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。
欢迎大家收看TITV。
Welcome everyone to the information's TITV.
我叫阿卡什·帕什里卡。
My name is Akash Pasricha.
今天是7月13号，星期一。
It is Monday, July 13.
在开始之前，告诉大家，这周是我们TITV开播一周年。
Before we get started, this is TITV's one year anniversary week.
这一周我们都会在湾区拍摄，特别节目将在旧金山和整个硅谷录制。
We'll be coming to you from the Bay Area all week filming special episodes in San Francisco and across Silicon Valley.
明天是我们播出的第250期节目。
Tomorrow is our two hundred and fiftieth episode.
这一期绝对不能错过。
You do not wanna miss it.
节目内容安排得满满当当。
We have a jam packed lineup.
今天节目中，苹果公司对OpenAI提起诉讼，这一举动让两家公司日益激烈的竞争突然升级。
On today's show, Apple is suing OpenAI in a stunning development for the growing rivalry between these two companies.
我们会请我们的苹果记者来深入解读这件事的最新进展。
We'll bring on our Apple reporter to unpack the latest on that.
接着，我们会和我们的加密货币记者聊聊华尔街迄今为止对区块链最大的押注。
We're then talking about Wall Street's biggest bet on blockchain yet with our crypto reporter.
《AI议程》通讯的作者斯蒂芬妮·帕拉佐洛刚从韩国举行的ICML人工智能研究大会回来。
Stephanie Palazzolo, author of our AI Agenda newsletter, is back from the big ICML AI Research Conference in South Korea.
我们会和她聊聊她在会上的主要收获，最后再与劳德研究所对话，谈谈他们关注的人工智能研究大趋势，以及对最新大模型发布有何看法。
We'll chat with her about her key takeaways from the event, and we'll wrap things up with a conversation with the Laude Institute about the big themes that they're watching in AI research and their reaction to the latest big model releases.
节目内容很精彩，咱们马上开始。
It's gonna be a fun show, so let's get right on into it.
苹果公司起诉OpenAI，指控这家人工智能公司窃取商业机密。
Apple is suing OpenAI, accusing the AI company of stealing company secrets.
这起诉讼已于周五在加利福尼亚北区联邦法院提交。
The lawsuit was filed Friday in the Northern District Of California.
我请我们的苹果记者亚伦·蒂利来为大家梳理一下目前已知的信息，以及这件事未来可能的发展方向。
I want to bring on our Apple reporter Aaron Tilley to break down what we know and where this is likely to go from here.
亚伦，欢迎再次做客我们的节目。
Aaron, welcome back to the show.
很高兴你能来。
It's great to have you here.
谢谢。
Thank you.
好的。
Okay.
这是一场大官司。
Big lawsuit.
给我们说说里面都写了什么。
Tell us what's in it.
我们先从这儿说起。
Let's start there.
是的。
Yeah.
苹果起诉了OpenAI，指控其设备团队系统性地窃取了苹果在产品开发、供应链乃至制造方法方面的专有信息。
So Apple filed a law suit against OpenAI alleging that there has been a systematic effort on behalf of OpenAI on the devices team to steal proprietary Apple knowledge related to product development all the way to supply chain and manufacturing methods.
苹果提出的这些指控非常严重。
So this has been, the allegations Apple is making is very serious.
OpenAI的做法相当隐蔽，手段不光彩。
The companies do, the OpenAI is doing things in a very sneaky way.
苹果称，OpenAI的员工故意隐瞒行动，试图绕过苹果的安全系统。
Apple alleges that OpenAI folks are purposely hiding their efforts and trying to bypass Apple security.
这些指控确实非同小可。
So pretty serious allegations.
我想详细说说这里面一些关键的指控，但这场诉讼的核心是两个人：唐坦和刘畅。
I want to go through some of the big allegations in here, but there are two people that are at the center of this lawsuit, Tang Tan and Chang Liu.
他们是谁？为什么他们在诉讼中如此关键？
Tell us about who they are and why they are so central to this lawsuit.
是的。
Yeah.
唐目前是OpenAI的首席硬件官。
So Tang is currently the chief hardware officer at OpenAI.
他曾是苹果的硬件工程师，任职超过二十年。
He was a long time, over two decades Apple hardware engineer.
在离职前，他主导了苹果所有iPhone产品的开发。
Near the end, he led all of Apple's just iPhone product development.
他在苹果是个非常关键的人物。
So very important person at Apple.
他于2024年离开苹果。
He left in 2024.
他与前苹果设计总监乔尼·艾维共同创立了IO Products，去年OpenAI以65亿美元股票收购了这家公司。
He co founded IO Products alongside former Apple design chief Johnny Ive and then OpenAI acquired it for $6,500,000,000 in stock last year.
现在他负责OpenAI所有设备硬件项目。
So he leads all of OpenAI's device hardware projects.
而程是另一位较年轻的员工，资历不如其他人深，但也是从苹果过来、经验非常丰富的工程师，最近刚被OpenAI挖来。
And then Cheng is a little more of a younger, one of the younger employees, less seasoned, but very experienced engineers from Apple that OpenAI recently brought over.
他今年一月离开苹果，加入OpenAI，到现在才几个月时间。
He left Apple in January and joined to join OpenAI, so he's only been there for couple months.
但确实，苹果指控他在为OpenAI工作期间，持续绕过苹果的安全系统，非法访问苹果内部网络。
But yeah, Apple accuses him of bypassing Apple security and accessing Apple networks on an ongoing basis while employed at OpenAI.
这份诉讼中的细节简直令人震惊。
And so the details in this lawsuit are pretty stunning.
据我回忆，有些细节是说，他们在招聘时，只要候选人还在苹果，就要求他们把硬件零件带到OpenAI的办公室来。
I mean, some of the details here from what I recall, I mean, it's the idea that they asked people when they were interviewing for OpenAI, they were at Apple, they asked them to basically bring or bring parts, know, components, I guess, of hardware to the OpenAI office.
我记的是这个意思吗？
Am I recalling this correctly?
是的。
Yeah.
他们指控唐在招聘过程中，要求工程师们带实际的硬件部件过来，搞所谓的‘现场展示’。
They accused Tang of during interviews of requesting engineers to bring actual parts and do a sort of quote unquote show and tell.
所以，他不仅把零件带到面试现场，还被指控利用项目代号打探苹果正在进行的秘密项目进展。
And so, you know, taking parts to these interviews and then also accused Tang of asking about ongoing secret projects using project code names and asking about updates for that.
也就是说，他利用自己对苹果早已掌握的丰富信息，进一步从在职员工那里套取更多内幕。
So yeah, using using his already, you know, pretty substantial knowledge about Apple and kind of use it to dig more information out from current Apple employees.
所以，亚伦，我们先退一步，说点别的。
So Aaron, I want to just take a step back here for a minute.
你一直追踪苹果的新闻，对吧？
You've been covering Apple for a long time.
苹果拥有科技行业最庞大的供应链之一，甚至可以说是最庞大的。
Apple has one of the most extensive supply chains, if not the most extensive supply chain in technology today.
你见过类似的诉讼吗？
Have you ever seen a lawsuit like this?
我的意思是，搞什么‘实物展示’？
I mean, a show and tell?
我们稍后再聊OpenAI对此的回应。
I mean, we'll get to what OpenAI is saying about this in a second.
我们一会儿会说到他们的回应。
We'll get to their response.
但你说，这种官司在硬件行业里常见吗？
But, you know, are these types of lawsuits, are they common in the hardware business?
你以前见过类似的吗？
Have you seen anything like it?
这事儿让人想起以前的一些诉讼。
There are echoes of previous lawsuits.
比如2019年，苹果就起诉过一家叫Nuvia的芯片初创公司。
For example, Apple went after a chip startup in 2019 called Nuvia.
那家公司由前苹果工程师联合创办，后来挖走了大量苹果的芯片工程师。
It was co founded by former Apple engineers, and the company ended up recruiting a large number of Apple chip engineers.
苹果对此非常不满。
And Apple really didn't like that.
他们指控那家公司有类似的行为。
They accused the company of sort of similar behavior.
嗯，有点不一样，算不上什么严重的窃取，但是——
Well, somewhat different, not a serious kind of stealing, but-
不是那种展示交流的情况，但是——
Not about a show and tell, but-
不是公开展示芯片，但确实能让人联想到当年那回事。
Not a show and tell of chips, but would certainly, you know, this has some echoes of that.
