your column, talked about this idea of power flexibility being something that all these data center companies are working for.
我的理解是，使用燃料电池反而限制了这种灵活性——就像我们之前说的，可以根据阳光强弱或实时能源状况，灵活切换不同的AI模型来降低成本，能源方面也是一样。
And so the way I understand it here is by using fuel cells, they basically have less of that flexibility because kind of like how we've talked about this is an analogy, so tell me if I'm getting it correctly, but we talk about routing to different AI models and which model is cheaper, depending on what the level of sunlight is or what the energy situation is on the ground, you can also route to different sources of energy to lower your costs.
但用了燃料电池，就做不到这一点了。
But by using fuel cells, can actually do that.
我说得对吗？
Is that right?
是的
Yeah.
你可以在任何站点加装其他形式的电力，但使用燃料电池的站点有个问题：如果频繁开关机，设备会加速损耗。
You can add other forms of power to any site, but there is a cost with a site with fuel cells that the equipment would degrade if you turn it on and off frequently.
这些离网电力站点必须自己承担大量设备维护的压力。
And these off grid power sites do have to take on a lot of their own equipment burden.
它们没法指望电网来提供额外的电力或维持稳定。
They can't rely on the grid to be there for the little extra power and to provide that stability.
所有这些都得靠自己搞定。
They're having to do all of it themselves.
这些都是成本很高的工程解决方案。
They're pretty expensive engineered solutions.
对
Right.
好的，所以新墨西哥州的情况就是这样。
Okay, so that's what's going on in New Mexico.
我还写到了一些内容，咱们先聊聊甲骨文，然后再谈Meta。
Also wrote about some I mean, we'll talk about Oracle and then we'll get to Meta in a second.
威斯康星州那边情况怎么样？
What's going on in Wisconsin?
威斯康星州也是人工智能发展的重要地区，甲骨文在那里的波特华盛顿还有一个非常重要的项目。
So Wisconsin is another big area of AI development, and Oracle has another very important project there in Port Washington, Wisconsin.
州电网监管机构坚持要求，新建发电设施和输电线路的费用必须由甲骨文及其合作伙伴承担。
The state regulators of the grid have been insistent that the expenses of building more energy generation as well as transmission need to fall on Oracle and its partners.
甲骨文已经在威斯康星州提起诉讼，质疑有关其信用评级所需抵押金的具体要求。
And Oracle has gone to court in Wisconsin challenging some requirements about how much collateral it has to post because of its credit rating.
这再次说明，这类项目的成本比大型云服务商原本预想的要高得多。
It is just another example of more expenses coming on these projects than perhaps the hyperscalers anticipated.
你提到了信用评级。
So you mentioned credit rating.
咱们聊聊这些延误的问题。
Let's talk about these delays.
这现在影响甲骨文的信用评级了吗？
Is it impacting Oracle's credit rating right now?
具体情况是怎样的？
How is that?
标普上个月把甲骨文的信用评级下调到仅高于垃圾级，理由是甲骨文要承担大量前期支出，而这些数据中心带来的收益要好几年后才看得见，毕竟项目还在建设中。
So S and P downgraded Oracle's credit rating to a notch above junk earlier this month and is citing a lot of ongoing upfront costs that Oracle has to pay for, while the reward of these data centers and the revenue that they'll generate is years away as they continue to build.
这确实是一笔巨额投入。
It's a very expensive endeavor.
如果你的财务压力比四大云服务商更大，那你的融资成本自然会更高。
When you are the company that has more financial strain than perhaps the big four hyperscalers, that does mean that your capital costs could be higher.
标普担心的是，甲骨文的前期投入实在太大了。
That is S and P's concern that Oracle has quite a lot of upfront investment.
那投资者现在很担心吗？
So are investors fretting then?
毕竟只比垃圾级高一点，这已经很接近了。
I mean, a notch above junk seems pretty close to junk.
依我有限的了解，信用评级降到这地步，确实不太妙。
In my limited understanding of credit ratings, that doesn't seem so good.
我接触的一些投资者觉得还不至于跌到垃圾级，但大家都盯着呢。
Some of the investors I've spoken with don't expect that it would go down to junk, but we're all watching it carefully.
没错，投资者也在担心资本成本和项目延期的问题。
And yes, I think investors are worried about capital costs as well as delays.
明明建了园区，却连电都通不上，这确实让人很不安。
Fact that have a campus but can't get power to it does trigger a lot of concerns.
但这不意味着他们解决不了。
It doesn't mean that they won't figure it out.
Oracle一直在不断调整，既在扭转公众看法，也直接向新墨西哥州民众解释，还修改了设计方案，想让项目能行得通。
Oracle has been doing all kinds of in the minute pivots, both with public opinion and trying to make their case directly to the people of New Mexico, as well as trying to change the design so that it would work.
对他们来说，把这事重新拉回正轨至关重要。
It's very important for them to get this back on track.
对。
Right.
我说了，我正看着这里的股价走势图呢。
And I mean, we should say, I'm looking at the stock chart here.
今年股价已经跌了35%。
I mean, the stock is down 35% this year.
光是去年，就跌了47%。
Last year alone, it's 47% down.
所以很明显，投资者确实很关注你提到的这些问题。
So, I mean, it seems like investors are certainly paying attention to these challenges that you've been writing about.
我还想再问你一个问题。
I want to ask you one more question.
我刚和格蕾丝聊了聊，SpaceXAI在孟菲斯建数据中心遇到的反对声音。
Just talked to Grace about the pushback that SpaceXAI has been seeing for its data centers in Memphis.
现在他们又在德克萨斯州寻找新的机会。
Now it's pursuing another opportunity in Texas.
那Oracle在新墨西哥州和威斯康星州建设施时，当地遇到的反对情况是什么样的？
What does the pushback look like locally for Oracle in New Mexico and Wisconsin, where it's building its facilities right now?
现在的情况跟以前完全不同了，以前大家只关心这个项目到底会是什么样子。
It's been an extraordinary shift from just the focus on what is this project going to be?
我有没有参与决策？
Did I have enough input into it?
很多人对土地、水源和空气质量可能受到的影响感到非常担忧。
A lot of fear and concern about the impacts on land, on water, on air.
这确实很难。
It's hard.
网上也流传着不少错误信息，但新墨西哥州本来就缺水，当地居民最担心的是怎么合理管理这些珍贵的自然资源。
There's misinformation out there too, but what is happening in New Mexico, it's a water scarce area and people are concerned about how they're going to manage natural resources.
甲骨文自己找到了水源，并且会自行处理用水。
Oracle sought out their own water source and will treat their own water source.
他们直接对公众说：我们一定会负责建设所需的所有基础设施。
They're going directly to the public saying, Look, we're going to be responsible for putting in the infrastructure that we need.
全美各地的科技公司都在这么做。
Tech companies are doing that around the country.
他们在威斯康星州也想如法炮制，但还是被质疑需要抵押多少担保物。
They're trying to make that same case in Wisconsin, but somehow still pushed back on how much collateral they've got to post.
每个地方的情况都不一样，处理起来都很微妙。
So it's a delicate dance in each place and each location is different.
安，我能问你一个问题吗？
Can I ask you one question, Ann?
我们一直盯着甲骨文，但……
I mean, we're so focused on Oracle.
你在通讯里也写过Meta。
You wrote about Meta as well in your newsletter.
甲骨文遇到的这些困难，是它一家的问题，还是整个数据中心建设普遍存在的问题？
Are the challenges that Oracle is facing, is this an Oracle problem or is this a data center build wide problem?
Meta在各地建厂时，遇到的也是类似的情况吗？
Are these similar problems that Meta is facing in its states wherever it's building?
这些问题大多超出了甲骨文本身的范围。
Most of these issues go beyond Oracle.
Meta 有一些项目进展得挺顺利的。
Meta has had some projects that are going pretty well.
但他们也得回应一些问题。
They've also got some questions that they're having to answer.
