← 返回任务列表

Big Tech Unites for Open Source AI—and Against Anthropic

331 段 · 1 位说话人 · 原片 30:50
F1
F10:00

今天这期 AI Daily Brief,我们要聊一个大型科技联盟公开力挺 Open Weights 模型。

Today on the AI Daily Brief, a big tech coalition throws its weight behind Open Weights models.

在那之前,先看头条新闻,NVIDIA 也公开力挺 Ilya Sutskever 的 Safe Superintelligence。

And before that in the headlines, NVIDIA also throws its weight behind Ilya Sutskever's Safe Superintelligence.

AI Daily Brief 是一档每日 podcast 和视频节目,关注 AI 领域最重要的新闻和讨论。

The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

好,朋友们,在我们开始之前,先快速说几个通知。

Alright friends, quick announcements before we dive in.
F1
F11:15

欢迎回到 AI Daily Brief 的 Headlines Edition,用大概五分钟带你看完每天需要知道的 AI 新闻。

Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around 5 minutes.

我们今天先从 NVIDIA 对 Safe Superintelligence 进行了一笔相当可观的投资说起。这家有点神秘的 startup,来自前 OpenAI 首席科学家 Ilya Sutskever。

We kick off today with NVIDIA making a substantial investment in Safe Superintelligence, the enigmatic startup from former OpenAI chief scientist Ilya Sutskever.

这个名字,这些年在节目里其实还真不太常提到,至少没提到太多。

Now, that's a name that hasn't come up on this show, at least very much, in years.

稍微往回说一点,给那些没有听我们这么久的朋友补个背景。Ilya 是二零二三年末 OpenAI 领导层风波中的核心人物之一。

Going back a bit for those who haven't been here for as long, Ilya was one of the central figures in the OpenAI leadership dispute in late 2023.

一开始,他是推动 Sam 下台的主要力量之一,后来在 Sam 重新回到 CEO 位置这件事上,他又站了回来。

First, he was a major driver in deposing Sam, then he came back on board with Sam coming back to the CEO role.

整个过程都非常混乱。

It was all very confused.

等到尘埃落定之后,他最终离开了公司,CTO Mira Murati 也一起离开了,两个人各自出去创办了自己的 AI lab。

When the dust had settled, he ended up leaving the company along with CTO Mira Murati, both spinning off to found their own AI labs.

在那段时间里,一个相对稳妥的判断就是,Ilya 大概率会在这个领域做出一些有意思而且新颖的东西。

As that was happening, the safe bet was on Ilya building something interesting and novel in the space.

他是一位在 AI 领域深耕了几十年的科学家,师从 Geoffrey Hinton 完成 PhD,也参与开创了 OpenAI 早期的很多工作。

He was the scientist with the decades-long history in AI, completing his PhD under Geoffrey Hinton and pioneering the early work at OpenAI.

与此同时,Murati 在融资的时候,一些投资人还公开表示不屑一顾。但差不多两年之后,Murati 的 Thinking Machines lab 已经招募了一支全明星级别的 AI 研究团队,而且正处在一个新趋势的中心:帮助公司去思考怎么微调它们自己的模型。

Meanwhile, some investors openly scoffed during Murati's fundraising campaign, and yet almost 2 years later, Murati's Thinking Machines lab has hired an all-star team of AI researchers and is at the center of this new trend of helping companies think about fine-tuning their own models.

而这一切发生的同时,我们几乎一直没听到 Safe Superintelligence 到底在做什么。

All of that happening while we haven't heard anything about what Safe Superintelligence is up to.

不过话说回来,这家公司最初的 mission statement 本身就暗示了要保持神秘。

But then again, the actual mission statement of the company implied secrecy from the start.

Thinking Machines 的目标更务实一些,是要构建一个未来,让每个人都能获得相关知识和工具,让 AI 为自己独特的需求和目标服务——这是他们自己的说法。相比之下,SSI 的目标要窄得多,Ilya 提出的是一个直奔主题的研究机构,要产出 safe superintelligence。

While Thinking Machines had a practical goal of building a future where everyone has access to the knowledge and tools to make AI work for their unique needs and goals—that's their phrase—SSI's goal was much more narrow, with Ilya proposing a straight-shot research organization to produce safe superintelligence.

事实上,从一开始 SSI 很明确的一点就是,Ilya 的立场是,整场游戏的核心就是 superintelligence,而任何你在通往

Indeed, what has been clear from the beginning with SSI is that Ilya was taking the position that the whole game is superintelligence, and that anything that you have to do for short-term commercial purposes on the way

superintelligence 的路上,出于短期商业目的不得不去做的事,归根到底都只是干扰项。

to superintelligence is ultimately just a distraction.

所以,没错,其他公司确实凭借数十亿美元的企业销售额,在私募市场一路冲到了上万亿美元的估值;但 Ilya 的判断是,那些东西其实反而可能会让人偏离最终的使命,至少偏离他心中的最终使命。

So while yes, other companies have raced off to trillion-dollar valuations in the private markets based on billions in enterprise sales, Ilya's bet is that that is actually potentially a distraction from what is the ultimate mission, or at least what his ultimate mission is.

至于这次 Nvidia 和 SSI 新交易本身的实际消息,其实非常简短。

Now, the actual news of this new deal between Nvidia and SSI is very brief.

在一份联合新闻稿里,Nvidia 说他们对 Safe Superintelligence 做了一笔相当可观的投资,但没有透露具体金额。

In a joint press release, Nvidia said that they had made a substantial investment in Safe Superintelligence without giving a number.

新闻稿还说,SSI 将获得 Nvidia 下一代 Vera Rubin 芯片的使用权限。

The release also stated that SSI would receive access to Nvidia's next-generation Vera Rubin chips.

SSI 表示,这笔合作会让他们在接下来十二个月里的算力资源提升到原来的十倍。

SSI said that the deal would 10x their compute resources over the next 12 months.

据背景消息人士透露,SSI 之前主要依赖的是 Google 的 TPU;而 Nvidia 则表示,他们是在获得了难得的机会,接触到 SSI 的技术之后,才决定进行这笔投资的。

In the background, sources said that SSI had previously been relying on Google's TPUs, and Nvidia said that they made the investment after gaining rare access to SSI's technology.

Jensen Huang 评论说,Ilya 在现代 AI 基础层面推动了许多开创性的突破。

Jensen Huang commented, Ilya has pioneered fundamental breakthroughs at the foundation of modern AI.

我们很期待看到 SSI 在 Vera Rubin 平台的支持下,还会发现哪些新的突破。

We're excited to see what new breakthroughs SSI will discover powered by our Vera Rubin platform.