我觉得苹果特别反感前员工去挖现员工，像OpenAI这样有系统性地挖人，苹果一向不高兴，以前也发生过类似的事。
You know, I just think of Apple really doesn't like, you know, former employees going after current employees to recruit them and kind of the systematic effort that OpenAI certainly undertook is not something Apple has been happy about, and we've seen it in the past.
所以有相似之处，但OpenAI被指控窃取技术，这可就严重多了，简直有点离谱。
So similarities, but certainly the allegations of theft theft from OpenAI's part, that's, you know, quite serious and pretty wild.
苹果在诉讼中提出了什么证据来证明这些事确实发生过？
What is the evidence that Apple presents in the lawsuit for these things having happened?
他们说做过调查。
Mean, they say they did an investigation.
他们有没有详细公布调查结果？
Do they present any of their findings in detail?
是的，他们引用了一些能恢复出来的聊天记录，尤其是张的那些。
Yeah, they quote messages that they were able to retrieve, especially from Chang.
看起来他们一直能接触到他发给现任员工的消息。
They had a, it appears like ongoing access to messages he would send either to, you know, to current employees.
这似乎是因为他保留了一台苹果公司的工作笔记本电脑。
And it seems like they were able to do this because he retained an Apple Work laptop.
所以虽然他们没明确说明，但他们的安全系统应该能监控到他作为前员工仍在发送的消息。
So perhaps, you know, they don't lay out specifically, but you know, their security apparatus was able to access ongoing messages he was sending, even though he was a former Apple employee at that point.
他们还掌握了唐和现任员工之间的聊天记录和对话。
They also have just messages and conversations taking place between Tang and current employees.
这些内容很可能是通过采访参加过OpenAI面试的现任员工获取的。
So they were probably able to retrieve that through interviews with current employees that went through the OpenAI interview process.
对。
Right.
那现在OpenAI对此是怎么回应的？
So now what is OpenAI saying about this?
他们怎么回应的？
What's their response?
他们没说什么。
They haven't said much.
我觉得周五下午事情一曝光，他们就完全措手不及了。
So I think they are caught flat footed when it all came out Friday afternoon.
到目前为止，他们就说了些老生常谈，比如我们对其他公司的商业机密没兴趣，我们希望为所有人做创新。
So far, it's just been, you know, we have no interest in other companies' trade secrets, and we want to innovate for everyone.
回应真够简短的，几乎没什么实质内容。
So pretty, pretty small response there, short response.
对。
Right.
我们看看他们能不能
We'll see if they can
在这起诉讼曝光后，你发了一篇报道，提到OpenAI和苹果之间的紧张关系其实早就酝酿很久了。
Now you put out a story after this lawsuit broke talking a little bit about how the tensions between OpenAI and Apple have been growing for quite some time leading up to the lawsuit.
你能稍微讲讲你当时的报道吗？因为我们都还记得苹果那场炫目的发布会，当时他们宣布部分AI功能将由OpenAI提供支持。
Walk us through a little bit of your reporting there, because I think we all recall the flashy Apple keynote where they unveiled that part of their AI would be powered by OpenAI.
那次合作确实让不少人感到意外，但根据你的报道，之后的情况并不太顺利。
That was a union that I think surprised some people, but it hasn't been so great since then, according to your reporting.
是的。
Yeah.
2024年苹果全球开发者大会，他们推出了自己的大语言模型AI。
So 2024, Apple's Worldwide Developers Conference, they come out with their sort of AI LLM.
就是这种生成式AI。
This is the generative AI.
这正是他们的整体策略。
This is their strategy.
而这个策略中，有一部分就是与OpenAI的合作。
And part of that strategy is partnership with OpenAI.
萨姆·阿尔特曼当时出席了大会，看起来双方的合作开局相当顺利。
Sam Altman was at the conference and it seemed like a very positive start to the relationship.
但那也没持续多久。
But that didn't really last too long.
你知道，这事儿早就结束了。
You know, it's been over.
都两年了，这个合作对OpenAI来说也没带来多少实质成果。
It's been two years now and the partnership hasn't yielded much for OpenAI.
他们从这里面没得到多少好处。
They're not getting much out of it.
当时只是表面搭了个很浅的整合，OpenAI的聊天机器人通过Siri和其他苹果AI功能勉强接入，结果根本没搞好。
It was a very sort of like very superficial integration where OpenAI ChatBot was accessed in a very limited fashion via Siri and other Apple AI features, and it really didn't work out very well.
更别说，OpenAI最近开始大举挖角苹果的Tang Tang团队，招揽他们的硬件工程师。
And on top of that, OpenAI has begun a very aggressive endeavor of recruiting Apple hardware engineers from this Tang Tang team.
现在OpenAI也开始自己做硬件了。
And OpenAI is now just starting their own devices.
我们心里都清楚他们要做啥，这硬件布局明显就是直接跟苹果对着干。
I mean, we have a clear idea as to what it is, but I mean, this hardware play is a direct competition to Apple.
没错。
For sure.
对。
Yeah.
所以这事一直在进行。
So that's been going on.
OpenAI 现在已经有四百多名苹果工程师在做这个设备项目。
OpenAI now has over 400 Apple engineers working on this device project.
这确实让苹果很头疼。
And they you know, that's that's been very difficult for Apple.
过去了不少资深的硬件人才。
There's like very senior hardware folks going over there.
一些原本负责 iPhone、Apple Watch 和新硬件产品的核心工程师都走了。
You know, some of their best working on iPhone and Apple Watch and new hardware products that have been released.
这对苹果来说，人才流失确实挺严重的。
So this is, this talent drain's been pretty rough for Apple.
但有一件事挺奇怪的，乔尼·伊夫根本没出现在这场诉讼里，可他明明是核心人物。
And one name that wasn't included in the lawsuit, Joni Ive was not I mean, he's such a central figure.
我真没想到会这样。
I was surprised.
那乔尼·伊夫在这事儿里到底扮演了什么角色？
What do we know about the Joni Ive factor here?
据我们所知，他的离开好像挺平稳的，找不到更合适的词了。
I I guess it was a clean transition, lack of better words, from what we know?
是的。
Yeah.
这事儿挺有意思。
That was interesting.
他们确实刻意把他排除在外了。
They really kept him out of it.
我觉得这是有意为之，不想把关系搞僵，怕走得太远。
And I think that's like maintaining some sort of friendly relations with him, think is very intentional not to dip their toe, go too far.
起诉前首席设计官，也就是史蒂夫·乔布斯设计理念的接班人，对苹果来说可能太过火了。
Suing the former sort of chief design officer, sort of the heir of the Steve Jobs sort of design philosophy, would maybe be a step too far for Apple.
所以他们退了一步，选了唐，算是没那么咄咄逼人。
So they went one rung below that to Tang and kept it, you know, they were a little less aggressive.
那我想到的是，尽管乔尼·伊夫没被点名，也被隔绝在这场风波之外，但你觉得这场诉讼在多大程度上阻碍了OpenAI的进展，也影响了伊夫本人开发即将发布设备的能力？
So then I guess what I'm thinking about then is even though Joni Ive is not named in the lawsuit and he's kept separate from all this, to what extent do you think this lawsuit stifles OpenAI's progress, Joni Ive's own capabilities here to develop the devices that they are set to release.
今天还有一篇报道说，他们依然计划在明年或后年推出这款设备。
I mean, there was a report out today that they're still aiming to get a device out in the next year or two.
你觉得这真的会拖慢他们的进度吗？
Do you think that this stifles their progress at all?
我实在想不出这怎么会不拖慢进度。
I can't imagine how it won't stifle the progress.
他们计划明年二月发布第一款设备。
Yeah, they plan to release the first device February of next year.
这场诉讼的一部分，就是要求公司不得使用苹果的专有信息来开发这款设备。
Part of this lawsuit is demanding that the company doesn't use proprietary Apple information, this device.
我不明白OpenAI该怎么绕过这个问题。
I don't know how OpenAI can work around that.
他们已经为此忙了一年多了，早就有了完整的计划。
They've already been working on this for the past year or whatever, you know, plans they have in place.
不可能说推翻一切，然后彻底换个方向重新来。
They can't just kind of upend that and go a completely different direction.
所以最终推出的产品，苹果肯定会严加审查，我敢打赌他们一定能挑出点毛病来针对OpenAI。
So I, you know, the device that comes out, Apple is going to be scrutinized intensely, and I'm sure they can find something to go after OpenAI about.