我们这篇文章里提到，Meta 在德克萨斯州的埃尔帕索和当地公用事业公司合作了一个项目，离新墨西哥州的那个项目很近。他们想出了一种可能成本更低、耗水更少的方式，来临时解决电力供应问题，直到公用事业公司能正式供电。
We wrote in this piece about a project that Meta is doing in partnership with a local utility in El Paso, Texas, very close to the New Mexico project, but in Texas, where they've come up with what may be a less cost intensive and less water intensive way to bridge the power till the utility can get power to them.
但与此同时，Meta 还在应对另一个故事，我觉得你接下来会听到更多关于他们如何处理这件事的细节。
But at the same time, Meta is also juggling a story that has started to I think you'll hear more about on how Are they handle way
我们说到细菌的事了吗？
we getting to the bacteria story?
到底怎么回事？
Is what it is?
快了。
We are.
对。
Yes.
在怀俄明州的夏延，Meta有一个数据中心。
So in Cheyenne, Wyoming, Meta has a data center.
他们当时在清理一些本应用于闭式冷却系统的管道。
They were cleaning out some pipes that are supposed to, be used for closed loop cooling.
当地水务部门声称，有细菌进入了废水系统（不是饮用水系统），并指出源头来自Meta。
The local water utility has alleged that there is a bacteria that got into the water system, the wastewater system, not the drinking water system, and has said that it's traced back to Meta.
Meta已经对被认定为污染源的决定提出上诉，但整个事件仍有不少疑点，这也引发了人们对一项重大创新的质疑——这些AI数据中心本应通过循环用水来大幅减少水资源消耗。
Meta has appealed a decision saying that it is the source, but there's some mysteries around how it all happened, but it has brought up a lot of questions about what was supposed to be a very big innovation, where these AI data centers use less water, recirculate fluid to cool their data centers.
未来可能会有更多关注聚焦在这些闭式冷却系统上，以及它们所在社区的废水系统是否能应对反复循环液体带来的沉积物积累，这值得持续观察。
There could be more scrutiny around these closed loop cooling systems and whether the wastewater systems in the communities that they are in are prepared to handle some of the buildup that comes from reusing these liquids again and again, it's something to watch.
这问题本来是可以解决的，但你知道，关键在于公众的看法。
It could be an addressable problem, but as you know, it's the public perception
没错。
that Yeah.
说的就是公众看法，就像我们刚才跟格蕾丝聊的，她说其中一个Colossus项目在孟菲斯花了整整122天。
Mean, public perception, and as we just talked about with Grace, mean, hundred and twenty two days, I think she mentioned, for one of the Colossus builds in Memphis.
我的意思是，这些项目推进得实在太快了，所以
I mean, these projects are moving at extraordinary pace, and so
是啊，但现在已经慢下来了
Yeah, but they're slowing They're down
慢下来了。
slowing down.
对。
Yeah.
所以媒体报道才这么重要，毕竟成本超支问题不容忽视。
Well, that's why the reporting is so important to consider, surely the cost overruns.
Ann，非常感谢你今天来做客。
Ann, I want to thank you for coming on.
这位是Ann Davis Vaughan，我们《The Information》AI基础设施通讯的作者。
That is Ann Davis Vaughan, author of our AI infrastructure newsletter here at The Information.
接下来的环节，我们邀请到合作伙伴Compute。
Our next segment is with our partner, Compute.
我们这档节目已经聊过不少新云服务商了。
We've covered a lot of Neo Clouds on this show.
它们大多严重依赖英伟达的产品。
Most of them rely heavily on NVIDIA.
但General Compute走的是另一条路。
General Compute is taking a different approach.
我想请这家公司的联合创始人兼CTO杰森·古迪森来聊聊他们的策略。
I want to bring on the co founder and CTO of the company, Jason Goodison, to talk about their strategy.
杰森，欢迎来到节目。
Jason, welcome to the show.
很高兴你能来。
It's great to have you here.
能来我也很高兴。
Great to be here.
谢谢你的邀请。
Thanks for having me.
我想先聊聊今天早上的大新闻。
I want to start with the big news this morning.
Anthropic 向 AMD 采购了两吉瓦的算力。
So Anthropic is buying two gigawatts of compute from AMD.
AMD 正在向 Anthropic 投资五十亿美元。
AMD is investing $5,000,000,000 into Anthropic.
你对这笔交易怎么看？
What did you make of that deal?
是的。
Yeah.
说实话，我真希望这能彻底颠覆整个市场格局。
Honestly, I'd love to say this kind of blew the entire game out of the market.
但事实上，这类交易我们早就开始看到了。
But the truth is we've already started to see deals like this happen.
对吧？
Right?
现在Meta已经和AMD达成了合作。
So we have Meta that's doing a deal with AMD already.
OpenAI也在和AMD合作，现在又来了类似的交易。
We've got OpenAI that's doing a deal with AMD, and now kind of more of the same.
我觉得原因在于，NVIDIA的利润率太高了，如果你选AMD，推理性能差不多，但可能训练上不如NVIDIA，毕竟软件生态有差距，不过推理这块确实能打平。
I think the reason is because and NVIDIA has such healthy margins that if you're doing AMD and you're kind of neck and neck on inference, maybe you're not as good at training because of the software ecosystem, but you're pretty much neck and neck on inference.
你可以算一笔账，这些交易对Anthropic来说更划算，能省不少钱。
Well, you can pencil some of these deals that are just more advantageous for Anthropic because they can save money.
他们的总体拥有成本会更低。
You know, their total cost of ownership would be lower.
所以这事儿挺明显的，为什么不这么做呢？
So it seems obvious, like, why wouldn't they do a deal like this?
这周AMD刚推出了新的机架系统。
And this is the new rack system that AMD has launched this week.
你觉得这个叫Helios的机架系统，能跟NVIDIA抗衡吗？
But what's your sense on how that rack system Helios is going to compete with NVIDIA?
是的。
Yeah.
我的意思是，我们正在探索GPU的替代方案。
I mean, so we're doing alternatives to GPUs.
所以我们可能不会去买它们中的任何一个。
So we're not going to probably buy either of them.
但事实是，这些GPU系统在总体拥有成本上表现非常出色。
But the truth is that these GPU systems are really good at at TCO.
TCO就是总体拥有成本，简单说就是运行AI系统的速度和成本有多低。
And TCO is total cost of ownership, essentially, how fast and how cheaply am I running my AI systems.
所以Vera Rubin和Helios系统在运行AI时速度其实很慢。
And so the Vera Rubin and the Helios system are really good at running AI very slow.
如果你只是想慢速处理大量token，那它们的成本确实更低。
And so doing it at a cheaper price if you wanna run tokens really slow.
我们公司目前关注的是采购那些以速度为优化目标的GPU替代方案。
What we're focusing on as a company is buying alternatives to GPUs that optimize on speed.
所以我们想实现的是每秒处理一千到两千个令牌。
So what we're looking to do is target one to 2,000 tokens per second.
Helios 和 Vera Rubin 都是非常出色的系统，特别适合那些想以低价生成每秒四五十个，甚至一百五十个令牌的用户。
And Helios and Verruban are both fantastic machines, and they're going to do really well at helping people compete on price when they want to generate 40 to 50, maybe 150 tokens a second.
但从商业角度来看，这根本不是我们关注的重点。
But fundamentally, that's just not really what we're focused on as a business.
所以你们是在寻找 GPU 的替代方案。
So you're looking at alternatives to GPUs.
那你们在和哪些公司合作？会把哪些芯片用在你们的设备里？
So who are you working with and which chips are you going to put in your facilities?
是的。
Yes.
我的意思是，我们跟大多数被称为 ASIC 的厂商都聊过。
I mean, we've spoken to most of the they call these ASICs, right?
我们主要接触的都是这些 ASIC 厂商。
We've spoken most to most of these ASIC providers.