目前 SSI 在这个领域里依然有非常强的存在感,而这次公告也说明这家公司正在取得进展。

Now SSI still looms very large in the space, and the announcement suggests the company is making progress.

SSI 表示,我们已经到了一个阶段:我们的研究值得被扩大规模了,而通过这次合作,我们将能够做到这一点。

SSI commented, we reached the point where our research is worth scaling, and with this partnership we will be able to.

Ilya 本人也补充了一句:是时候把 SSI 的规模拉起来了。

Ilya himself chimed in, adding, time to scale that SSI.

SSI 联合创始人 Daniel Levy 还补充说,深度学习这件事,往往就是一支很精干的小团队在操控一台大电脑。

SSI co-founder Daniel Levy added, deep learning happens when a small crack team operates a big computer.

而现在,这台电脑变得更大了。

The computer just got bigger.

现在再说另一个跟 Nvidia 有关的故事。

Now one other Nvidia story.

接着之前他们要为 OpenAI 的一个设施做兜底支持的消息来看,这种 AI 融资里的循环叙事,现在又强势回来了。

Following up from the news that they were going to backstop an OpenAI facility, the circular AI financing narrative is back with a vengeance.

周一,Nvidia 股价下跌了百分之四,因为有人指出,最新这七亿五千万美元的交易,对一些人来说,实在是有点过于循环了,让他们不太买账。

Nvidia stock was down 4% on Monday as people pointed to a new $750 million in deals that are, for some, just a little too circular for their tastes.

这两笔交易包括和 SK Group 的一项合作,双方会彼此开展超过五亿美元的业务;另外还有那笔两亿五千万美元的兜底协议,帮助 OpenAI 建设一个新的数据中心。

The 2 deals include a partnership with SK Group, which will see the companies doing more than $500 million in business with each other, as well as that $250 million backstopping deal to help OpenAI construct a new data center.

实际上,去年年底到处流传的那张循环交易关系图,我们甚至都看到更新版了,所以接下来几周,估计你还会继续看到这个老话题卷土重来。

Indeed, we even got an update of that circular deal chart that was flying around all of late last year, so expect to see that little chestnut come back over the coming weeks.

不过,说到芯片,Nvidia 依然是独一档的存在,这大概也是为什么中国正这么努力地想改变这个局面。

Still, when it comes to chips, Nvidia remains in a class of its own, and that is perhaps why China is working so hard to change the equation.

《The Wall Street Journal》报道说,中共的相关人员正在推动采用中国本土技术。

The Wall Street Journal reported that CCP operatives are pushing the adoption of domestic Chinese technology.

去年夏天,Huawei 举办了一场闭门会议,演示了他们最新一代的芯片。

Last summer, Huawei held a closed-door meeting to give a demo of their latest generation of chips.

据说,这显示出他们在追赶 Nvidia 方面取得了相当惊人的进展。

showing apparently dramatic progress in catching up to Nvidia.

这已经足以让官方更有底气,采取更激进的做法了。

That was enough to give the party the confidence to get more aggressive.

根据《The Wall Street Journal》的说法,中国的 AI 沙皇向中国最大的 AI 用户发出了一个很直接的警告:谁要是抗拒使用国产芯片,谁就是叛徒。

According to The Wall Street Journal, China's AI czar delivered a blunt warning to China's largest AI users: anyone who resisted employing domestic chips was a traitor.

而现在,这种转变已经相当明显了。

Now the shift has been fairly dramatic.

二零二一年,在美国出口管制之前,中国有百分之九十依赖美国芯片。

In 2021, prior to U.S. export controls, China was 90% dependent on U.S. chips.

到了二零二五年,这个数字降到了百分之六十,而 Morgan Stanley 认为,未来五年里,这个比例可能会进一步降到最低百分之二十五。

By 2025, that number fell to 60%, and Morgan Stanley believes it could fall to as low as 25% over the next five years.

Huawei 副董事长 Eric Xu 对原因说得非常直白。

Huawei Deputy Chairman Eric Xu was very clear on the cause.

他评论说:“如果不是美国把我们的国家、我们的公司,还有我们的行业逼到墙角,我们根本不会去做这样的事情。” 当然要说明的是,虽然他们确实取得了进展,但 Huawei 的芯片和 Nvidia 相比,依然明显落后;主要原因是,西方那边的进步也同样在加速。

Commenting, "If the U.S. hadn't forced our country, our company, and our industry into a corner, we never would have done something like this." Now to be clear, even though they've made progress, Huawei chips are still lagging significantly behind Nvidia, and that's primarily because progress has only accelerated in the West as well.

Brookings 的研究员 Kyle Chan 说,这就像是在追一列子弹头列车。

Kyle Chan, a fellow at Brookings, said it's like trying to catch up with a bullet train.

即便中国整个科技生态都在集中力量解决这个问题,这依然是一项非常艰巨的挑战。

Even with China's tech ecosystem focused on the problem, it's still a daunting challenge.

而其中最大的问题之一,单纯就是产能。

And one of the biggest issues is simply throughput.

中国一直拿不到 ASML 生产的芯片制造设备。

China has been denied access to chip fabs made by ASML.

ASML 是台湾和西方的垄断供应商,也是唯一一家能够制造最尖端晶圆厂设备的公司。

the monopoly supplier for Taiwan and the West and the only company able to manufacture cutting-edge fabs.

目前,中国整个供应链每个月大概能生产两万八千颗先进逻辑芯片,只是 TSMC 产量的一小部分。

China's entire supply chain can produce around 28,000 advanced logic chips per month currently, a fraction of the output of TSMC.

不过 according to Bernstein,这个数字在接下来的三年里,预计每年都会翻一倍。

Still, that number is forecast to double each year for the next 3 years, according to Bernstein.

单是 Huawei 一家,就预计今年要出货一百五十万颗 AI 芯片,大约是他们二零二五年总量的两倍。

Huawei alone expects to ship 1.5 million AI chips this year, roughly double their total for 2025.

不过,Bernstein 预计,在芯片竞赛中,中国至少到二零三零年之前,仍会继续落后于美国。

Still, Bernstein expects that in the chip race, China will continue to lag behind the U.S. through at least 2030.

而且,技术追赶本身也正在快速发生。

The technological catch-up is happening quickly as well.

周一,The Information 报道说,一家总部在上海的公司已经开始量产深紫外光刻机,这是先进晶圆厂里的一个关键组成部分。

On Monday, The Information reported that a Shanghai-based company has begun mass-producing deep ultraviolet lithography machines, a key component in advanced fabs.