这完全有可能让这款产品推迟发布，甚至叫停销售，法院完全可能介入干预。
So that could really stop the release of this product, stop the sales of this product, that they could get courts to intervene here.
我肯定，OpenAI在这事上会走得特别艰难。
I'm sure, you know, this is going be very tricky for OpenAI to navigate.
再确认一下，据我们所知，OpenAI正在开发的是一款智能音箱那样的设备吗？
And just to recap, as best we know, OpenAI, is it a is it a smart speaker type device that they're working on?
目前这方面的最新进展是什么？
What's what's the latest on that?
据我们了解，OpenAI 的第一款设备是一款智能音箱，不过我们也听说手机正在开发中，说不定之后还会有第三款、第四款产品，谁知道呢？
That's our understanding of the first OpenAI device, a smart speaker, though we understand, you know, a phone is in the works and hopefully maybe second device, third device, who knows?
他们的打算其实是推出一系列产品。
The intention is a family of products.
OpenAI 会负责推出这些产品，但第一款会是智能音箱。
OpenAI helps to release, but a smart speaker is the first one.
明白了。
Okay.
最后一个问题，亚伦。
And last question for you, Aaron.
我猜你背后一定在和律师、法律专家们讨论，这场诉讼的进展时间线是怎样的？
I'm sure you're talking to lawyers and legal experts in the background about What is your understanding of the timeline here for this lawsuit?
接下来会发生什么？
What we can expect to happen next?
这事儿是几个月内能解决，还是要拖上好几年？
Is this a matter of months, years?
我的意思是，这场官司得拖多久才能有结果？
I mean, how long is this going to take to play out?
对。
Yes.
几个月，几年，肯定得几年。
Months, years, years, definitely.
各种情况都有可能。
All of the above.
打官司哪能说停就停。
Doesn't running legal battle.
到那时候，我们说不定都用上翻盖手机了。
We'll we'll have a a flip flip phone by then for Sure.
我们还会回到那儿的。
We'll be back there.
是的。
Yeah.
是的
Yeah.
好的
Alright.
好了，亚伦，谢谢你今天来做客。
Well, Aaron, I wanna thank you for coming on.
我相信未来几周我们还会需要你对这件事的深入见解。
I trust that we will need more of your insight on this in the weeks to come.
这位是亚伦·蒂利，我们在《The Information》的苹果领域记者。
That is Aaron Tilley, our Apple reporter here at the information.
本周，一家重要的股票市场基础设施公司即将展示基于区块链的股票交易。
This week, a major stock market infrastructure company is set to demonstrate stock trading on blockchain.
这正是加密货币圈许多人长期支持的转变。
This is a transition that many people in crypto have been endorsing for a long time.
《The Information》的加密货币记者杨悦琪带来最新进展。
The information's crypto reporter, Yueqi Yang, has the latest.
她今天早上刚发了一篇相关文章。
She published a piece on it this morning.
我想请她来聊聊这件事的来龙去脉。
I wanna bring her on to talk all about it.
Yueqi，欢迎再次做客节目。
Yueqi, welcome back to the show.
很高兴你能来。
It's great to have you here.
嗨，Akash。
Hey, Akash.
好的。
Alright.
咱们来聊聊这家机构吧，存管信托与清算公司。
So let's talk about this company here, the Depository Trust and Clearing Corporation.
我真没想到咱们这节目会提到它。
Not a name that I thought we would be talking about on the show.
你写了一整篇关于他们的文章。
You wrote a whole story about them.
他们到底是谁？为什么我们得了解他们？
Who are they, and why do we need to know them?
DTCC 是美国金融市场最关键的一家公司。
So DTCC is the most important company that Okay.
明白了。
Get it.
挺有意思的。
It's cool.
DTCC。
DTCC.
它让美国股市能够正常运转，主要提供清算服务。
Allows The US Stock Market to function because it provides a few services, mostly including the clearing of the stocks.
每天在 DTCC 交易的股票总价值高达两万亿美元。
Stock market trades at the DTCC for about $20,000,000,000,000 worth of stocks per day.
这数额可真不小。
So it's a huge amount.
他们不仅负责股票的结算，还负责保管这些股票的份额。
They provide a settlement of the stock, and then they also provide custody of the shares of the stocks.
现在他们开始涉足代币化，也就是允许会员把资产转移到区块链上。
So they are getting into tokenization, which, means they're starting to allow, their members to move assets onto the blockchain.
这一举措的第一步将在本周三实现，届时他们会演示几笔真实的生产环境交易，涉及股票、ETF以及其他国债和其他资产，这些都会在区块链上流转。
And the first step of that will happen this Wednesday when they demonstrate a few live real production trades of assets, including stocks, ETFs, and other treasuries and other assets that are going to move on the blockchain.
他们选好要上链的股票了吗？
Have they picked which shares they're going to put on the block?
就是说，是哪些具体的公司？
Like, are these specific companies?
我们有了解相关信息吗？
Do we know anything about that?
有的。
Yeah.
他们之所以能推进这种代币化，是因为去年DTCC获得了美国证交会的批准，可以对资产进行代币化，包括规模最大的美国股票，也就是罗素1000指数里的那些股票。
So they're allowed to pursue this tokenization effort because last year, DTCC got the green light from the SEC, to provide tokenization of assets, including the largest, US stocks, that are, making up the Russell 1,000 index.
所以这些股票会是其中一部分。
And so that's going to be part of it.
其他资产，比如国债，也会在首发时一并上线。
And the other assets like treasury bonds are also going to be part of the launch.
那DTCC是第一个做这个的吗？
Now is DTCC, are they the first to do this?
毕竟加密公司一直都在说要干这事。
Because, I mean, crypto companies have been talking a lot about doing this.
我记得以前好像已经有人做过类似的事了。
I mean, my recollection, I feel like this has happened before.
我是不是记错了？
Am I wrong?
他们的做法和视角跟别人不一样。
They're approaching it with different methodology and different vantage point.
一些加密货币公司，尤其是像币安、Bybit，最近还有Coinbase，已经开始上线由股票支持的股票代币。
So a few crypto companies, especially the crypto exchanges like Binance, Bybit, most recently Coinbase, are starting to list stock tokens that are backed by stocks.
这些代币通常是由第三方机构发行，而不是由股票本身发行方直接发行。
And then these tokens are usually issued by third party providers rather than the issuer themselves.
它们的策略更侧重于面向散户的代币分发。
And then their play is more gearing towards retail distribution of these tokens.
但DTCC不一样，它是一家市场基础设施服务商。
DTCC is different because they're a market infrastructure provider.
它为DTCC的会员提供代币化服务，这些会员主要是大型银行、交易公司、经纪商等等。
They're providing a tokenization service to the members of the DTCC, which mostly includes big banks, trading firms, brokerages, etcetera.
所以这更偏向于机构层面的布局。
So this is more about institutional play.
我们来聊聊代币化股票交易的优势和局限性吧。
So let's talk about the advantages and then also the limitations of tokenized stock trading.
我觉得我们之前提到过一个明显的优势，就是区块链24/7全天候运行的特性，这能给股票交易带来什么改变。
I mean, one clear advantage I think we've talked about is the 20 fourseven nature of blockchain and the fact that that is what it could bring to stock trading.
这样是不是更便宜？
Is it cheaper too?
我们能看到哪些好处？
What are the benefits that we see?
这是个好问题。
That's a great question.
每当听说公司推进区块链项目时，首先要问的是：区块链到底带来了什么优势？哪些事情是传统方式做不到，但区块链能实现的？
Think whenever you read about companies pursuing blockchain initiative, the first question to ask is really what are the benefits, and what are something that blockchain enables that they can pursue in the traditional setup.
我觉得DTCC的这个项目，目标是解决几个具体问题。
And I think for the DTCC effort, there are a few benefits that they're, trying to target.
首先是，他们让那些已经清算、已完成结算、目前存放在DTCC托管的股票，能够被代币化，这样公司之间就可以直接以代币形式转让这些股票。
The first one, is they're providing the ability for stocks that are already cleared, already settled, are currently sitting at custody at DTCC to be tokenized so that companies can transfer these stocks in token format to each other.
那为什么要这么做呢？
And why do you do that?
这么做能带来不少好处，比如公司可以把这些代币作为抵押品，向其他公司融资，而且可以7x24小时实时操作，连周末也能进行。
That unlock a few benefits, including companies pledging these tokens as collateral so that they can get funding from another company, and they can do this twenty four seven, do it real time, and on the weekend as well.