我们最看好的是我们的SambaNova芯片。
And the ones that we're the most excited about are our SambaNova chips.
我们已经部署了一套使用SambaNova上一代硬件的集群。
We already have a cluster live with, SambaNova's last generation hardware.
目前我们为Minimax等模型提供服务的速度，比竞争对手快了五倍。
We already serve models like Minimax about five times faster than competitors.
我们现在能做到每秒处理大约500个token。
We're we're doing it about 500 tokens per second.
但我们真正期待的是明年年初即将推出的SN50芯片。
But what we're really excited about is the s n 50 chip that's coming out, at the beginning of next year.
我们有一套大型集群Q1，很快就会上线。
We have we have a large cluster q one that's that's gonna come out.
这些芯片的功耗是每机架20千瓦，所以我们可以直接部署到美国任何现有的数据中心。
And, these these chips are gonna be 20 kilowatt rack draw, so we can basically put them into any existing data center across America.
它们能在大型前沿模型上实现每秒1到2000个token的处理速度。
They're gonna be able to operate at one to 2,000 tokens per second on large frontier models.
我们能在单个机架上部署一个万亿参数的模型。
We're gonna be able to put a 10,000,000,000,000 parameter model on a single rack.
所以说到效率和速度，这些东西完全把其他竞争对手甩在了身后，我们对此非常兴奋。
So really, when it comes to efficiency and speed, these things really just blow the rest of the competition out of the water, and we're really excited about it.
你们为什么选择SambaNova？
Why did you pick SambaNova?
我们之前已经请过Rodrigo来我们节目好几次了。
I mean, we've had Rodrigo on our show a couple times before.
我记得他们以前的芯片主要专注于运行开源模型，而最近几周，开源和开放权重模型显然火了起来。
I know that their chip traditionally has been it's been very focused on running open source models from what I recall, and open source, open weight models have obviously taken off, at least in the conversation, the last couple weeks.
你们选择SN50和SambaNova，是因为这个原因吗？
Was that a reason that you pursued the SN50 and Salmonova specifically?
除了这个，你们选他们还有没有别的原因？毕竟还有其他芯片公司，比如D Matrix、ETS、Cerebrus。
Were there other reasons that you picked them over some of the other chip companies?
咱们聊聊D Matrix，ETS，还有Cerebrus。
I mean, there's let's talk D Matrix, there's ETS, there's Cerebrus.
我老是念不对这个名字。
I can never pronounce it properly.
你们当初为什么
Why did you
选择SambaNova呢？
go with SamaNova?
是的。
Yeah.
我觉得原因在于，这是一次千载难逢的机会，能以极快的速度实现前沿水平的智能。
I think the answer is because we think that there's a weird combination here of once in a lifetime opportunity to do frontier level intelligence extremely fast.
举个例子，从拨号上网到宽带的转变——我这么说可能有点老了，但想想以前等三十秒才加载一个页面，现在谁受得了？
So when you think about, for example, the difference between dial up Internet and broadband, I might be dating myself a little bit here, but, you know, sitting and waiting thirty seconds for a page to load, is it would be considered unacceptable now.
可现在要是等个十到二十分钟，甚至一小时，才等到一个编程助手跑完，大家反而觉得理所当然，可这种事未来肯定不会继续下去。
But, you know, sitting and waiting for ten, twenty, maybe even an hour for a coding agent to run would seem par for the course, and that just doesn't seem like something that will happen in the future.
至于Cerebrus，我觉得确实挺有意思的。
So Cerebrus, I think, is really interesting.
那里的芯片工程师都非常出色。
Machined engineers there are are exceptional.
事实上，你会发现每家ASIC公司都有顶尖的工程团队，因为他们做的事本身就是世界上难度最高的工作之一。
In fact, what you'll find is every ASIC company has an exceptional engineering team because what they're doing is fundamentally just one of the hardest things in the world.
Cerebrus运行速度很快，但我们发现模型规模上还是有个瓶颈。
And so Cerberus can run really fast, but we found that there's a bit of a limitation on the model size.
随着你不断增加晶圆数量，成本简直会飙升到天上去。
And as you scale up to more and more and more of these wafers, the the cost just goes through the roof.
如果你想运行一个超大的模型，那成本高到根本承担不起。
So if you did wanna run a really large model, you'd be serving at just like a cost that you just could not afford to.
而SamaNova的独特之处在于，它既能超快运行大模型，价格还合理又有竞争力。
And so there's a unique combination with SambaNova where you can run large models extremely fast and actually at a reasonable and competitive price.
我觉得，这就是我们一直等待的突破口——快速推理的完美结合点。
And so I think that's the sweet spot that we've been waiting for for this breakout success of fast inference.
我觉得，我们正站在宽带时代的门槛上。
And I think we're on the cusp of our broadband moment.
现在很多人都在说，使用开源权重模型能大幅降低成本。
People are talking about open weight models as a way to lower their costs significantly.
我只是想确认一下，我对这个的理解有没有偏差。
I just want to make sure I gut check my understanding of this.
开源权重模型真的能减少计算资源的消耗吗？
Do open weight models, do they use less compute?
而且，这会不会意味着芯片和云服务的市场规模其实会变小？
And also, does that inherently mean that maybe the market size for the chip business and cloud services is lower
比我们原先想的还要小？
than we thought?
完全不是这么回事。
No, not at all.
这里面有个杰文斯悖论，我稍后再回过头来谈开源权重模型的问题。
There's a Jevons paradox to this, and I'll get back to the open weight model question.
但我们发现，随着每个token的成本不断下降，实际使用量反而越来越多。
But what we found is that the the cheaper the tokens have gotten over time, the more usage has actually happened.
所以现在随着token变得更快、更便宜，大家用得也越来越多了。
And so we're seeing, as tokens are getting faster and cheaper to use, you're just using more of them.
你可以想象一下，用一个编程助手来搭建网站。
And you could consider imagine using a coding agent where you're building a website.
为什么一次只建一个网站呢？
Why build one website at a time?
为什么不同时生成十个版本，然后挑出你最喜欢的那一个？
Why not build 10 variants of it and pick the one you like the most?
这些事情现在还看不到，但我相信未来一定会出现。
These are things that you're just never seeing right now, but I expect to see in the future.
开源模型的优势在于灵活性高，你可以深入进去，搞清楚它的运作机制，然后根据自己的需求做调整。
And so open weight models, they're really flexible because you can kinda get in there and get dirty and understand how they work and just tune them for your own use cases.
所以现在很多人用它们来提升在私有数据集上的评估效果。
And so you're seeing people use them to get even better evals, evaluations, you know, on their their private datasets.
比如说，你是一家做法律AI的公司。
So let's say you're a company that's doing law AI.
你得在这方面做到顶尖。
You wanna be really good at that.
所以你会专门设计一套测试套件，然后用一个开源模型。
So you make a test suite that's optimized on that, and then you use an open weight model.
别直接用Anthropic的现成模型，而是能自己调整，完全贴合你的需求。
Instead of using one of Anthropic's models, you know, off the shelf, you get to tune something specifically for your use case.
还有一点要说，假设你有专有数据。
And the last thing to say on that too is, let's say you have proprietary data.
如果你用Anthropic，就意味着要把数据发给他们。
You know, if you're gonna use that with Anthropic, you're sending the data to Anthropic.
你只能祈祷希望Anthropic别把你的数据悄悄用到他们下一个模型里，不然你的核心资产就全没了。
And, you know, you're kind of praying and hoping that Anthropic doesn't just, you know, kind of bake it into their next model because that's all of your IP gone.
所以现在大家用开源模型，既能保护数据隐私，又能提升成本效益，还能获得更精准、更贴合实际场景的AI性能。
And so these open weight models are something people are using to keep their data private and also get better better unit economics and just better AI in general that's optimized on on their use cases.
没错。
And Right.
这就是我们看到的巨大转变所在。
That's how we're seeing a big shift to it.