这是中国第一次接触到这项技术,而这项技术在此前一直因为那些出口管制而被挡在门外。

This is the first time that China has had access to this technology that was previously kept from them under those export controls.

总的来说,这些报道都凸显出,中国的技术底座已经在多大程度上被动员起来,围绕着这个想法在推进。

Overall, all of this reporting highlights the extent to which China's technology base has been mobilized around this idea.

在过去这些年围绕出口管制的争论里,大体上一直有两个阵营:一派认为,美国的利益最好通过限制中国获取 chips 来实现,希望借此拖慢他们在 AI models 上的进展;另一派则认为,更好的策略是让 Nvidia chips 大量进入中国,占住整个 tech stack,让中国继续依赖美国供应。

Throughout the export control debates over recent years, there have broadly been 2 camps: those who think the U.S. interest is best served by restricting China's access to chips in hopes of slowing down their progress on AI models, and those who believe that a better strategy would be to flood the country with Nvidia chips, owning the tech stack and keeping China dependent on U.S. supply.

现在很明显,这事儿已经翻篇了。

It's clear now that that ship has sailed.

中国正在发展他们自己的 chip 经济,而且速度非常快。

China is developing their own chip economy, and they're doing so at a very rapid pace.

与此同时,Apple 正在和 Micron 就中国 chips 展开一场 lobbying 大战。

Apple, meanwhile, is duking it out with Micron in a lobbying battle for Chinese chips.

上个月有报道称,Apple 一直在游说白宫,希望获准从中国制造商那里采购 memory chips,其中就包括 CXMT。

Last month, reports stated that Apple was lobbying the White House for approval to purchase memory chips from Chinese manufacturers, including CXMT.

他们的理由是,memory 的价格已经完全失控了,除非政府愿意看到卖五千美元的 iPhone,否则就必须找到解决办法。

Their argument was that memory prices were completely out of control, and unless the administration wanted to see $5,000 iPhones, a solution needed to be found.

现在,The Wall Street Journal 报道说,美国 memory 制造商 Micron 也加入了这场争论,正在游说白宫维持现有的限制。

Now, The Wall Street Journal reports that U.S. memory maker Micron has waded into the debate, lobbying the White House to keep restrictions in place.

Micron 是美国唯一一家有相当规模的 memory 制造商,其他大型生产商的总部都在 Korea。

Micron is the only sizable U.S. manufacturer of memory, with the rest of the large producers headquartered in Korea.

据报道,Sanjay Mehrotra 警告说,如果允许中国制造商进入美国市场,可能会重创本土产业,让它走上和美国 steel 业以及制造业相同的路。

Sanjay Mehrotra has reportedly warned that allowing Chinese manufacturers to enter the U.S. could decimate the local industry, courting the same fate as the U.S. steel and manufacturing industry.

当然,从技术上讲,Apple 并不需要白宫批准才能拿到中国 chips。

Now, technically, Apple doesn't need White House approval to access Chinese chips.

目前唯一的限制,是 Pentagon 的 entity list。

The only restriction at the moment is the Pentagon entity list.

而这个名单限制的其实只是军方供应链。

Which only limits the military supply chain.

不过话又说回来,在这个高度紧张的时刻,Apple 也不想绕过白宫。

Then again, Apple does not want to go around the White House during this highly tense moment.

不过,Journal 也指出了 Trump 政府面临的一个两难局面:他们现在正面对不断上升的通胀问题,而 memory 价格飙升又在边际上进一步造成了压力。

Still, the Journal notes the conundrum for the Trump admin: they are currently facing down a rising inflation problem, with the spike in memory prices hurting on the margins.

与此同时,这届政府也已经明确把支持和推动本土 chip 制造列为优先事项。

The administration has also made it a clear priority to support and promote local chipmaking.

如果允许中国 chips 进入无处不在的 iPhone,可能会削弱这个政策方向。

Allowing Chinese chips for the ubiquitous iPhone could undermine that agenda.

Micron 给出的解决方案,是提供更多支持,加快建设更多国内 chip fabs。

Micron's solution is further assistance to fast-track the construction of more domestic chip fabs.

除此之外,Apple 和 Micron 其实都很努力地想要获得总统的好感。

And beyond that, both Apple and Micron have worked hard to get in the president's good graces.

一位白宫发言人并没有对这一情况给出更多说明,只是表示,政府将会——原话是——在保护国家安全的同时,为美国人民争取投资和经济纾困。

A White House spokesperson shed no light on the situation, stating that the administration would pursue, quote, investments and economic relief for the American people while safeguarding our national security.

不过,在上周的一场国会听证会上,白宫科学与技术办公室主任 Michael Kratsios 对议员们表示,最重要的事情是,我们必须在美国本土建立产能。

Still, during a congressional hearing last week, White House Director of Science and Technology Michael Kratsios told lawmakers, the most important thing is we have to build capacity here.

我们显然不想和任何被列在 entity list 上的人做生意。

We obviously don't want to be doing business with anyone that's on the entity list.

Evan Kurtzall 认为,这在可预见的未来就是这个世界的运作方式。他写道,Apple 想把 CXMT 的零部件放进自己的供应链,而 Micron 则在游说把它们挡在门外,双方都把这包装成国家安全问题。

Evan Kurtzall thinks this is just the way of the world for the foreseeable future, writing, Apple wants CXMT parts in its supply chain, and Micron is lobbying to keep them out, and both are dressing it up as national security.

未来几年的 chip 政策大部分恐怕都会是这样:美国公司在 Washington 为采购来源彼此争斗,而地缘政治则是在后面再附加上去的。

Chip policy for the next few years is going to be mostly this: American companies fighting each other in Washington over sourcing, with the geopolitics attached afterward.

相关领域里还爆发了一场大冲突,所以接下来我们就先收掉这些头条,进入今天的主节目。

There was a big fight in some related parts of the world, so with that we will close the headlines and move on into our main episode.
F1
F112:30

欢迎回到 AI Daily Brief!

Welcome back to the AI Daily Brief!

哎呀,这场 open 和 closed AI 之争真是已经烧到一个新高点了。

Boy oh boy has the open versus closed AI debate reached a new fever pitch.

这里面有太多东西值得拆开来说了。

There is so much to unpack here.

这个话题现在完全主导了 AI 圈内部人士和 AI 开发者之间的讨论。

This is a topic that is completely dominating the insider AI and AI builder conversation.

所以我们来看看究竟发生了什么,也试着把这一切理清楚。

So let's figure out what's going on and try to make sense of it all.