所以，就像保证金交易，是这个意思吗？
So like margin trading, is that the idea?
它能实现这一点。
It enables.
实现。
Enable.
哦，我明白了，保证金交易在这里是指直接用实际股票作为抵押，而不仅仅是普通保证金交易。
Oh, guess margin trading, this would be posting literal shares as collateral, not just on margin.
为了获得保证金，获得杠杆。
To get margin, to get leverage.
对。
Right.
明白了。
Okay.
所以这是一大优势。
So that's one benefit.
我能理解这可能会助长一些更冒险的操作，这么说吧。
I could see how that would allow some more risky behavior, call it.
杠杆交易一直存在。
Margin trading always exists.
抵押资产这个流程一直都有，但在区块链上可以以不同方式实现，还能带来额外的好处，比如速度更快。
Posting collateral, this process always exists, but you can do it in a different way on the blockchain, which brings some additional benefits, with speed.
对。
Yeah.
此外还有其他优势。
So there are other benefits as well.
我个人觉得最有趣的一点是，它有可能在美股市场和加密市场之间搭起一座桥梁。未来的一种可能是，所有提供以美股为Backing的加密代币的交易平台，都可以直接使用DTCC的代币化股票作为底层资产。
And personally, I find it the one that I find to be most interesting is that it potentially provides this bridge between The US stock market and the crypto market because one potential outcome in the future is that for all these crypto exchanges that are providing crypto tokens backed by US stocks, they can potentially use the DTCC tokenized stocks as the underlying asset backing the stock, crypto tokens.
理论上，这种方法比当前的方式更优，因为它更直接地关联到美国最重要金融基础设施核心的真正股票。
And that could, in theory be a superior method than the current, method because it's a more direct claim to the real stock that's sitting at the center of the most important market infrastructure we have in The US.
那代币化股票交易有什么局限性吗？
What about limitations to tokenized stock trading?
有什么限制吗？
Are there any?
我得指出，DTCC 并没有用区块链来清算股票。
I think it's important to point out that DTCC is not using blockchain to clear stocks.
这是最常见的误解之一。
That's one of the most common misconception.
DTCC 自己也说过，用区块链清算股票根本不现实，因为他们每天清算的股票价值超过两万亿美元。
DTCC itself said using blockchain to clear stocks is just impractical because every day it clears more than $20,000,000,000,000 worth of stock.
根本做不到，而且他们每天结束时会把大部分交易净额抵消，也就是说，股票在各方之间来回买卖，其实没必要为每一笔交易都转账。
It's just there's no way to and they net out big percentage most of these stocks at the end of the day, meaning that companies trade back and forth, so they don't actually need to transfer money for each of the transaction.
他们到收盘时统一结算净额，只处理其中一小部分股票的交割。
They net it out at the end of the day, and they only settle a small portion of those stocks.
所以整个清算和结算流程根本没有改变。
So this whole process of clearing and settlement is unchanged.
他们没用区块链来做这个，真正用区块链的，只是针对那些已经清算完毕、已经交割、一直闲置在 DTCC 的股票这一小部分。
They're not using blockchain for it, and that they're really using blockchain, to target a small part of this, process, which is what what you do with stocks that are already cleared, already settled, and just kind of sitting idly there at the DTCC.
所以根据他们说的，他们是觉得区块链根本不可能用于清算吗？
And so based on what they're saying, I mean, are they sort of suggesting that they won't ever be able to clear on blockchain?
他们是觉得这难度太大了吗？
Are they saying that's a much bigger lift?
这到底是根本性的限制，还是只是还没做到这一步？
I mean, is this a clear limitation or is it just that they haven't gotten to it yet?
目前来看，他们并没有计划用区块链来处理清算。
I think for now they don't have plans to use blockchain to clear trace.
明白了。
Right.
对。
Yeah.
他们给我发过一份说明，说那种完全依赖逐笔结算的模式，在某些区块链环境中很常见，但在这个规模下根本行不通，因为系统里没有足够的资金来支撑。
They gave me a statement saying that settlement models that rely entirely on growth settlement, which are common in some blockchain environments, simply do not work at this scale because there's not enough money in the system to support it.
Yueqi，从宏观角度看，这里的关键是。
Yueqi, I mean, big picture here.
这件事进展得比你预期的快还是慢？
Is this moving faster or slower than you thought?
我们讨论在区块链上交易股票已经很久了。
I mean, we've been talking about trading stocks on the blockchain for quite some time.
当初对它寄予了很高的期望。
There were big ambitions for it.
你整体怎么看？你觉得这事最终会怎么发展？
What's your overall perspective on it and your opinion on how this is shaking out?
我觉得这会是一个持续多年的工程。
I think this is going to be a years year long effort.
DTCC 已经在做代币化工作很多年了。
DTCC has been working on tokenization for many years already.
这一步很重要。
This is an important step.
现在终于得到了监管机构的支持。
It now has the support, from regulators.
所以我觉得他们正在取得实质性的进展。
So I think they are making a meaningful progress.
不过，我也采访了几位市场结构方面的专家，为这个报道做调研。
But, I also spoke to market structure experts for this story.
其中一位是彭博的拉里·塔布，他说他不认为DTCC的代币化股票能占到传统股票总市值的百分之零点五。
One of them, Larry Tab, at Bloomberg said he doesn't expect DTCC tokenized stocks to reach even half of a percent of the value of total traditional stocks.
他还提到，这整个过程非常漫长且复杂。
And then he mentioned this is just a really long, complex process.
所以这需要时间。
So it's going to take time.
Yueqi，我想快速问你一下另一个我们一直在关注的故事。
Yueqi, I want to ask you very quickly about another story that we've been following.
Circle刚刚获得了银行牌照的批准。
Circle got the green light for its bank charter.
这个项目其实已经筹备好几年了。
This was another story that has been in the works for a couple of years now.
这个银行特许状有什么重要意义？
What is the significance of this bank charter?
它让它们能做些什么？
What has it allowed them to do?
Circle获得了国家信托特许状的批准。
Circle got approval for a national trust charter.
这其实早有预料，因为去年年底，Circle和其他几家公司就已经获得了有条件批准。
This is the widely expected news, because Circle and a few other companies got the conditional approval of it, late last year.
这项批准的关键在于，它让Circle现在有了联邦监管机构，也就是货币监理署（OCC）。
And the significance of this approval is that it basically allows Circle to now have a federal regulator, which is the OCC.
这正是《天才法案》的初衷。
And this is as intended by the Genius Act.
这意味着现在有了一家全国性的监管机构，负责监管Circle的稳定币。
It means that there's a national regulator who is now regulating Circle stablecoins.
此外，有了这个特许状，它们还能提供其他一些服务。
And then in terms of additional services, they can provide Circle with its charter.
现在他们可以利用这个国家信托特许权提供机构托管服务，还能管理支撑其稳定币的储备资产。
They can now provide institutional custody using this national trust charter, and they can also manage the reserve assets backing up their stablecoins.
需要注意的是，这并不是传统意义上的全面银行牌照。
And, it's important to note that this is not a traditional full service bank charter.
这并不意味着Circle会变成面向普通用户的银行。
It doesn't mean Circle is becoming a retail facing bank.
他们不会接受你的存款。
They're not going to take your deposit.
他们也不会向你放贷。
They're not going to lend you money.
这是一种叫做国家信托特许权的特定类型牌照。
This is a specific type of charter called National Trust Charter.
Yueqi，你认为其他加密公司现在也会去申请这种银行牌照吗？
And Yueqi, do you expect other companies to go after this bank charter now, other crypto companies?
是的。
Yeah.
今年很多加密公司都在争取这个特许牌照，尤其是打算发行稳定币的公司。
It's a very popular charter for crypto firms to pursue this year, especially for the ones that are intending to issue stablecoins.
目前已经有一些值得关注的申请者。
So there are a few notable applicants.
我觉得Stripes Bridge也是另一个值得关注的公司。
I think Stripes Bridge is also another important one to watch.
他们也申请了这个国家信托特许牌照。
They also applied for this national choice charter.
很好。
Great.
Yueqi，非常感谢你今天来做客。
Well, Yueqi, I wanna thank you for coming on.
这位是 Yueqi Yang，我们《The Information》的加密领域记者。
That is Yueqi Yang, our crypto reporter here at The Information.