我想再回到你当初决定不只用SAMAnova、也不用NVIDIA这个选择上。我们这边的调研数据显示，根据记者们收集的信息，NVIDIA在推理市场的份额其实一直在上升。
So I want to get back to your decision to not just use SAMAnova, but not use NVIDIA, because we've done some reporting here at the information that if you look at NVIDIA's share of the inference market, it's actually been going up according to our reporters' estimates and data sources that they've consulted.
既然客户还是这么需要NVIDIA来做推理，那你为什么依然觉得不选NVIDIA对你来说是明智的呢？
And so in a world where customers still want more NVIDIA for inference, why then do you still believe that it's wise to not go with an NVIDIA for your offer?
这真是个非常好的问题。
I mean, it's a fantastic question.
我觉得答案在于，现在整个算力市场供不应求，而NVIDIA的供应链非常强大，他们早就领先了所有人。
I think the answer is that we're in a compute dry market, and NVIDIA has really robust supply chains, and they've got basically a head start on everyone.
大家需要的是算力，而他们能提供算力。
So people need tokens, and they produce tokens.
所以事情其实很简单。
And so it's pretty simple.
你只要给我更多算力就行。
You know, give me more tokens.
但我认为未来会发生的是，我们正在为未来的需求做准备，也就是我们需要更快的推理速度。
But what what I think is gonna happen in the future is, what we're doing is we're optimizing for where the puck is going, right, which is we want fast tokens.
这个快速推理的市场才刚刚起步。
And this market of fast tokens is just kind of it's nascent.
现在才刚开始兴起。
It's just starting to take off.
你看到了Cerebras的IPO，这说明市场确实关注这类技术。
You've seen the Cerebras IPO, is market indication that people care about this stuff.
显然，OpenAI已经在用Cerebras运行他们最新的模型。
Obviously, OpenAI is running some of their latest models on Cerebras.
这个市场才刚刚开始，我们决心要主导它。
So that market is just getting started, and we really wanna own that market.
这就是我们的方向。
So that's where we're going.
这才是未来趋势所在。
That's where the puck is going.
但AMD和NVIDIA现在正在激烈竞争。
But, AMD and NVIDIA, you know, they're fighting it out here.
你还能看到，NVIDIA通过一些循环融资操作，实际上在某种程度上刺激了需求。
You're also seeing NVIDIA is theoretically, you could say subsidizing demand to some degree by doing some of these circular financing deals.
就连AMD，说到今天发生的事，你也知道吧？
And even AMD, now that we're discussing what happened today, you know?
今天早上，他们向Thropic投资了50亿美元。
This morning, they invested 5,000,000,000 into it, Thropic.
所以你看到的是，这种补贴现象吗？
So you're seeing like, you're seeing subsidization?
是的。
Yeah.
这正是推动市场加速的原因之一。
That's Part of the reason that's accelerating it.
而这些ASIC供应商，没有一家有实力做这种事。
And these ASIC providers, none of them have the balance sheet to do that.
他们为什么要这么做呢？
And why would they?
对吧？
Right?
他们才刚起步而已。
They're still getting started.
建造ASIC芯片所需的营运资金简直高得离谱。
Working capital requirements for building ASIC is unbelievable.
所以他们才刚起步。
And so they're just getting started.
因此，我们公司的做法是与债务融资方合作，来资助这些ASIC芯片的研发。
And so what we're doing as a business is we're partnering with we're partnering with debt financers to help fund these ASIC chips.
我们并不指望ASIC公司自己承担这笔投资，毕竟他们没有足够的资产负债表来支撑。
And we do not expect that the ASIC is gonna underwrite the investment because, you know, they don't have the balance sheet for it.
这正是我们目前所做事情特别困难的地方。
So that's that's particularly what's difficult about doing what we're doing.
所以我们现在还处于非常早期的阶段，但正因如此，它才会朝着未来趋势的方向快速增长。
That's why we're so early in it, but that's why it's going to grow to where we're going where the puck is going essentially.
那你觉得薇拉·鲁宾望远镜会显著改变这个局面吗？根据你看到的情况，它会让人们更想要英伟达吗？
And you don't think that the Vera Rubin will materially change that game or make people want NVIDIA even more based on what you're seeing?
我觉得确实如此，毕竟Grok收购案已经发生了，对吧？
Well, I think so obviously the Grok acquisition happened, right?
全球能做快速推理的硅芯片，其实就那么寥寥几款。
So there's only a handful of pieces of silicon in the world that can do fast inference.
有Cerebras，有Grok，还有SambaNova。
You've got Cerebras, you've got Grok, you've got SambaNova.
Grok的收购案挺有意思，我们现在也开始看到一些消息，比如Semi Analysis有报告提到，薇拉·鲁宾、Grok机架、Kyber这些项目的交付时间都已经推迟了。
The Grok acquisition was an interesting one, and we're starting to see you know, there was some report from semi analysis about how there, there's some delayed timelines on the the Verirubin already, for the the Grok, rack, the the Kyber.
不过公司那边已经否认了这些说法，具体会怎样，还得再看看。
And, you know, that was rejected by the company, so yet to be seen what actually happens there.
但说实话，不谈太技术的东西，Grok芯片其实并不适合当前大型模型的发展方向。
But the truth is, and not to get too technical here, but the Grok, chip is not really the optimal chip for how, big models are getting these days.
展开剩余字幕（还有 265 条）
他们把所有东西都集成在芯片上了。
So they put every single thing on the chip.
他们这么做有个问题，就是用了所谓的注意力和FFN解耦技术。
The the difficulty with what they're doing is they're doing what's what's called attention FFN disaggregation.
我不打算讲得太深，简单说就是，每次Grok芯片生成一个词元，都得来回往返于VeriRubin机架和Grok芯片之间。
I don't wanna get too into the weeds, but the the TLDR on it is that every time they produce a token on a Grok chip, they have to go back and forth to the the VeriRubin rack and back to the Grok chip.
你想想看。
So think about it.
如果是一个80层的神经网络，就得来回跑80次，这简直太离谱了。
If you have an 80 layer neural network, you're gonna go back and forth 80 times, which is just insane.
而SN50芯片呢，我们会直接在这块硅片上一次性完成整个解码过程。
Whereas on the s n 50 chip, what we're gonna do is just kind of run the entire decode on on this silicon all at the same time.
再说一遍，我不打算太技术化，但关键是，这芯片压根不是为这种用途设计的。
Again, not to get too technical, but the point is that the the chip is just not optimized for the use case.
我觉得它要跑大型模型，肯定会特别吃力。
I think it's gonna have a really hard time running large models in particular.
而且就算它真能做到最优，我们短期内也看不到。
And even if it was going to be optimal at it, we're not gonna see it for a while.
所以我们现在拿到的这些SN50机架都是风冷的。
And so these SN50 racks we're getting are air cooled.
每个机架功耗有20千瓦。
They're a 20 kilowatt draw.
至于我们说的Kyber机架，功耗最高能到600千瓦，简直不可思议。
The Kyber racks that we're talking about are, you know, up to 600 kilowatts, which is insane.
它们还使用
They also use
所以你的意思是，鲁宾系列芯片和Grok芯片的功耗需求，正如你所说，会非常大。
So basically, what you're saying is the, I mean, the power that is going to be demanded by the Rubin family of chips, and the Grok chips, from what you're saying, that's going to be significant.
这会成为一个难题，NVIDIA必须面对，这也是你开始转向其他方向的原因之一，没错。
That's going to be a problem that a challenge that NVIDIA will have to deal with, and that's one of the reasons why you've started to diversify away from it, which Exactly.
对。
Yeah.
我的意思是，你看。
Mean, look.
我们这节目里一直聊这个话题，我觉得很多人都在找替代方案，所以……
It's we we've we've been talking about it on the show, and I think I think people are certainly looking for alternatives, and so I
我觉得这
think it's
确实是个明智的策略。
it's certainly a smart strategy.