NVIDIA CEO Jensen Huang 在 Twitter/X 上发的第一条帖子,就拿到了六千二百万次观看。他那封解释为什么开放模型很重要的公开信,很明显说明这场讨论正在变得多么重要。

With 62 million views on his first post ever on Twitter/X, NVIDIA CEO Jensen Huang's open letter about why open models matter clearly shows just how significant this conversation is getting.

不过我们先退一步,给那些没有像我们中有些人一样近乎疯狂、甚至可以说病态地密切关注这件事的人,梳理一下我们到底是怎么走到今天这一步的。

Taking a step back though, let's try to map how we got here for anyone who hasn't been paying insanely, perhaps terminally close attention as some of us have.

在 Washington,围绕 Chinese models 的担忧,几个月来一直在不断升温。

In Washington, the concern over Chinese models has been building for months.

GLM 5.2 的发布——当然,它正好是在 GPT-5 离线的时候发生的——还带来了一些很夸张的标题,说这个模型和 Claude 4 一样强。

The release of GLM 5.2, which of course happened while GPT-5 was offline, came with some dramatic headlines claiming the model was just as powerful as Claude 4.

比如那个一向被吹捧为很有新闻操守的机构 Forbes,就写出了这么一句神作:大家系好安全带,现在坏人也有了一个和 Claude 4 一样强大的模型。

Take this gem from that vaunted institution of journalistic integrity Forbes: Buckle up, the bad guys now have a model as powerful as Claude 4.

我也不好意思说另一位 AI 播客主持人的坏话,但如果我哪天也发出这种标题,我希望你们立刻退订,越快越好。

Far be it from me to speak ill of another AI podcast host, but if I ever drop a headline like that, I hope that you will unsubscribe as soon as humanly possible.

不管怎么说,虽然对真正用过这个模型的人来说,GLM 5.2 显然并没有强到能和 Claude 4 相提并论,但对 Washington 的政策制定者来说,它已经够用了。

In any case, while GLM 5.2 clearly wasn't as powerful as Claude 4 for anyone who actually used the model, it was good enough for policymakers in Washington.

没过多久,我们就第一次听到一些风声,说 Trump administration 正在考虑一项新的 executive order,要禁止 open-source models,至少是 Chinese 那一类。

Shortly afterwards, we got the first rumblings that the Trump administration was considering a new executive order that would ban open-source models, at least of the Chinese variety.

然后当然,Kimi K2 就发布了,这也确实让这些担忧看起来更有道理了。

Then, of course, we had the release of Kimi K2, and it certainly made these concerns seem more justified.

K2 是当时发布过的规模最大的 open-source model,参数量达到了二点八万亿。

K2 was the largest open-source model ever released, weighing in at 2.8 trillion parameters.

它在 Arena 的 Frontier Code Arena 上,以相当明显的优势击败了 Fable。

It beat Fable on Arena's Frontier Code Arena by a fairly wide margin.

当然,这也并不意味着它在整体上就已经接近 Fable,但对 Washington 里那些非技术背景的政策制定者来说,这种事已经足以让他们从私下担心,直接变成全面恐慌了。

And of course, while that also doesn't mean it's anywhere close to Fable across the board, this sort of thing was enough for non-technical policymakers in Washington to move from quietly concerned to a full-on freakout.

我们还看到了 Treasury Secretary Scott Bessent 和高级技术官员 Michael Kratsios 的公开表态,提前释放出要禁止 Chinese open-source models 的信号。

We got public comments from Treasury Secretary Scott Bessent and senior tech official Michael Kratsios foreshadowing a ban on open-source Chinese models.

这里很重要的一点是,这些表态谈的并不是模型的原始性能。

Now importantly, these comments did not discuss raw model performance.

相反,官员们说,这项政策的目的是阻止 model distillation,而且还可能会对已知曾获取过 Nvidia chips 的 Chinese companies 实施制裁。

Instead, officials said the policy was intended to stop model distillation and could see sanctions applied to Chinese companies known to have accessed Nvidia chips.

这种说法里一个很典型的例子,来自 Treasury Secretary Scott Bessent。他在推特上写道,我们支持 open-source AI,以及它所释放出来的创新。

A canonical example of this sort of rhetoric came from Treasury Secretary Scott Bessent, who tweeted, we support open-source AI and the innovation it unlocks.

但 open source 不代表可以随意掠夺 American IP。

But open source is not open season on American IP.

当 PRC firms 进行秘密的、工业规模的 distillation attacks,越界构成 IP theft 的时候,制裁和 entity list designation 都会被摆上桌面。

When PRC firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and entity list designations will be on the table.

而且这不只是 White House 那边的声音。

And it wasn't just from the White House.

在 Congress,我们开始看到一连串法案被提出来,同时也出现了很多抱怨,说政府在对抗 Chinese AI 这件事上做得还不够。

In Congress, we began to see a flurry of bills being put forward and complaints that the administration wasn't doing enough to combat Chinese AI.

在这一切满天飞的时候,Commerce Secretary Howard Lutnick 成了政府里看起来唯一一个还在主张,应该用竞争来应对 Chinese AI 的官员。

With all of this flying around, Commerce Secretary Howard Lutnick emerged as the sole administration official seemingly still advocating for competition as the right way to deal with Chinese AI.

有传言说,他一直在推动政府支持 U.S. open source,而且据报道,最近几周他还跟多家 labs 见了面。

He was rumored to be pushing for the administration to support U.S. open source and was reportedly taking meetings with multiple labs in recent weeks.

而在上周晚些时候,他的这套主张又得到了一个重大助力,因为有消息披露,Kimi K3 在关键的 cybersecurity benchmarks 上明显落后于 Mythos。

And late last week, his agenda got a major shot in the arm when it was revealed that Kimi K3 lagged significantly behind Mythos on crucial cybersecurity benchmarks.

就在这一切发生的同时,业界也出现了一些比较温和的反对声音。一个叫 Little Tech Association 的行业组织呼吁政府不要禁止 Chinese open source。

Now, as all this was happening, we saw some lightweight pushback from the industry when a trade group called the Little Tech Association urged the administration not to ban Chinese open source.

这个组织代表的是 startup community 里的一部分人。他们认为,这个领域里很多工作都是建立在 Chinese models 之上的,如果真的把它们禁了,我们可能会看到一场 US startups 的大灭绝。

The group, which represents a part of the startup community, argued that a lot of work in the field was building on top of Chinese models, and we could see a mass extinction event for US startups if they were banned.