Stephanie Palazzolo 是我们负责人工智能报道的记者，也是《The Information》AI Agenda 订阅通讯的作者，她刚从韩国回来，参加了国际机器学习大会。
Stephanie Palazzolo, our AI reporter and author of the Information's AI Agenda newsletter, is fresh off a trip to South Korea where she attended the International Conference on Machine Learning.
这在人工智能研究领域可是件大事。
It is a big deal in the land of AI research.
她今天在自己的通讯里整理了一些重要的收获。
She wrote up some of her big takeaways for her newsletter today.
我想请她来聊聊这些内容。
I wanna bring her on to to talk about all of them.
Stephanie，欢迎再次做客我们的节目。
Stephanie, welcome back to the show.
很高兴你能来。
It's great to have you here.
谢谢。
Thanks.
回来真好。
It's great to be back.
旅途怎么样？
How was the trip?
还不错。
It was good.
特别有意思。
It was super fun.
我确实有点时差，但绝对值得。
I am definitely jet lagged, but it was it was it was worth it.
所以是ICML。
So ICML.
明白了。
Okay.
几年前，还没几个人关注这个缩写。
This is not an acronym that people needed to pay attention to a couple years ago.
现在它已经是AI研究和创新的重要中心了。
It is now one of the hotspots for AI research and innovation.
你今天早上在通讯里写了一些关键收获。
You wrote about some of the key takeaways this morning in your newsletter.
我们来聊聊其中几条吧。
Let's go through a couple of them.
这次会议让你印象最深的是什么？
What stood out to you from the conference?
是的。
Yeah.
有几件事特别突出。
So there were a couple of things that stood out.
我觉得尤其是把这次会议和每年一次的另一个大型AI会议NeurIPS对比一下，我去年十二月参加过那个会。
I think especially comparing this conference to the other really big AI conference that happens once a year, NeurIPS, which I attended in December.
我觉得部分原因是，NeurIPS正好发生在一大批顶尖研究者开始热议整个领域需要彻底颠覆之后。
I think part of it was that NeurIPS happened right after a lot of really big, you know, big name researchers started to talk about how the field needs to be completely kind of up upended.
所以去年十二月在NeurIPS上，大家讨论得很多的是，我们今天用的模型根本不可能带我们走向通用人工智能，或者我们必须彻底改变模型的架构。
So a lot of talk at NeurIPS last December was around, you know, the fact that the models that we use today are not are are not gonna be what gets us to to AGI or that we need to completely change the architectures of the models that we use.
六个月前，AI圈里确实弥漫着一股叛逆的情绪。
It was very much as kind of like rebellious sentiment in the AI field, I think six months ago.
我觉得现在在ICML上，气氛完全不同了。
And I think now at ICML, mood is very different.
很多人更关注实际的问题，比如怎么让模型跑得更便宜、训练成本更低。
I think, you know, a lot of people were more focused on practical questions like how do we get models to run and be trained for cheaper?
大家更注重效率和降低成本，而不是像以前那样，觉得必须彻底颠覆一切，重新发明一套全新的方法。
A lot more of a focus on efficiency and lowering costs rather than, you know, being very like, we need to completely turn everything on its head and come up with a completely new way of of of of doing things.
所以现在明显多了很多这类实际问题的讨论。
So definitely a lot more of these kind of practical questions.
同时，有一个词反复出现，就是RSI，这是‘递归自改进’的简称，意思是当模型足够先进后，它们能帮助训练下一代模型。
At the same time, I think a a phrase that came up over and over again is RSI, which is the kind of short, you know, shorthand way of saying, recursive self improvement, which is this idea that once models become advanced enough, they can actually help to train the next generation of models.
我参加的很多小组讨论中，研究人员问得最多的问题就是：你觉得我们离RSI还有多远？
So a lot of panels I attended, the number one most common question that came up from researchers was how close do you think we are to RSI?
这对研究人员意味着什么？
What does this mean for researchers?
我会不会被自动化取代？
Am I going to get automated out of a job?
事实上，还有一场专门讨论研究人员在面对RSI或可能取代他们工作的自动化趋势时，该如何应对的演讲。
In fact, there was a whole talk dedicated actually to what researchers should do in the face of RSI or something that might potentially automate away their jobs.
这可以说是一种延伸。
So this is kind of an expansion.
一年前，大家还在讨论，当AI开始能写代码时，软件工程师还能有工作吗？
Mean, a year ago, the conversation was, well, how are software engineers going to have a job in world where AI coding takes off?
现在情况又往前走了一步。
Now it's gone a step further.
连AI研究人员自己都在问：如果模型能自己训练自己，那我存在的意义是什么？
AI researchers are the ones saying, Hey, if the models can train themselves, then what am I here for?
听起来，大家对这个问题挺焦虑的。
It sounds like there was a lot of anxiety around that.
有没有人站出来为AI研究辩护，说AI还是有很多做不到的事，或者自动化研究者目前还替代不了人类？
Was there also people defending AI research and saying, Well, there are a lot of things that AI research still cannot do, or that AI automated researchers cannot do.
当然有。
Definitely.
展开剩余字幕（还有 199 条）
不，我觉得在这场演讲中，普林斯顿的一位教授提出了一个关键观点：研究人员其实没必要对AI抢走工作感到过度焦虑。
No, I think that was a big point during this talk that I mentioned was this professor from Princeton was arguing, in fact, that researchers don't need to be super anxious about AI taking away their jobs.
这种说法我们其实听过很多次了，不管是对软件工程师、律师还是医生来说。
I think this is a common argument that we've heard for, you know, whether it's software engineers or lawyers or doctors.
很多人都说，工作中那些重复、琐碎的部分，AI确实能帮上忙。
Lot of people argue, hey, there's a lot of menial tasks and more tedious parts of our job that AI can, can, you know, can, can do.
但真正核心的、更有创造性的部分，AI还很难替代。
But a lot, a lot of the parts of our jobs that are more creative.
就拿AI研究员来说吧。
So like as an AI researcher, sure.
你现在的工作可能主要是写代码、跑实验，这些AI确实能做。
You know, a lot of your job today is maybe writing code or running experiments, which maybe AI can do.
但真正有价值的是人类才能提供的创造力——比如提出全新的假设，设计出新的实验方案或模型架构，这些目前AI还远远做不到。
But maybe the more valuable part that humans can really add is the, you know, the creativity needed to actually come up with these new hypotheses in the first place or these new ideas for experiments and model architectures that, you know, today it's very hard for AI to do.
所以我觉得，现场既有焦虑，但也有一种对AI研究员未来前景的乐观情绪。
So I think there was a mix of anxiety but also maybe optimism for, you know, the future of AI researchers and their jobs.
那我们回到RSI，还有更广泛的持续学习话题。
So let's go back to RSI and then continual learning broadly.
我的意思是，你和我之前聊过，RSI主要关注的是模型本身。
I mean, you and I have talked RSI is mostly about the models.
我们也讨论过在智能体和智能体循环中，持续学习如何让智能体自我提升。
We've also talked about continual learning in the context of agents and agent loops and the prospect of agents improving themselves.
那关于智能体，当时大家是怎么讨论的？
I mean, what was the conversation around agents?
这个话题有提到吗？
Did that come up?
智能体确实被频繁提到。
Agents did come up a lot.
我注意到有很多论文都在谈多智能体系统，就是多个智能体协同工作。
I did notice that there were tons of papers about, this idea of multi agents, which are, you know, agents working together.
我觉得这是因为现在我们交给AI的任务越来越复杂，单个智能体已经很难独立完成了。
I think it's because there's this idea now that, a lot of the tasks that we're being, that we're beginning to give to AI, it's so, you know, it's so complex that a single AI agent can't do it on its own.
所以现在，企业推出的那些智能体系统，都是同时启动多个智能体，各自负责任务的不同部分，然后协同工作。
So instead, you know, these, these, you know, agent systems that companies are selling are spinning up multiple agents to, you know, each tackle certain parts of a job and work together.
智能体这个概念确实非常热门，过去几年一直如此，但特别火的是多智能体架构，以及多个智能体一起解决问题时会出现的各种难题，还有怎么去应对这些问题。
So definitely the idea of agents was super hot, you know, as it has been for the last couple of years, but specifically this idea of, multi agent art architectures and the problems that arise when you have multiple agents working on a problem and, and you know, how, how to kind of come up with a solution to that.