杰森，谢谢你来参加我们的访谈。
Jason, I want to thank you for coming on and speaking with us.
这位是杰森·古迪森，General Compute的联合创始人兼首席技术官，欢迎来到TI TV。
That is Jason Goodison, the co founder and CTO of General Compute here on TI TV.
亚马逊支持的AI工作室Fable正在制作并发行第一部登陆美国影院的AI生成视觉长片。
Amazon backed AI studio Fable is producing and distributing the first AI generated visual feature film to hit US theaters.
这部电影叫《神不赠礼》，由YouTube创作者伦敦执导。
The movie is called Gods Don't Give Gifts, and it is by YouTube creator, London.
现在加入我们的是Fable的首席执行官爱德华·萨奇和这部电影的导演扎克·伦敦。
Joining me now is Fable CEO Edward Saatchi and director of the film, Zack London.
欢迎两位的到来。
Welcome to the both of you.
很高兴你们能来。
It's great to have you here.
是的。
Yeah.
非常感谢。
Thanks very much.
扎克，跟我们说说这部电影吧。
So, Zack, tell me about the movie.
这里面内容挺多的，值得好好聊聊。
Well, there's a lot to to dig into there.
简单来说，《神不赠礼》是一部科幻选集片。
But the gist of it is, gods don't give gifts is a sci fi anthology.
故事发生在一个意识可以自由转移、记忆可以买卖、整个人生经历都能瞬间下载并闪现的世界。
It's set in this world where consciousness can be swapped between bodies, memories can be bought and sold, whole lifetimes can kind of be downloaded and flashed in a in a moment.
我们已经解决了死亡，还有那些曾经拖累人类的其他问题。
Basically, we've we've solved death and all the other, you know, ailments that hold us down.
但影片讲述了四个不同的故事，既有这项技术的受害者，也有受益者。
But it follows four different stories, kind of both like the the victims and the beneficiaries of this technology.
从被强行赋予意识的生物，到追逐幻影记忆的机器人，再到被塞进新身体的蜘蛛，还有活了上万年、对一切早已麻木的贵族，全都涵盖其中。
So that's everything from, you know, the creature that has had sentience thrust upon it, to an android chasing, you know, phantom memories to a spider that's been kind of thrust into another body to world weary aristocrat who's lived 10,000 lives and is completely jaded by it.
而且这一切都是由人工智能生成的。
And it's And and it's all it's all AI.
人工智能到底是什么？
What is AI?
什么又不是人工智能？
What isn't AI?
能给我们详细讲讲吗？
Walk us through that.
是的。
Yeah.
不是AI的部分是剧本和写作。
So what is not AI is the the writing and the scripts.
我和一位合作者一起写了整个剧本。
Myself and a and a co writer wrote the whole thing.
所有录音中的声音，我们都使用了真实的配音演员。
The voices, we use actual lives voice actors for everything that we record in the studio.
配乐方面，我们请了三位不同的音乐人来创作。
The score itself, we used three different musicians to score the film.
所有的音效和环境声，则是由一组拟音师团队完成的。
And then all the audio and SFX is a team of Foley artists.
所以明确一下，画面是AI生成的，但每一帧都经过了极其精细的制作。
So to be clear, the visuals are AI, albeit painstakingly constructed.
而所有音频、剧本、导演、剪辑和后期制作，全都是人类完成的。
And then, basically, all the audio and the actual script and directing and editing and post production is human.
我想问问你，具体是怎么做出这些东西的？
And, I mean, I want to ask you about how you actually make this.
我们这个节目经常聊模型，你的YouTube频道里也有一些你过去用AI做的素材。
I mean, we talk about models a lot on this show, and we have some footage of of past stuff that you've made using AI on your YouTube channel.
那你到底在用哪些模型？
But what models are you actually using?
用的是哪些软件？
What applications are there?
是的。
Yeah.
如果你要梳理整个技术栈，就会发现 everything 都在变，根本没法锁定一个固定的流程。
So, I mean, if you if you covered the the tech stack, you know, everything's in flux, like, all the time, which makes it really hard to pin down any concrete workflow.
整个制作过程中，技术变化特别大。
Over the course of production, it's changed dramatically.
有时候一两个月，感觉就像经历了好几年的技术迭代，新功能和工具一个接一个地冒出来。
You know, you have months where it feels like decades worth of, you know, features and tools are just released, like, relentlessly.
所以很难确定一个确切的工具组合。
So it's hard to pin down an exact tool set.
但核心来说，我们用的还是那些基础的图像生成工具。
But I would say at its core, you know, we have the kind of bread and butter image generation tools.
我觉得Midjourney在所有这些工作中都至关重要。
Midjourney, I would say, is pretty critical to all of it.
至于动画部分，主要靠Seed Dance。
And then for animation, very much cling and seed dance.
另外还有一整套其他工具，包括AI和非AI的，用来把所有东西拼接起来，再不断调整，直到我们觉得达到了那种独特鲜明的视觉风格——真正能呈现我们想表达的世界，用我们想要的视角来讲故事。
And then, you know, a whole suite of other tools, both AI and non AI, to kind of stitch it all together and, kind of modify it to the point where we feel like we've achieved what I hope is this, like, very distinct and differentiated aesthetic that really kind of captures the world and tells the story through the lens that we want to tell it with.
所以现在主要就是这两个，Midjourney和Seed Dance，我听说过Midjourney，但另外那个还真没听过。
So so mostly right now, it's it's it's those two mid journey, seed dance, I mean, these are tools that I I that I mean, mid journey I've heard of, but I've not heard of the other two.
我想问一下，这些工具背后用的是哪些模型？
I guess those which models are are those using?
Seed Dance现在绝对是动画领域的市场领导者。
Seed Seeddance is by far the market leader right now in animation.
那是字节跳动的动画工具。
That's that's ByteDance's animation tool.
好的。
K.
我们也会用Cling。
And then we also use Cling as well.
Cling。
Cling.
明白了。
Okay.
对。
Yeah.
了解了。
Got it.
爱德华，你能说说这部电影的发行计划吗？
So, Edward, tell me a little bit about the plan for distribution for this movie.
是的
Yeah.
从一开始，我们这些对AI电影充满期待的人，就一直在等，等有一天AI拍的电影能真正达到院线放映的水准——毕竟，随便拼凑九十分钟的内容硬推上去，谁都会做。
So, you know, right from the very beginning, for those of us who've been excited about AI cinema and AI movies, we've been waiting for the moment that a film would be good enough to go out in the cinemas because you can string together ninety minutes of anything and and, you know, try to try to push it out there.
这部片子才是真正的大师之作。
This is actually a masterpiece.
它很动人。
It's moving.
它很感人。
It's touching.
它还特别幽默。
It's it's witty.
我觉得这部电影有资格扛起重任，让普通观众真正认识并接受AI电影。
And it's the kind of film that I think, you know, deserves to carry on its shoulders the the kind of goal of introducing, the ticket paying public to, AI movies.
所以我们定在10月30号，星期五上映。
And so we're gonna be on on Friday, October 30.
你可以去附近的电影院，买点爆米花，再来罐健怡可乐，舒舒服服地看场电影。
You're gonna be able to go, to your local multiplex, get some popcorn, get some Diet Coke, and, and watch the movie.
有没有大型电影公司参与这次发行？
Is is there I mean, is there a major movie studio involved in this release?
他们主动联系过你们吗？
Have you been approached by them?
是的。
Yeah.
我们正在和几家制片公司接洽，但同时也用到了Iron Lung这个模式。
We are talking to different studios, but we're also using the Iron Lung model.
Iron Lung是MarketPlayer做的。
So Iron Lung is by MarketPlayer.
他是个很有名的YouTuber。
He's a big YouTuber.
今年一月，他用一种特别的方式，把他的电影《Iron Lung》搬上了院线，靠的就是他在YouTube上那群热情的粉丝。
And in January, he distributed his film Iron Lung in, in theaters, using a strategy that kind of harnessed the the attention and the excitement of his fan base on YouTube.