这个组织里最有名的成员是 Y Combinator。但虽然 Y Combinator 在创业者圈子里地位很高,大家还是很清楚,这个组织恐怕并没有足够的政治分量——这么说吧——去挡下一项带有重大地缘政治影响的政策。

The most prominent member of the group was Y Combinator, but as vaunted as Y Combinator is among startup founders, it seemed pretty clear that this group lacked the political heft, let's say, to block a policy with major geopolitical implications.

然后到了周五,Big Tech 也带着一封获得广泛支持的公开信加入了这场战斗。

Then on Friday, Big Tech joined the fight with a widely supported open letter.

这封信一开头讲的是 open source 的历史,以及 free and open software 如今是如何支撑着世界上每一个关键系统的。

The letter opened by discussing the history of open source and how free and open software now underpins every critical system in the world.

信里一开始写道,在上世纪八十年代,早期的 open source software 先驱挑战了当时流行的看法——也就是只有公司把代码牢牢控制在自己手里,软件才会进步。

In the 1980s, it began, early open source software pioneers challenged the prevailing belief that software would advance only if companies kept tight control over their code.

Open source 带来的不只是降低软件成本。

Open source did more than lower the cost of software.

它建立了一个共享的知识基础,在这个基础上,一代又一代的美国工程师和企业家构建起了他们的制度主权。

It created a shared foundation of knowledge on which generations of American engineers and entrepreneurs built their institutional sovereignty.

这封信接着指出,美国现在在人工智能问题上,也面临着一个类似的选择。

Continuing, the letter argues, The United States now faces a similar choice with artificial intelligence.

我们在 AI 领域的领导地位,评判标准不该是某一个 frontier AI model,而应该看美国是否建立起一个强大、开放,并且能渗透到各个行业的生态系统。

Our AI leadership will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector.

这对于在全国范围内创造创新和繁荣的机会至关重要。

This is essential for creating opportunities for innovation and prosperity across the country.

这需要扩大 AI 的可及性,鼓励竞争,打造强健的 application layer,并且让美国人对他们所依赖的这项技术拥有更大的控制权。

It requires expanding access to AI, encouraging competition, robust application layers, and giving Americans greater control over the technology they rely on.

Open-weight models——也就是任何人都可以下载、检查、修改,并且在自己的基础设施上运行的 AI model——是这个基础中的重要组成部分,因为它们让先进的 AI 更容易获取、更容易适配,也更广泛可用。

Open-weight models—AI models that anyone can download, inspect, modify, and run on their own infrastructure—are an important part of that foundation because they make advanced AI more accessible, adaptable, and widely available.

接下来,这封信进一步梳理了 open-weight models 的好处,指出它们能够让控制权在整个行业里更广泛地扩散,提升竞争,并把 AI 更广泛地分发给所有消费者。

Now, from there, the letter runs through the benefits of open-weight models, noting that they allow diffusion of control across the industry, increased competition, and broad distribution of AI to all consumers.

它当然也谈到了风险,主要是把强大的 AI 提供给 cyberhackers 的风险。

It also does discuss the risks, primarily the risk of making powerful AI available to cyberhackers.

不过,这封信评论说,对这种风险的正确回应,并不是禁止 open weights。

The letter commented, however, the right response to this risk is not to prohibit open weights.

在一个网络安全攻击者会使用先进 AI 的世界里,防御方也需要获得能力相当的模型,这样他们才能检测、模拟并应对新出现的威胁。

In a world where cybersecurity attackers use advanced AI, defenders need access to models with comparable capabilities so they can detect, simulate, and respond to emerging threats.

Open models 能扩大防御能力,提高透明度,也能让许多团队一起发现并修复漏洞。

Open models broaden defensive capability, increase transparency, and allow vulnerabilities to be discovered and remediated across many teams.

事实上,这封信还表示,开放性本身就可能是 AI 安全的重要推动因素,因为它能确保整个生态系统里不存在单点故障。

In fact, the letter stated, openness could be a major contributor to AI safety by ensuring there is no single point of failure in the ecosystem.

这封信还提出了一个也许有争议的立场,试图区分 model distillation 和真正的 IP theft。

Taking a perhaps controversial position, the letter sought to distinguish between model distillation and actual IP theft.

信中写道,distillation,也就是利用一个模型的输出,来帮助训练或改进另一个模型,是一种被广泛使用的技术,用于模型改进、评估和验证。

It stated, distillation, or the practice of using one model's outputs to help train or improve another, is a widely used technique for model improvement, evaluation, and validation.

它体现了一种悠久传统,那就是从已有技术中学习、在其基础上继续构建并加以改进——而这种传统,自 open-source software movement 兴起以来,就一直在推动创新。

It reflects a long tradition of learning from, building upon, and improving existing technologies—a tradition that has helped drive innovation since the rise of the open-source software movement.

相比之下,试图从 closed models 中非法提取参数值的做法,确实会引发合理的担忧。

By contrast, unlawful efforts to extract values from closed models raise legitimate concerns.

这些担忧应该通过有针对性的法律和商业框架来解决,而不是对那些在 AI 创新中扮演重要角色的技术手段施加一刀切的限制。

Those concerns should be addressed through targeted legal and commercial frameworks, rather than sweeping restrictions on techniques that play an important role in AI innovation.

这封信最后总结说,AI 时代可以成为一个繁荣的时代。

The letter concludes, The age of AI can be one of prosperity.

只要做出正确的选择,open-weight AI 就能够扩大机会、强化竞争、延续美国的技术领导力、降低风险,并确保这项非凡技术的好处能在我们的经济体系中被广泛共享。

With the right choices, open-weight AI can expand opportunity, strengthen competition, extend American technological leadership, mitigate risk, and ensure the benefits of this extraordinary technology are shared broadly across our economy.

这样的未来值得去建设,而美国也应该在建设这个未来的过程中发挥领导作用。

That future is worth building, and the United States should lead in building it.

而让这封信不只是一些随意公开表态的原因,是它很快就获得了广泛支持。

Now what made this letter more than just some random open words was the support that it quickly garnered.

就像我在节目一开始提到的,Nvidia CEO Jensen Huang 率先表态,用他有史以来第一条 X 帖子分享了这封信。

As I mentioned at the top of this show, Nvidia CEO Jensen Huang led the way, using his first-ever X post to share the letter.

Huang 还补充说,AI 将改变每一个行业,驱动每一家公司,并且会由每一个国家来构建。

Huang added, AI will transform every industry, power every company, and be built by every country.

Open models 能加强安全和网络安全,加速创新和扩散,并实现主权。

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

这个世界既需要 frontier-closed models,也需要 frontier-open models。

The world needs both frontier-closed models and frontier-open models.