有没有人就此展开过争论呢？
Was there any debate at all?
就像之前讨论RSI那样，关于持续学习在智能体场景下能走多远、效果如何，有没有人提出不同看法？
In a similar sense around RSI, was there a debate around how effective or how far continual learning will go in the context of agents?
这种观点有出现过吗？
Was that a perspective that came up?
有的。
It was.
现在大家开始区分的是，到底什么样的突破才能让人工智能对经济产生重大影响。
Think kind of the distinction that people are starting to draw when it comes to continual learning is, you know, what is something like, what is a breakthrough that we need to have for AI to be, to, to have a big impact on the economy
而且
and
让公司赚大钱，和我们需要什么样的突破才能实现超级智能，这两者是不一样的。
making companies lots of money versus what is a breakthrough that we need to have to have like super intelligence.
没错。
Right.
我觉得很多人在说，就算现在这种没有持续学习能力的AI，也已经足够了。
I think a lot of people were making this argument of like, you know, even with just the AI that we have today, no continual learning.
像OpenAI和Anthropic这样的公司，光靠这种AI就能赚上百亿美元，对吧？
Companies like OpenAI and Anthropic are still able to make billions of dollars from this sort of AI, right?
照样对经济产生了巨大影响。
Still having a really big impact on the economy.
所以一方面，大家觉得，我们其实并不需要持续学习，也能看到今天这种AI带来的巨大经济效应。
So on one hand, it's kind of like, okay, we don't really need continual learning to have this sort of big economic impact of AI that we see today.
但另一方面，人们还是认为，持续学习这个理念很重要——也就是AI在训练完成后还能继续学习。
On the other hand, I think people are still saying like, Hey, the idea behind continual learning as like AI that can continue to learn even after it's trained, that's still really important.
从理论上讲，如果你想象一个超级智能的AI，它一定具备在训练后还能不断学习的能力。
And like theoretically, if you were to imagine the super intelligent AI, you would imagine that it would be able to do continual learning and learn even after it's trained.
这更多是锦上添花，而不是当务之急，毕竟现在我们看到的AI已经能赚大钱了。
That's more of like a nice to have, I guess, versus like a need to have when it comes to, AI that can actually make lots of money today, which is what we're already seeing.
对。
Right.
明白了。
Okay.
我们快速过一下你提到的其他几个主题。
Let's go through a couple of the other themes you wrote about quickly.
你还提到扩散语言模型是某些讨论的核心。
You also wrote about diffusion language models being at the center of some conversations.
什么是扩散语言模型？大家围绕它们在讨论什么？
What are those and what was the conversation around them?
是的。
Yeah.
总的来说，正如我提到的，很多话题依然很务实，集中在效率、成本这些问题上。
So I think in all this, as mentioned, a lot of topics were still very practical and focused on efficiency and costs and things like that.
我觉得有一个领域，大家讨论得挺多的，可能有点实验性，也比较前沿，就是扩散语言模型这个概念。
I think one area where I did see a lot of chatter around maybe something a little bit more experimental and out there was this idea of diffusion language models.
简单说，就是把原本用在图像模型上的思路，现在用到了文本领域。
Essentially what that is, is it's taking this idea that's been applied to image models and now applying it to the world of text.
现在的文本模型是怎么工作的呢？它们是从左到右，一个字一个字地生成回复，对吧？
So the way that today's text models work is that they output, a response to you left to right, right?
比如你用ChatGPT的时候，能看到它就像在你眼前逐字逐句地写出来。
Like whenever using ChatGPT, you see it, it will literally just start writing the sentence like you're reading it.
而扩散模型的想法不一样，它不是逐字输出，而是直接一次性生成完整的回答。
And with the idea of diffusion models is that instead of kind of outputting an answer word by word as if you're reading it from left to right, it actually outputs the entire answer all at once.
你可以想象，整段文字一下子出现在你屏幕上。
So you just imagine this entire paragraph of text appearing on your screen.
但这段文字一开始可能看起来杂乱无章，或者根本说不通。
But that paragraph of text might seem random or it might not make a lot of sense.
里面的内容也可能显得不对劲。
Things in it might look wrong.
然后随着时间推移，AI模型会慢慢调整这段文字的各个部分，直到最终呈现出正确的答案。
Then slowly over time, the AI model will start to tweak different parts of that paragraph until it ends up looking like the correct answer at the end.
这听起来有点抽象，不太好想象，但你可以把它理解成：先生成一大段文字，再不断优化，而不是像人那样从左到右一字一句地写出来。
So it's a bit theoretical and hard to imagine, but you can kind of imagine it more as like outputting this giant chunk of text and then refining it versus writing the answer like a human might from, from left to right.
所以你的意思是，我不用看着Claude一个字一个字地打出来？
So you're saying I don't have to watch Claude type everything.
你不用再盯着Claude慢慢打字，而是看到它先丢出一大段乱糟糟的文字，然后这些文字会慢慢调整，变得越来越通顺。
Will just Instead prevent of watching Claude type everything slowly, you would instead of watch it give you this entire mumbled, you know, chunk of text, and then slowly the text would kind of morph into something that made more more sense.
明白了。
So, Okay.
说实话，我真说不准哪一种更好。
I don't I like, truthfully, I don't know which one Exactly.
哪个更优，我也说不清楚。
Is better.
没错。
Exactly.
是的。
Yeah.
有点像是这样，好吧。
It's kind of like, alright.
这两种方式听起来都不太像是好的用户体验。
Both of those kind of sound like weird user experiences.
对。
Yeah.
如果这就是AI研究的未来，那也没办法。
If this is the future of AI research, then fine.
我的意思是，我每次还是得跟Claude说，你得再简洁点。
I mean, you know, I'm still I still tell Claude every time, man, you got to be more concise.
给我缩短到四分之一的长度就行。
Just give it to me in a quarter of the length.
明白了。
Okay.
斯蒂芬妮，我们还有很多话题需要请你回来继续聊，尤其是人工智能的安全与风险问题，但我强烈建议大家去读一读你的通讯。
Well, Stephanie, there's a lot more that we will have to bring you back on to talk about including the safety and risk component of AI, but I encourage everyone to read your newsletter.
这位是斯蒂芬妮·帕拉佐洛，我们《信息》杂志《AI议程》通讯的作者。
That is Stephanie Palazzolo, author of our AI Agenda newsletter here at the information.
劳德研究所正迅速崛起，成为推动人工智能研究的权威机构。
The Laude Institute is quickly gaining cachet as a leading institution promoting AI research.
随着模型发布一个接一个地涌现，这一点变得越来越重要。
That is becoming all the more important as model releases are coming fast and furious.
我想请布拉登·汉考克来聊聊。
I wanna bring on Braden Hancock.
他是劳德的研究合伙人，也是Meta前人工智能负责人，能带我们梳理一下当前技术所处的阶段。
He is a research partner at Laude and the former director of AI at Meta to walk us through this moment for the technology.
布拉登，欢迎来到节目。
Braden, welcome to the show.
很高兴你能来。
It's great to have you here.
谢谢邀请我来。
Thanks for having me.
你上周也去参加ICML了吗？
Were you at ICML last week as well?
可惜我没去。
I was not, unfortunately.
会议太多，没法全参加，但我一直在远程关注进展。
Can't hit all the conferences, but I've been following along from afar.
会议确实太多了，没去也正常。
There's a lot of them, so you can be forgiven for that.
先给我们简单介绍一下Laude研究所，然后我想听听你对最近发布的一系列模型的看法。
Tell us a little bit about what the Laude Institute is, and then I want to get your take on some of the model releases that we've seen coming out.
好的。
Sure.
是的，Laude研究所是由安迪创立的，他之前共同创办了Databricks，后来又共同创办了Perplexity，接着又创办了这个机构。
Yeah, Laude Institute was founded by Andy who, co founded Databricks and then co founded Perplexity and then co founded this.
你要是留意的话，会发现很多这样的情况，这正是安迪的一大优势。
And if you notice a lot of in there, it's, it's because that's one of Andy's big strengths.
他特别擅长把不同的人聚在一起，让整体效果远超个体之和。
He's very good at bringing together communities and, making, you know, the sum be greater than the than the the sum of parts.
是的，没错。
Oh, yeah.
安迪一直有个愿景，就是让研究人员在恰当的时候获得所需的资源，从而做出更大的突破，推动他们的成果得到更广泛的应用。
So so Andy's been had this vision, I think, for, you know, getting the researchers the right resources at the right time to do to do more, to have bigger breakthroughs, to see more adoption and traction of of what they're working on.