而扎克在他的各个频道上，总播放量已经达到了六亿次。
And Zack, across his channels that has had 600,000,000 views within his world.
我觉得《回廊》和《迷宫 obsession》这两部片子，加起来预算才一千一百万美元，但作为YouTube创作者的作品，总收入却达到了九亿美元。
I think with Obsession of Backrooms, you're kinda seeing, two films that had a cumulative budget of 11,000,000, YouTube filmmakers, with a cumulative revenue of 900,000,000.
所以现在，其他发行方、制片公司和院线都开始行动了——之前他们对铁肺电影在一月的做法还有点犹豫，现在却都抢着想知道，怎么把扎克这种创作者和他粉丝的热度，成功带到电影院里。
So the distributors, the the other distributors, the the studios, and the exhibitors are now, whereas they were slightly resistant to what Iron Lung was doing in January, now there's a frenzy of, you know, how do we actually see if we can harness YouTube excitement around an IP and a filmmaker like Zack into cinemas.
看起来大家现在都在想，
And that it seems as though people
埃德，你们当初是决定要启动这个项目吗？
Ed, did you did you guys decide to embark on this project?
这个决定是在《回廊》和《迷宫 obsession》爆红之前，还是之后？
Was it mean, was this before or after the the smash success of Backrooms and Obsession?
说来有趣，我们俩当时是一起去看的《铁肺》。
Well, funnily enough, we both we both together went to the cinema to see Iron Lung.
那大概是九个月前的事了。
This was kind of nine months ago.
所以，也许那时候就埋下了种子。
So the seed the seed was was maybe planted there.
我觉得你看到的是两个趋势。
I think I think you've got two trends.
对吧？
Right?
一方面，你们两位YouTube创作者已经超越了《星球大战：曼达洛人》，显示出新一代正在取代旧一代；另一方面，还有AI。
Like, one is you two YouTube filmmakers crushing Star Wars Mandalorian and, you know, showing that maybe there's a a kind of changing of the god and changing of the generations, and then you've got AI.
我们确实受到《Backrooms》和《Obsession》的鼓舞，但更重要的是，这部片子得让世人明白，为什么大家对AI这么兴奋。
So I think we we were partly, you know, encouraged by obsession backrooms, but also, you know, that's that this does need to be a film that that shows the world why we've all been so excited about AI.
这是一部独立电影。
You know, this is an indie.
它更像是一个实验。
It's it's a kind of, you know, it's it's a it's an experiment.
这将是第一个这样的作品。
You know, this would be the first.
这是人们第一次在大银幕上看到一部AI制作电影的预告片。
This would be the first time people will see a trailer for an AI movie.
没错。
Right.
在其他电影之前上映，而且还是第一部拥有烂番茄评分和Metacritic评分的AI电影。
In theaters before other movies, first time for a rotten tomatoes score for an AI film, Metacritic score.
但我相信扎克和我都非常希望这种艺术形式能被认真对待，设定很高的标准，我觉得这正体现了AI真正能带来的突破。
But I think Zack and I both are just really passionate about this art form being taken seriously and and having a very high bar, which I think gods don't give gifts suddenly rises to of actually
制作这部电影花了你们多少钱？
making How great work of how much did it is it costing you to make this movie?
我的意思是，这些模型价格可不低。
I mean, the models are they're they're pretty expensive.
是的。
Yeah.
模型的成本要看用的是哪一种。
I mean, the the models, it it it depends on what model you're using.
但确实不是免费的。
But, yeah, it's certainly not free.
还有人工成本，当然少不了。
And then, of course, there's personnel cost.
现在网上有种说法，说我们这部预算五亿美元的好莱坞大片，周末花五十美元就做出来了。
I mean, I think there's a there's a narrative out there that unfortunately is like, you know, we made this $500,000,000 Hollywood budget, you know, budget film for for $50 in a weekend.
这明显是在煽动情绪。
And it's kind of rage baits.
但说实话，要做出我们这种水准的作品，或者说是目标水准的作品，没那么简单。
But I think the reality is to kind of pull something off at the the caliber that we're they were pulling it off at or, like, we're aiming for.
这得靠一个团队才能完成。
I mean, it takes a a team to do it.
我有一整支团队，包括视觉设计师、动画师、后期制作组——剪辑、混音、调色、配音演员，还有负责配乐的音乐人。
You know, I have a full team of of visual designers, animators, a post production crew, you know, that includes editing, sound mixing, color grading, voice actors, obviously, musicians to score it.
所以，这不只是靠几个模型就能搞定的事。
So it's, you know, it's not it's not just a matter of credits.
还有人参与其中，真正动手做这些的是人。
There's there's the human involvement there as well, the actual people to to to build this.
虽然现在生成内容确实很快，但我们已经为此忙了七八个月了。
And while it's true that you can generate things really quickly, you know, we've been working on this for seven or eight months now.
这可是一个不小的制作工程。
It's it's quite a big production.
所以我真不想轻描淡写地忽略背后付出的努力。
So I don't really wanna kind of downplay the effort it takes.
我觉得很有必要澄清一个误解：以为只要把内容丢进一个盒子，啪一下就能出一部电影，直接投到大银幕上。
And I think it's important to look kind of, like, dispel that misnomer that, you know, this is some instantaneous type of movie into a box and boom, we have a feature film that we can just, you know, slap on the big screen.
换句话说，我们在这档节目里经常听到模型和算力成本有多高，但对你来说，最大的开销还是员工工资，目前来看，人力成本远高于使用模型本身的费用。
So in other words, you know, when we hear so much on on this show, at least, of of the the models and the compute costing a ton, I mean, for you, the major expense has still been payroll for all your staff, and that's been the majority of the expense compared to the cost of actually using the models right now.
是的。
Yes.
明白了。
Okay.
那你是怎么看待产权、版权，还有底层训练数据这些问题的呢？
And how do you sort of think about, you know, ownership structure, copyright, you know, the underlying training data?
这现在正是好莱坞最头疼的大问题。
This is the this is the big question in Hollywood right now.
作为创作者，你自己是怎么看这个问题的？
How do you think about that as as a creator yourself?
是的。
Yeah.
说实话，我可不敢说自己是AI领域知识产权方面的权威专家。
Look, I mean, I'm not gonna pretend to be, you know, the world's foremost scholar on on IP as it as it pertains to AI.
但我觉得，从以往的判例来看，只要AI生成的内容没有经过充分修改，基本上是不能受版权保护的。
But I think of the precedents that have been established, it's basically that, you know, any kind of raw output from AI that's not sufficiently modified is not really copyrightable, right?
如果你没体现出任何创作意图、所有权，或者没做实质性改动，那就不算你的作品。
If you haven't really kind of demonstrated any form of authorship or ownership or modification of it, right?
而如果能证明你对内容做了足够大的调整，让它真正成了你自己的创作，那情况就不一样了。
Whereas where it's landed in favor of those using AI is if you can kind of demonstrate that you've sufficiently modified it to the extent that you've made it your own.
如果我只是直接用谷歌、字节跳动或者MidJourney这类工具生成一个内容，我就没法说这东西是我原创的。
So if I were to just generate something, you know, let's say raw out of a Google product or a ByteDance product or mid journey, I couldn't make the claim that this is mine.
但当我们考虑到，每个角色其实都是由成百上千张图在不同工具里拼贴、调整、再结合人工配音、脚本创作、动画制作、剪辑和合成出来的，
But when we factor in, you know, that every character is is the consequence of, you know, hundreds of of images across different tools, photo bashed together, manipulated across a variety of tools, voice acted based on scripts that are human written, cast into other animation tools that are done, edited, composited together.