Big Tech 的领导者们一个接一个地加入进来,联名签署了这封信。

One by one, the leaders of Big Tech came on board to co-sign the letter.

Sundar Pichai 转发了 Jensen 的帖子,并补充说,我非常高兴代表 Google 支持这件事。

Sundar Pichai retweeted Jensen, adding, very happy to support this on behalf of Google.

我们长期以来一直受益于 open source,也是 open source 的重要贡献者,事实上,我们也一直通过 Google DeepMind 提供 Gemma 的 open-weight models。

We have long benefited from open source, are big contributors to open source, and in fact have consistently made open-weight models with Gemma available from Google DeepMind.

Satya Nadella 也发文表明 Microsoft 的支持,他写道,open-weight models 是健康 AI 生态系统中不可或缺的一部分。

Satya Nadella signaled Microsoft's support, posting, Open-weight models are essential to a healthy AI ecosystem.

他还表示,我们正与业内其他各方一道,规划一条路径,让 open-weight models 在保护国家安全的同时,增强美国的竞争力,并扩大经济机会。

Together with others across our industry, we are outlining a path for open-weight models to strengthen American competitiveness and expand economic opportunity while protecting national security.

Mark Zuckerberg 转发了 Nadella 的帖子,还评论说,开源是一股积极而且重要的力量,既能赋能人们,也能防止权力过度集中。

Mark Zuckerberg shared Nadella's post, commenting, Open source is a positive and important force for both empowering people and preventing centralization.

很自豪能支持这件事。

Proud to support this.

后来 Elon Musk 转发了 Jensen 的内容,说,这件事我完全支持。

Elon Musk later retweeted Jensen and said, This has my full support.

Jensen 说得对。

Jensen is right.

基本上,这封信得到了几十家大大小小科技公司的支持,他们都有切身利益,要确保开源不会被政府关掉。

Basically, the letter enjoyed support from dozens of companies across both big and little tech, all with a vested interest in making sure open source isn't shut down by the government.

Palantir 的视角挺有意思,他们评论说,二十年前,Palantir 还是一个新进入者。

Palantir had an interesting perspective, commenting, 20 years ago, Palantir was a new entrant.

今天,我们和科技行业里一些最大、也正在崛起的玩家站在一起,反对那种由少数几个封闭的 frontier models 统治一切的未来。

Today, we stand with some of tech's biggest and emergent players to reject a future where a select few closed frontier models rule them all.

主权和繁荣需要美国在一个强大的 open-weight AI 生态里保持领导地位,这种生态会强化竞争,让客户保持掌控权,并把创新带来的好处扩散到整个经济。

Sovereignty and prosperity require American leadership in a strong open-weight AI ecosystem that strengthens competition, keeps customers in control, and drives the benefit of ingenuity across the economy.

而且说实话,这场讨论里有一种声音,纯粹就是特别激动、特别兴奋。

And honestly, one strand of the discourse was just gushing and excited.

Not Boring 的 Packy McCormick 写道,资本主义真是太美了。

Not Boring's Packy McCormick wrote, capitalism is just so beautiful.

市值加起来八万亿美元的公司站出来支持 open-weight models,因为这对他们的生意有好处,也让他们脸上有光,而且完全只是巧合的是,这对我们消费者还有几乎所有企业也更有利,除了那几家实验室。

$8 trillion worth of market cap coming out in support of open-weight models because it's good for their business and makes them look good, and totally coincidentally also happens to be better for us consumers and practically every business except for a couple of labs.

我们等于白白享受到了 Microsoft 和 Nvidia 出于自身利益带来的好处。

We just get that benefit for free with Microsoft and Nvidia's self-interest.

真是让人眼眶一热。

It brings a tear to the eye.

今天是我对 American AI 的未来感觉最乐观的一天。

Today is the most positive I have felt about the future of American AI.

这么多公司能以统一的方式站到一起,对抗 regulatory capture,看着真是不可思议。

The coming together of so many companies in a united manner to fight against regulatory capture is incredible to watch.

说真的,这比今天发生的其他任何事都更重要。

Honestly, this is bigger than anything else going on today.

Opus 5 做 ARC puzzles,跟这件事比起来根本不算什么。

Opus 5 doing ARC puzzles is nothing compared to this.

我不确定科技圈里以前有没有什么事情,能像这条发出来的信息一样,得到这么广泛的支持和一致立场。

Not sure if anything in tech has ever gotten as much broad-based support and alignment as this posted message.

现在关键是,美国应该真正站出来,继续推动 open-weights innovation。

The key now is that America should actually step up and continue to push open-weights innovation.

这对整个市场都有好处,包括那些 frontier labs。

It's good for the entire market, including the frontier labs.

Open weights 能通过为客户的具体需求提供更多选择,加快 AI 在整个经济中的扩散;也能降低某些工作负载的成本;而且你还可以针对一些工作领域去微调模型,因为这些领域未必适合放进一个通用的 horizontal frontier model 里。

Open weights accelerates the diffusion of AI across the economy by providing more options for specific customer requirements, can lower the cost of certain workloads, and you can tune models for areas of work that don't always make sense in a horizontal frontier model.

甚至连 Denny's 都支持了这条信息,还分享了一张维恩图,说他们和 Nvidia 的重叠之处,就是都知道保持开放有多重要。

Hell, Denny's even supported the message, sharing a Venn diagram with the overlapping space between them and Nvidia being about knowing the importance of staying open.

当然,明显缺席的是美国那些 frontier labs,OpenAI 和 Anthropic,至少一开始是这样。

Now, the notable absence was, of course, the frontier US labs OpenAI and Anthropic, at least initially.

没过多久,Sam Altman 就回应了这封信,评论说,我希望美国在 AI 上赢,不管是开源模型还是专有模型,我也很高兴看到这个。

It didn't take long before Sam Altman responded to the letter, commenting, I want the US to win in AI, both in open source and proprietary models, and I am glad to see this.

当时 OpenAI 还不是这封信的联署方,但在收到大量反弹之后,他们确实被加进了联署名单。

At the time, OpenAI was not a signatory to the letter, but after receiving a ton of pushback, they were in fact added to the list of signatories.

这样一来,就只剩下一家了。

And that left one.

前 AI 沙皇 David Sacks 写道,除了 Anthropic 之外,整个科技行业都已经公开表态支持 open source AI。

Former AI czar David Sacks wrote, The entire tech industry, save for Anthropic, has come out in favor of open source AI.

那么,接下来会发生什么呢?

So, what happens next?

Anthropic 会改变它的游说策略吗?