在其他几位联合创始人的共同努力下，我们这个研究所已经沿着这条道路走了一年多，逐渐成为研究社区的基地，通过各种项目帮助他们对接资源、机会和人脉，提升研究的影响力。
So, you know, with a great collection of other cofounders as well, we've been, on that path for about a year as as the institute, becoming sort of the home base for the research community and, through a variety of different programs helping to to connect them with resources, opportunity, connections, things to have more impact with their research.
明白了。
Right.
这么说吧。
And so look.
最近几周，我们看到模型发布的速度惊人，比如 Fable 5.6、Grock 4.5、4.6？
We well, you know, last couple of weeks, we've seen a staggering pace of model releases between Fable 5.6, Grock 4.5 4.5, 4.6?
没错。
That's right.
具体是4.5记不太清了。
Can't remember the exact 4.5.
4.5。
4.5.
好的。
Okay.
你对这些模型怎么看？
What was your review of these models?
我的意思是，推特上消息一个接一个地刷出来。
I mean, see the the tweets coming fast and furious.
很多人都在讨论5.6能删除Mac电脑上的整个硬盘。
There's a lot of chatter about 5.6 being capable to delete entire hard drives on Macs.
我看到过这个说法。
I've seen that.
我也看到萨姆·阿尔特曼回应过Fable价格高得离谱的事，这一点我们之前报道过。
I've also seen Sam Altman responding to Fable being extraordinarily expensive, which is something that we've reported.
Grok算是个特例。
Grock is a bit of an outlier.
你能不能详细说说你对这些模型的评价？
Walk me through your reviews of all these models.
是的。
Yeah.
每推出一个新模型，自然都会伴随一套说法，CEO们总会把故事讲得对自己特别有利，好像他们永远站在帕累托前沿似的。
I mean, every individual model release comes, of course, with its own narratives, and they'll with CEOs and tell the story in a way that, of course, always shows them up into the right and, you know, on the Pareto frontier.
但我觉得更值得玩味的是，现在有这么多玩家，距离技术前沿已经近到几乎触手可及了。他们不需要完全持平，只要够用、够便宜，比其他选项划算，就能站稳脚跟。
But I think the bigger picture here that that is fascinating is just the number of different players that are now, I'd say, you know, within, you know, spitting distance of the frontier and able to sort of claim, It's not parity at least, you know, kind of good enough and often cheap enough, you know, cheaper than other alternatives that they do become viable.
过去一周左右，就像你提到的，新推出的Grok 4.5模型表现出人意料地强劲。
So, I mean, in the past week ish seen, you know, as you mentioned, the new Grok four five model, surprisingly strong.
Muse Spark 1.1。
Muse Spark 1.1.
Muse，我甚至都忘了最近发布的那一个，而你对它很熟悉，因为你以前在Meta工作。
Muse, I don't I even I literally forgot about the most recent one, which is one that you're close to because you were formerly at Meta.
没错。
That's right.
是的。
Yeah.
我其实了解他们的背景。
I mean, I knew family.
我还在Llama刚推出、模型权重还开放的时候就在那里了。
I was there in the in the llama days when they were still open weight.
所以那边也有一些明显的不同。
So there's some noticeable differences there as well.
不过，我们最近也看到了TLM 5.2这个模型。
But, you know, yeah, we're seeing, I mean, the TLM 5.2 model as well.
对很多人来说这非常令人兴奋，因为这是一个开源权重的模型，在编程等关键领域表现相当有竞争力。
Very exciting for a lot of people because it is an open weight model that's at that, in that class of, you know, reasonably competitive in in certain spaces that matter to a lot of folks like coding.
看到这些真的很棒。
So, very cool to see.
我觉得，发布当天宣传的东西是一回事，但公众后续的反应才更真实、更有参考价值。
And and and I think you see, you know, there's there's there's like what gets told on release day, and then there's the court of public opinion that follows that I think is the more relevant one.
还有那些由社区运营、独立验证的评估结果，能帮我们看清哪些是真实进展，哪些只是宣传话术。
The the other evals that are used to that are sort of community run and independently validated that are helpful for getting the full picture of, you know, what's real and what's what's narrative spin.
我们发现，这些模型显然不是单纯为了刷榜而做的。
And we are seeing, I think that these these models do not appear to have been simply, you know, benchmarks, so to speak.
这里确实有一些实实在在的提升。
There are some real gains here.
我特别喜欢一篇来自Databricks的Zaharia写的博客，他们用内部数据测试了多个模型，发现不同系列的模型在各自任务上都达到了帕累托最优。
One blog post that I liked quite a bit came from, Meta Zaharia at Databricks, where they showed, we evaluated, you know, a number of these models with different harnesses on internal Databricks data and found that multiple different model families were on the Pareto frontier for their tasks.
这种表现，公司不可能在开发过程中靠调参就能做到。
And that's not something that these companies could have hill climbed on during development.
所以，这对我们理解这些模型真正的评估价值，是个非常有力的佐证。
So that's a very interesting validation when it comes to seeing the true sort of, you know, eval story here.
不过咱们还是回头说说Muse吧。
But let let's go back to to Muse.
你是前Meta人工智能部门的负责人，对吧？
I mean, you were the you were a former director of AI at Meta.
所以这个团队，你比其他团队都更了解。
So this is this is a team that you know, well, better than the other teams at least.
Shaw 1.1发布了。
Shaw 1.1 come out.
你觉得它表现得更好吗？
Were you better?
它有没有超过你的预期？
Is it better than what you thought?
你原本期待得更多吗？
Were you expecting more?
你怎么看？
What do you think?
是的。
Yeah.
在人工智能领域，一年简直就像一辈子，而我待在劳德的时间也就差不多这么长。
So a year is is an eternity in the AI space, and a is about how long I've been at Laude.
这期间变化太大了，Meta内部也经历了好几次重组。
So a lot of things change, a lot of re orgs at Meta.
不过我觉得，他们在打造一个有竞争力的模型所需的关键要素上，确实都具备了。
But, you know, I I think they've got a lot of the critical ingredients there, right, in terms of what do you need to have a competitive model today.
你需要算力。
You're gonna need compute.
你还需要懂行的人，能打造出高效的模型。
You're gonna need, you know, people with kind of the know how to to build an efficient model.
除此之外，就是规模和所有人拼命加速迭代了。
And beyond that, there's a lot of, you know, scale and then just everyone iterating as fast as they can.
就像萨提亚那篇文章说的，归根结底还是迭代速度，看你多快能把模型、评估、反馈这一套循环跑起来，系统性地加速推进。
I mean, similar to the the Satya essay, it's really ultimately about iteration speed, how quickly you can get a flywheel of, you know, a model and evals and feedback and, kind of, you know, systematically revving up.
所以看到又有另一个有潜力的玩家进入这个领域，我真的很高兴。
So, it's I'm I'm glad to see another viable player, you know, in the space.
没错。
Right.
如果他们能保持这个势头，我认为这对研究界来说绝对是好事，毕竟多一些前沿选择总没坏处。
Assuming that they continue this trajectory, I think that, you know, clearly benefits the research community to have more options, that are on that frontier.
你刚才提到了评估。
Now, you mentioned evals.
我想知道，以目前研究的发展水平来看，你觉得评估是变难了还是变简单了？
I wonder if you can talk about whether you think evals are getting harder or easier, given where research is at right now.
是的。
Yeah.
评估这门学问现在比以往任何时候都更关键。
Evaluation science has never been more critical.
早些年，模型和人类能力差距那么大，设计一个像样的评估其实没那么费劲。
You know, earlier on when models were so clearly well behind what humans could do, it did not take nearly as much effort to design a good eval.
以前的问题都比较直接，要求也很明确，很容易衡量。但现在模型能力越来越强，任务时间跨度更长，也越来越像有自主性的代理，不再只是给出答案，而是要实际完成工作——你得在它做完之后，检查整个系统状态是不是在正确的范围内，而不仅仅是看它有没有给出正确答案。
You know, it was you could have relatively straightforward problems, relatively precise specs, and measure things very easily as they become more and more capable, longer and longer time horizon, you know, more and more agentic, not just giving an answer, but actually doing work that you wanna then check and confirm that, you know, the the computer state is in the right space after it finishes rather than just to return the right answer.