我觉得我们早就远远超出了判断人类创作或修改的底线，对我来说，这明显属于我们真正创造并有权主张版权的内容。
I think we've gone kind of so far past the benchmark of what demonstrates, like, human authorship or modification that to me, it seems unequivocally that this lies well within what is kind of copyrightable content that we've that we've created and kind of claim ownership over.
没错。
Right.
埃德，我想问你个问题，关于奈飞和本·阿弗莱克公司做的那个合作，这跟你在Fable做Showrunner的目标其实挺相关的。
Ed, I wanna ask you a question about the deal that Netflix has done with Ben Affleck's company, and it sort of relates to what you're trying to build at Fable with Showrunner as well.
你觉得奈飞最终想通过这个达成什么目的？
I mean, what do you think Netflix ultimately wants to do with that?
你一直聚焦在互动媒体的未来，让观众能直接告诉平台：我想看这样的内容。
And You've pinned yourself to this future of interactive media where audiences can tell platforms, Hey, this is what I want to see.
我想成为故事里的一部分。
I want to be inserted in this type of storyline.
我问你一个问题，作为观众，我当然可以说我想看动作片、爱情喜剧之类的，但其实我也不太确定自己到底想看什么。
The question I have for you is, I mean, as a viewer, mean, sure, I could say that I want an action movie, I want a rom com, whatever, but I don't actually know what I want, I think.
这让我想到一个问题，确实如此。
So it sort of leads me to the question, yeah, sure.
然后我就开始想信息茧房的事了，如果人们只看那些他们已经熟悉的内容，会怎么样？
Then I get into the about echo chambers, and if people are just watching more of stuff that they already know to have existed.
你是怎么看待这些围绕你正在打造的产品的核心问题的？
How do you think about all those meaty questions at the center of what you're building?
是的。
Yeah.
我觉得Netflix和本·阿弗莱克公司那笔交易是个重要转折点，更多是关于视觉效果的。
So, you know, I think that deal with Ben Affleck's company was a big turning point, more about visual effects.
我们该怎么调整视觉效果，或者说是后期制作阶段的改动？
So, you know, how do we make visual effects changes or, you know, post post post production.
你到底该怎么修改一个东西呢？
How do you tweak something?
你怎么去修改它呢？
How do you change something?
我觉得这很有价值。
I think that's that's valuable.
但我们一直试图更激进地思考这项新技术带来的影响，因为它本质上是可以混搭重组的。
But, we've we've tried to be a little bit more radical about what the implications of this new technology are and that it's, it's innately remixable.
我觉得，不管人们怎么争论它什么时候会普及、消费者行为会怎样，如果你听不进媒介本身在告诉你什么，那你肯定会走偏。
And I think that outside of any debate about, you know, when's it gonna happen, consumer behavior and all that, if you don't listen to what a medium is telling you about itself, you're gonna go wrong.
早期的电影人拍的都是被拍下来的戏剧，根本没去体会电影这门媒介本身的特性。
So people at the early days of cinema, they made filmed plays, film theater, and they weren't listening to what the medium was doing.
而那些激进派却说，不对，这才是蒙太奇。
And the radicals who said, no, this is montage.
这完全是另一回事。
This is completely different.
你能压缩时间。
You can compress time.
通过剪辑交错，你可以在观众心中制造潜移默化的效果。
You can create subliminal effects in the audience by intercutting.
你不能只是把摄像机往那儿一放就拍东西。
You don't just put a camera down and film things.
真正赢的是那些懂这个道理的人。
Those are the ones who won.
我们的观点也是这样：当你真正理解这个媒介本身，它本质上就是可重组的。
So that's our theory as well, that when you when you truly look at the medium, it is remixable.
它是互动的，你该顺着这个方向，听从自己的直觉。
It is interactive, and, you should follow that and follow your instincts.
我们设想的Netflix版AI，是人人都能创作自己的剧集，创作衍生作品。
So we see a Netflix of AI where people create their own shows, where people, create derivative work.
实际上，随着这个发布，大家马上就能参与互动，我们还会围绕《八卦小鬼》这个IP举办一场两万美元的竞赛。
And, actually, alongside this announcement, people are gonna be able to play, and and, we're gonna do a $20,000 contest around the Gossip Goblin IP.
没错。
Right.
或者任何由扎克或《八卦小鬼》拥有的衍生作品。
Or any derivative work owned by Zack, owned by Gossip Goblin.
这只是一个实验，想看看能不能为电影的初期推广带来帮助——多点营销、多点素材、多点内容，也能提升粉丝参与度，同时看看会不会在宇宙观里催生出一些有意思的作品。
Just an experiment to kinda see if that helps for for a start, helps the, like, launch of the film, more marketing, more assets, more content, more fan engagement, but also if it leads to some interesting work within the universe.
扎克·凯恩·帕森斯说过，他并不太喜欢人工智能和好莱坞。
Zack, Kane Parsons has said that he's not really a fan of AI and Hollywood.
既然他一直是YouTube创作者进军好莱坞的先锋，你对此有什么看法？
How does that make you feel given that he's very much been a trailblazer for for content for for YouTubers entering Hollywood?
这挺好的。
That's cool.
他当然有权利持有自己的观点。
You know, he's he's entitled to to his opinion.
你见过他吗？
Have you have you met him?
你和他聊过吗？
Have you talked to
我从来没跟他聊过。
I've never I've never spoken to him.
明白了。
Okay.
我不能说他对此有什么了解。
I I can't say that he's been acquainted.
我的意思是，这让你感觉怎么样？
But I mean, how does that how does that make you feel?
至少在业内，他一直是个开拓者。
Mean, he's he's been a trailblazer so far, at least in the industry.
当然。
Sure.
其实不少开拓者都已经接受了AI。
Then plenty of trailblazer blazers have have embraced AI.
你看。
Look.
我觉得这挺讽刺的。
I I think it's funny.
我的意思是，无论哪一方，都有不少挑衅行为，这很明显。
I mean, from from both sides of the aisle, there's a lot of, like, provocations, obviously.
这话题可真是火药味十足，至少这么说不为过。
You know, it's it's a fiery issue to say the least.
你看，尼尔·布洛姆坎普、达伦·阿罗诺夫斯基，甚至乔治·卢卡斯，都出来说了：嘿，这趋势来了。
And so, you know, you see Neil Blomkamp or, you know, Darren Aronofsky or even George Lucas going out and saying, hey, look, it's coming.
不管怎样，我们迟早都会接受它。
We're gonna embrace it in one way or another.
我完全理解人们对AI的抵触。
I totally get the pushback against AI.
我觉得部分原因只是源于误解。
I think some of it is just comes from misinformation.
但也有不少担忧，确实是很有道理的。
And some of it, I think, is very well founded.
但直接说它完全不好，或者完全好，这其实并不妥当。
But I think unequivocally saying that it's bad or that it is unequivocally good.
我觉得这把一个本来很复杂的问题，硬生生简化成了非黑即白的单一结论。
I think it just basically takes a nuanced issue, and just compresses it down until, like, it is just, you know, a black and white monolithic thing.
人工智能很糟糕。
AI is bad.
人工智能很棒。
AI is great.
人工智能会毁掉好莱坞。
AI is gonna kill Hollywood.
我觉得根本没必要去讨论这些极端说法，很多人就是故意说些极端的话来博眼球。
I think it just it's not a particularly productive conversation to even get into the like, those extremes, and people can make polarizing statements for the sake of it.
但说到底，凯特·帕森斯怎么想，我其实根本不在乎。
But I I mean, you know, at the end of the day, I I don't really care what what what Kate Parsons has to say about this.
对。
Right.
而且这个失误是
And is is the miss
是关于失业方面的错误信息吗？
is the misinformation, is it around, like, job loss?
你认为现在流传的错误信息到底是什么？
Like, what what what is what is the misinformation as you see it that that that's out there right now?
我觉得错误信息多得数不清。
I mean, I think there's a plethora of misinformation.
有些是故意散布的。
Some of it deliberate.