Will Anthropic change its lobbying efforts?

大概率不会。

Not likely.

现在,煤气灯操控开始了。

Now the gaslighting begins.

没人想要禁止开源,我们只是想限制谁能用它。

Nobody is trying to ban open source, we just want to limit who can use it.

我们只是想限制谁能为它做贡献。

We just want to limit who can contribute to it.

我们只是想限制这些模型到底能有多强。

We just want to limit how powerful these models can be.

我们只是想确保我们游说推动的那些护栏措施,没法被移除。

We just want to make sure the guardrails we lobbied for can't be removed.

最后的结果其实都一样。

The net effect will be the same.

他们不把开源彻底废掉,是不会停手的。

They won't stop until they kneecap open source.

行业里其他人都得像盯鹰一样死死盯着这帮家伙。

The rest of the industry needs to watch these guys like a hawk.

因为行程安排的关系,我现在是在周一下午录这个。

Now, because of travel, I'm recording this on Monday afternoon.

等你听到这一段的时候,Anthropic 完全有可能已经改口了,不过我猜这种事大概率不会发生。

It's entirely possible that Anthropic has shifted by the time you're hearing this, but my guess is that that's not going to happen.

坦白说,考虑到 Anthropic 一贯的立场,他们不支持这个,其实也不让人意外。

And frankly, the fact that they don't support this shouldn't be surprising given Anthropic's views.

他们已经公开说过,他们正在跟政府合作,防止 distillation attacks。

They've openly said they are working with the government to prevent distillation attacks.

所以很多人都认为,Anthropic 私下里是在游说推动禁止 open model。

So most assume Anthropic is lobbying for an open model ban behind closed doors.

先把正在进行的那些游说放一边不谈。

Hold aside any lobbying that's going on.

CEO Dario Amodei 已经不止一次谈到 open model 整体带来的风险,尤其是 Chinese model 带来的风险。

CEO Dario Amodei has spoken repeatedly about the risks posed by open models in general and Chinese models in particular.

在六月接受 Bloomberg 采访时,Amodei 说:"我担心的是,如果达到 MItM 级别的网络攻击能力,变成任何人都能下载得到的东西。"

During an interview with Bloomberg in June, Amodei said, "What I worry about is if we get MItM-class cyber abilities available for anyone to download."

我觉得这确实是个很严肃的担忧。

I think it's a serious concern.

所以,这一切到底该怎么理解呢?

So how to make sense of all of this?

嗯,有一种解读,某种程度上也呼应了 Packy 刚才说的,就是这事说到底,不过是大家都在为自己的利益发声,伙计。

Well, one interpretation, which kind of harkens back to what Packy was saying, is that this is just everyone talking their book, man.

Bucco Capital 写道:"我对这件事的看法其实很简单。"

Bucco Capital writes, "My read on this is quite simple."

客观来说,OpenAI 和 Anthropic 正在把其他所有人都打得落花流水。

Objectively, OpenAI and Anthropic are kicking everyone's butts.

你能看到他们收入增长的陡峭曲线。

You can see the revenue ramps.

事实上,这封公开信反而让我觉得,他们的表现比我原来想的还要好,而其他人是真的非常害怕。

In fact, the open letter makes me think they are doing even better than I thought, and everyone else is genuinely really scared.

那些可能会输的人看到了一个机会,或者更糟一点,是看到门正在关上,所以他们觉得自己现在必须行动。

The potential losers see an opening, or alternatively worse, see the door closing, and they feel compelled to act now.

地缘政治、道德和民族主义,大多都只是噪音。

Geopolitics and morality and nationalism are mostly just noise.

我不觉得公司真的在乎那些东西。

I don't think corporations really care about that stuff.

真的不怎么在乎。

Not really.

但他们非常在乎输,而他们现在就在输。

But they really care about losing, and they are losing.

Sacra data 的研究负责人 Steve Hou 转发了 Bucco 的那篇帖子,并补充说:"这个角度挺有意思。"

Sacra data head of research Steve Hou reshared that post from Bucco and added, "Interesting take."

不过问题是,在 open model 和 closed model 的争论里,到底是哪一方在诉诸那些地缘政治、道德和民族主义的论点呢?

The question is, which side of the open versus closed models debate is invoking the geopolitical, moral, and nationalistic arguments, though?

如果最好的开源模型主要都是 Chinese 的,那支持开源模型这件事,看起来就和地缘政治、民族主义方面的担忧是相冲突的。

To the extent that the best open models are predominantly Chinese, it seems that the support of open models runs counter to the geopolitical and nationalistic concerns.

更 broadly 来说,我觉得也有可能,OpenAI 和 Anthropic 靠着更简单、更好的 UX,也因此带来更强采用,已经让 AI 变成了一场他们自己才会输掉的游戏。

More broadly, I think it's possible that OpenAI and Anthropic could be running away with it on the easier and better UX, hence adoption, that AI has become their game to lose.

我对这封公开信的另一个解读是,企业领导人和联署人担心的,与其说是 OpenAI 和 Anthropic 原始的竞争实力,不如说是美国联邦政府的权力,以及基于非经济、地缘政治和 NatSec 理由,对 Chinese 开源模型实施禁令;这样一来,就会在美国本土开源模型还没来得及长出来之前,进一步强化,甚至彻底坐实这两家实验室的市场主导地位。

My other interpretation of the open letter is that rather than being scared of the raw competitive powers of OpenAI and Anthropic, corporate leaders and signatories of the letter are worried about the power of the US federal government and a ban of Chinese open models on non-economic, geopolitical, and NatSec grounds, thereby reinforcing, if not cementing, the market dominance of the two labs before US-native open models have a chance to sprout up.

现在,最近的一次 podcast 露面里,Sam Altman 说了一段话,或许更能说明,OpenAI 参与这件事不只是受公众舆论施压,他们自己的真实信念也确实和这件事有交集。他说:“我觉得当下这场斗争是,我们究竟会走向一个 AI authoritarianism 的世界,还是一个自由的世界?”

Now, perhaps reinforcing the idea that OpenAI wasn't just pressured by public opinion to join this, but that it does intersect with their actual beliefs, in a recent podcast appearance, Sam Altman said, "I think the fight of the current moment is, are we going to head to a world of AI authoritarianism or liberty?"

人类每一次拿自由去交换安全,长期来看都是净损失。

Every time that humanity has traded off its liberty for safety, it's been a long-term net loss.

我们会把这个交到人们手里。

We are going to put this in the hands of people.

我们会赋能他们。

We're going to empower them.