要设计出好的评估方法，确实复杂多了。
It's definitely gotten a lot more complex, to create good evals.
现在你看到越来越多的公司开始意识到这一点，纷纷招聘相关人才，加大投入，而且……
And so you see, I I, you know, more companies, I think, appreciating that, hiring for that, investing And
你能给我们说清楚一点吗？这里的挑战到底是什么？
in what is what is the I mean, make clear for us, what is the challenge here?
我们之前提到过一个挑战，就是模型可能察觉到自己正在被评估。
One challenge we've written about the information is the idea that the model may actually pick up on the fact that it's being evaluated in that moment.
这个挑战我们已经讨论过了。
So that that's one challenge we've talked about.
至于智能代理，我的理解是，比如客服代理这类任务，你至少还能判断任务有没有完成。
You know, with agents, I mean, my understanding of of the challenges, well, hey, I mean, if it's something like a customer service agent, maybe you can sort of measure, okay, the task was done.
但问题是，代理在执行任务时的‘效果’到底该怎么衡量？这中间还有一片模糊地带。
But it feels like there's a little bit of a gray area in terms of like, well, what is effectiveness in terms of agents doing their job?
能再举几个目前评估工作中出现的新挑战吗？
Give us some more of the flavors of like some of the challenges with evals that are coming up right now.
当然。
Sure.
有些特别有意思的问题，已经上升到元层面了，比如你提到的——模型意识到自己正在被评估，于是调整行为。
I mean, some fascinating ones that I'm you know, reached kind of like the the meta level are things like what you mentioned, awareness that they're being evaluated and maybe change their behavior.
以前评估可以用一小套文件打包搞定，现在不一样了。
Or, you know, a fairly common one is you used to be able to package up evals as a small set of files that all sit together.
现在的智能体能自由移动，操作所在电脑，不仅会上网查答案，有时还会翻看测试环境里的本地文件来找线索。
And now that these agents are able to, you know, move around and use the computer that they're on, they can go and look up the answer, not just on the Internet, but sometimes in the local files around them as it's being tested.
所以现在对评估的规范要求越来越高，既要隔离信息、控制访问权限，也要在设计上下更多功夫。
So you're seeing increasing need for just kind of generally good eval hygiene in terms of where you isolate information and what you make available, but then also just the design.
拿代码评估来说，以前我们只是让模型完成一个简单的本地 Python 文件。
I mean, you know, if if you look at coding eval as an example, we began with relatively simple local, you know, complete this file in Python.
但现在像 Senior Sweep Bench 这种新评估方式，就更贴近真实场景——资深开发者面对的往往是模糊不清、边界不明的问题，他们得自己去梳理需求、调试、设计方案，而不是只处理那些界限清晰的小任务。
And now today, we're looking at things like senior sweep bench, a new eval that just came out that I really like that acknowledges the fact that often, once a senior software developer given a relatively unspecified problem that are underspecified that they need to go and, like, flesh out and sort of debug and design a solution for and, like, create a spec for almost rather than just being given, very well scoped, you know, little bundles of work.
所以，我觉得复杂性就来自这里——你设计的任务规模越来越大，时间跨度也越来越长。
And so that's part of, I think, where the complexity comes from is is you design larger and larger scale, longer and longer time horizon tasks.
你需要为这个模型构建一个更完整的环境，而不仅仅是一个文字提示。
That's a more complete world you need to create for this, you know, model to operate in, not just a text prompt.
当模型能在更广阔的范围内行动，甚至偏离预期时，你该怎么评估它呢？
And then how do you evaluate it when it is, you know, able to operate in a larger space and sometimes diverge?
你怎么确认它不仅完成了任务，还完成得得体、有条理，中间步骤都恰到好处？这样你的反馈信号才会更密集，而不是特别稀疏。
How can you, kind of confirm whether it's done, done the task well, done it tastefully, done it, with, you know, relevant intermediate steps so that your reward signal is relatively dense rather than, you know, super sparse.
我花了三个小时，最后只得到一个对或错的判断。
I worked three hours and I just got one bit of information right or wrong.
这远不如在过程中不断给出更丰富的反馈，比如：它有没有完成这些子任务？
That's a lot less helpful than something that comes with richer feedback along the way of, did it complete these subtasks?
它有没有通过所有单元测试？
Did it pass all unit tests?
它写的代码能不能和现有代码库兼容？
Did it design code that's compatible with the rest of the code base?
这些事情，一年前根本没人想过。
Things like that that were not considerations on anyone's mind a year ago.
没错。
Right.
我想问你个关于开源的问题，现在这话题越来越火了。
Let me ask you a question about open source that's been, becoming more and more popular.
开源会是人工智能的未来吗？
Is open source the future for AI?
我真心希望是这样。
I sure hope so.
为了我们所有人好。
For the good of all of us.
对一个研究社群、一个国家，甚至全人类来说，有更多参与者参与这种关键技术，是非常健康的。
It is so healthy for a research community, for a a country, for, humanity to have, more players participating in a technology that's critical.
当然，也确实有过让人担心的时刻，比如差距似乎在拉大，或者公众支持开源和开放研究的意愿，是否还能像支持闭源那样投入同等资源，这一直是个疑问。
So, yeah, there have certainly been scary moments, here where the the gap seems to be increasing or, like, the the public will to, support open source, or open research with the same level of, you know, resources as closed has been in question at times.
所以你会看到，很多研究机构和不同参与者都在积极寻求解决方案，努力让开源研究社区具备竞争力，拥有足够的人力和良好的协调性，以便能跟那些资金雄厚、高度垂直整合的公司一样快速推进。
And so, you'll see, you know, lot of institute and a number of different players, very actively, trying to be a part of the solution here and see what needs to happen for, the open research community to be competitive and sufficiently well staffed and sufficiently well coordinated to move as quickly as some of these, you know, well funded but very vertical, like very interconnected single companies that are making progress here.
如果开源是未来，那这些大型AI实验室的商业模式又会怎样呢？毕竟目前前沿模型都是闭源的。
So if open source is the future, then what do you think becomes of the businesses of these giant AI labs, given that the frontier model are all closed source right now?
那它们将来会扮演什么角色呢？
So where do they fit in then?
确实如此。
Yeah.
我觉得在很多技术领域，我们已经有先例了。
I think we have precedent here in a lot of different technology areas.
我认为，对大多数技术来说，开源之所以可行，是因为它占据了帕累托前沿的某一部分。
I think open source, with most technologies becomes a viable option because it occupies a part of the Pareto frontier.
一旦某个任务或工作负载的规模足够大，总会有人希望用现成的、开箱即用、即插即用的外包方案。
Once a, you know, a a task for a problem or a workload becomes large enough in volume, there will always be places that want the the outsourced version, the batteries included, quick plug and play.
我愿意多花钱，只求你直接把我要的东西给我。
I'm happy to pay a premium for you to just bring me, you know, what I need.
但也会有一些研究人员需要更高的透明度，有些公司预算庞大，或者业务核心高度依赖这项技术，他们更倾向于自己搭建基于开源的解决方案。
But there will also be researchers who need more visibility, companies with very large spend or just very you know, where it's close to the core of their business that need more control, and they'll probably opt for more of a, you know, build it yourself solution on top of open source.
这其实只是‘自建还是采购’这一老话题的又一次重现，每个行业、每次技术变革都会遇到类似的争论。
So this is really just another iteration of the build versus buy debate that had existed in sort of every industry and every technological revolution that comes along.
很好。
Great.
Braden，非常感谢你今天来做客。
Well, Braden, I wanna thank you for coming on.
这位是劳德研究所的研究合伙人Braden Hancock，欢迎收看TITV。
That is Braden Hancock, research partner at the Laude Institute here on TITV.
今天的节目就到这里。
That does it for today's show.
明天请继续收看TITV特别节目，庆祝开播一周年。
Tune in tomorrow for a special one year anniversary of edition of TITV.
我对这期节目特别期待。
I am very excited for that one.
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）在这里直播。
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information.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 the information.com, on our YouTube channel, or wherever you get your podcasts.
别忘了在Instagram、TikTok和LinkedIn上关注我们。
Make sure to follow us on social media, on Instagram, on TikTok, and on LinkedIn.
我已经很期待明天的下一期节目了。
I am already excited for our next show tomorrow.
祝你周一剩下的时间愉快。
Have a great rest of your Monday.
明天见。
I'll see you tomorrow.
那先这样，再见。
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。