有些则是因为网络讨论本来就是这样，你懂的。
Some of it just because, you know, it's like it is a that's what happens with, you know, Internet discourse.
它总是容易走向最极端、最对立的观点。
It just falls into it gravitates towards, you know, the most polarizing extremes.
比如说失业问题，我认为大家关注的只是非常狭窄的一种应用场景——比如，我的独立电影团队，如果没有这项技术，我们根本不可能以这样的规模来讲故事。
I mean, if we're talking about, for instance, job loss, I I think it's basically focusing on a very, very particular niche application of it, which is, hey, our indie filmmakers, basically, my whole crew wouldn't be working in this space if it wasn't for basically this ingredient technology that's, you know, enabled myself and others to kind of tell stories at scale that would otherwise not be possible.
但如果我们谈人工智能导致的失业问题，这远远不只是生成式AI带来的影响。
But if we talk about job loss from AI, that is an issue that goes way beyond generative AI.
现在却好像在说创作者才是罪魁祸首，而不是去正视AI背后那些更宏大的地缘政治影响——比如Anthropic、ChatGPT、Kimi这些大模型，以及它们如何重塑我们所知的世界。
And now it's basically saying like, hey, creators are at fault for this, as opposed to the, you know, basically, like, geopolitical massive kind of implications of AI as in, you know, the the the broader space of Anthropic and Chatuchakutti and Kim three and the, basically, the world as we know it influenced by LLMs and, you know, basically super intelligence.
但就算把这些放一边，我们还得面对用水量、土地占用，还有这些技术到底会带来什么后果这些问题。
But I think even putting that aside, you get to issues like water usage or land usage or the implications of of what are the consequences of this.
当然，AI也不是什么完美的工具，这我们都知道。
And then it's like, yeah, we're it's not you know, AI is not, this perfect tool.
但你要是看看数据图表，比如高尔夫球场的用水量，简直吓人。
But also, if you look at, like, a bar chart of, like, well, fuck.
高尔夫球场消耗的水量，比AI训练多出好几个数量级。
Golf courses consume, like, many orders of magnitude more water.
再比如所有数据中心的总耗能，到底有多少是真正用在生成式AI上的？
Or of all data centers, what amount is actually being used for the purposes of generative AI
没错。
Right.
相比之下，平台却在不停上传那些无聊透顶的‘跟我一起准备’视频，每二十分钟就二十四小时不间断地堆满内容。
Versus hosting the fucking twenty four hours of content uploaded every, you know, twenty minutes to a platform of just the most inane, you know, get ready with me video.
所以我觉得这有点挑着说，没错，AI确实有很多问题。
So I think it's kind of cherry picking it, which is like, yes, there are many flaws in AI.
从任何角度看都是这样。
From any angle, look at it.
任何新技术都会带来后果，这再正常不过。
There's consequences as there are for any new technology.
但很明显，生成式AI只是特别显眼地成了靶子，让人把对这种神秘甚至有点阴森的新技术的所有恐惧、焦虑和紧张都投射到它身上。
But it seems that clearly this is Gen AI is just kind of poking out in a very obvious place to kind of project all of our fears and anxieties and tensions about what is this kind of strange and kind of sometimes sinister new technology.
对。
Right.
我理解你的意思是，既然工具已经存在了，创作者们只是在努力适应这个新技术带来的新世界，这确实很有道理。
And what I hear you saying is it's, you know, if the tool exists, then this is sort of creators trying to figure out how to live in this new world where this technology exists, which certainly is a good point.
我想问你一个问题。
Want to ask you one thing.
你再说一遍？
Say that again?
快开始吧。
Get going.
不用了。
No.
我只是想说，我想在这儿讲点题外话。
Was just ask I just wanna I wanna take a sermon here.
喂。
Hey.
我这辈子就看过两部恐怖片，扎克。
I've seen, like, two horror movies in my life, Zack.
《月光光心慌慌》第一部和第二部。
Halloween one, Halloween two.
那时候我上十年级。
Was in tenth grade.
它们把我吓到了。
They scared me.
从那以后，我再也没看过任何恐怖片。
I I never I have never watched a horror movie since then.
你觉得为什么现在观众突然对这类片子感兴趣了？
Why are they having a moment right now with audiences, do you think?
尤其是在YouTube创作者当中。
Certainly, among the YouTube creators.
这是个好问题。
That's a great question.
我觉得《诸神不给恩赐》严格来说不算恐怖片。
I don't think that gods Don't Give Gibs is a horror movie per se.
虽然它选在万圣节上映，所以你能理解为什么会有这种印象。
Though it is coming out on Halloween, so, you know, I could see how that might suggest it.
也许有人会觉得它像一部恐怖电影。
It could perhaps be perceived as a horror film.
它也可以被看作是一个警示故事。
It could be perceived as a cautionary tale.
也可能只是被当成酷炫的科幻作品来看。
It could just be perceived as just cool, speculative sci fi.
为什么恐怖片现在这么火？
Why are horror movies having a moment?
我不清楚。
I don't know.
这话题太大了，你大概没时间深聊。
That's a bigger conversation that you probably don't have time for.
我觉得恐怖片这个类型特别适合低成本制作。
I do think that the horror genre is like, a, lends itself to smaller budgets.
所以低成本下反而更容易尝试各种新点子。
And so therefore, it's kind of more open to experimentation on a on a smaller budget.
我觉得这正是AI在科幻领域让我觉得特别激动的地方——要达到我们目前这种规模和精细度，传统制作需要的预算高得离谱。
And I think actually, that's kind of what to me is so exciting about AI in the context of sci fi, which is, you know, you would need a prohibitively large budget to pull off what we're doing at the scale and scope and polish that we're executing it on.
根本不存在独立制作的科幻片这个类型。
There is no indie sci fi genre.
就算有，恐怕你也真不想看。
And if there is, unfortunately, you probably don't wanna watch it.
就像恐怖片现在正火一样，科幻片也终于有机会打开更广阔的世界，能讲一些更大胆、更离奇的故事了，尤其是现在，我们可以讲一些真正深奥、古怪又烧脑的故事。
And so I think that, like, well, the same way that, obviously, horror is having a moments, building off of that, the fact that sci fi can now kind of be opened up to the broader worlds, that kind of more ambitious and out there stories can be told, especially now where we can kind of tell a story that I think is actually like, quite heady and weird and cerebral.
但同时，画面质感又像那种——我真不想用这个词——但确实有大片的精致感。
But with the kind of visual polish of a a I hate to be, like, use this term, but, like, a blockbuster film aesthetic.
这其实是个‘糖衣炮弹’。
But, you know, it's a Trojan horse.
在这层光鲜外表下，藏着的是一些观众平时在这样的电影里根本想不到的、更酷的深层信息。
Within that, there's a, I think, a much cooler messaging than audiences might be used to from a a film that looks like this.
没错。
Right.
非常感谢你们两位来做客。
Well, I wanna thank you both for coming on.
这确实是个很有趣的话题。
It's certainly a fascinating topic.
这里是爱德华·萨奇和扎克·伦敦，欢迎收看TITV。
That is Edward Saatchi and Zack London here on TITV.
今天的节目就到这里了。
That does it for today's show.
提醒一下，我们每周一到周五早上10点（太平洋时间），下午1点（东部时间）直播。
A reminder, we are on this stream Monday through Friday at 10AM Pacific, 1PM eastern.
如果错过直播，节目内容也可以在theinformation.com、我们的YouTube频道，或者你常用的播客平台收听。
If you can't make it then, episodes are available on theinformation.com, on our YouTube channel, or wherever you get your podcast.
别忘了在X、Instagram、TikTok和LinkedIn上关注我们。
Make sure to follow us on social media on X, on Instagram, on TikTok, and on LinkedIn.
我已经迫不及待想看明天的节目了。
I am already excited for our next show tomorrow.
祝你剩下的星期三过得愉快。
Have a great rest of your Wednesday.
那先这样，再见了。
Bye bye for now.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。