我们会让社会表达自己的想法,并按照他们想要的方式来使用这项技术。

We are going to let society express its ideas and use this technology in the way they want.

Business Insider 从那段对话里提炼出的意思是,Altman 最大的担忧之一,是只有少数几家公司控制 AI。

Business Insider took from that conversation the idea that one of Altman's big fears is a small number of companies controlling AI.

而且确实,对一些开源倡导者来说,这就是一个可以把各种情绪都好好感受一遍的重要时刻。

And indeed, for some open advocates, it was just a big moment to feel all the feels.

Nathan Lambert 写道,今天对开源模型来说是个大日子。

Nathan Lambert wrote, a big day for open models.

我感觉自己终于能稍微放松一点,沉浸一下这个时刻了。

I feel like I can relax a little, soak it in.

一个强大得惊人的联盟站出来说,我们相信,我们在乎,这在任何议题上都很少见。

An incredibly powerful coalition said, we believe, we care, That's rare on any topic.

Mudit Jayasakera 写道,我真的希望,六个月后,当又一次周期性地转向 closed source 的摆动出现时,我们大家都还记得当下这场讨论里的这种能量和这些价值观。

Mudit Jayasakera writes, I really hope that in 6 months, when there is, yes, another cyclical swing towards closed source, we all remember the energy and values that are currently in the discourse.

当两条前沿路线之间的界线这么模糊的时候,大家很容易团结到 open source 这边来。

It's very easy to rally around open source when the line between the 2 frontiers is this blurry.

但我们之所以这么久以来一直支持这个主张,是因为如果要建立一个安全、而且被民主化的智能未来,它就应该由很多人来塑造,由很多人来做 red-team 和解读,也由很多人来提供服务。

But the reason we've rallied behind this thesis for so long is that in order to build this safe and democratized future of intelligence, it should be shaped by many, red-teamed and interpreted by many, and served by many.

是的,我真的相信,那个 AI 更加分散扩散的世界,才是最安全、也最有成果的世界;而且这个世界应该由我们这个领域里更多杰出的人共同参与,而不是只靠那些关在 closed doors 后面的人。

Yes, I really do believe the safest and most fruitful world with diffused AI is the one contributed to by more of the brilliant people in our field than the ones just behind closed doors.

现在,说到这会把 Anthropic 置于什么位置,哎呀,这又是另一个 Rorschach test。

Now, when it comes to where this leaves Anthropic, boy is this another Rorschach test.

看你站在什么角度,他们要么是反派,要么是整件事里最吃到 aura 红利的一方。

Depending on your perspective, they are either the villain or the main recipient of aura from this whole affair.

AI 评论员 Andrew Curran 转发了 David Sacks 的帖子,并说,针对 House of Anthropic 的 crusade 还在继续升温。

AI commentator Andrew Curran reshared David Sacks' post and said, the crusade against the House of Anthropic continues to build.

这件事越来越被塑造成:Anthropic 独自对抗整个行业。

It is increasingly framed as Anthropic standing alone against the entire industry.

对有些人来说,就算他们并不同意,也还是会觉得 Anthropic 这样坚持自己的立场挺酷的。

And for some, even if they didn't agree, said that it was pretty cool that Anthropic was sticking to its guns.

Packy McCormick 又写道,我得说,Anthropic 本来完全可以轻轻松松把信签了,然后再私下去游说,而我尊重他们没有这么做。

Packy McCormick again wrote, I will say it would have been easy for Anthropic to just sign the letter and then lobby behind the scenes, and I respect that they didn't.

他们相信某些东西,而且就一直坚持,哪怕代价不小。

They believe something and they're sticking to it, even when costly.

使用化名的 AI 创业者兼评论员 Signüll 认为,观点的多样性是一件好事。

Pseudonymous AI entrepreneur and commentator Signüll thinks that the diversity of perspective is a good thing.

不管怎么说,他写道,我很喜欢 Anthropic 和很多其他公司有不一样的世界观。

For what it's worth, they wrote, I love that Anthropic has a different worldview than many of the other companies.

为什么我们不想要那种有真实原则、而且会对未来得出不同结论的创始人呢?

Why wouldn't we want founders with genuine principles who arrive at different conclusions about the future?

你不一定要同意 Anthropic 的看法,才能尊重他们是在按照自己相信的东西一以贯之地行动。

You don't have to agree with Anthropic's views to respect that they're acting consistently with what they believe.

真要说的话,整件事反而让他们更有 aura 了,因为他们又一次愿意为了原则站出来。

If anything, this whole episode gave them more aura because they were willing to again stand on principle.

这种情况一次又一次地都是真的。

Time and time again, this has been true.

对这个国家来说,一个健康的生态应该是:那些从根本上持有不同愿景的人,彼此竞争,来证明自己是对的。

A healthy ecosystem for this country is one where people with fundamentally different visions compete to prove themselves right.

Signüll 还补充了最后一点:如果你是政策制定者,或者你是签了那封信的 CEO,那你现在就得准备好接受一种可能性——开放权重模型最终可能会卷入一起严重的安保或者安全事件。

And Signüll added, one last thing, if you're a policymaker or a CEO who signed that letter, you now have to be prepared for the possibility that an open-weight model could eventually be involved in a serious security or safety incident.

如果真发生了,那 Anthropic 的立场看起来就会比它现在很多批评者愿意承认的,更有先见之明。

If that happens, Anthropic's position will look a lot more prescient than many of its critics are willing to admit today.

但这就是坚持原则立场的本质。

But that's the nature of taking a principled stance.

有时候你会错。

Sometimes you're wrong.

但如果后来的事实验证了你的推理,历史往往会更记住这一点。

But if events validate your reasoning, history tends to remember that a lot more.

有点让人意外的是,在“Anthropic 因为这件事在这一边显得更有 aura”这件事上,居然还有一个新加入的人是 OpenAI 的 Roon。他转发了我刚才分享的 Denny's 那条帖子,还加了一句:我觉得,在资本主义里每家公司都抱团起来,乖乖签请愿书去抱怨封闭模型,他们根本没意识到自己把 Anthropic 衬托得有多酷。

One perhaps surprising entrant to the Anthropic is getting aura from this side of the room came from OpenAI's Roon, who retweeted that post I just shared from Denny's and added, I don't think every corporation in capitalism banding together meekly signing petitions complaining against closed models realizes how cool they're making Anthropic look.

又一次开启了无限 aura farming。

Unlimited aura farming once again.

有意思的是,Ethan Mollick 还引用了 Roon 的另一条推文,来说明实验室里很多人想问题的方式,真的跟我们其他人不太一样。

Interestingly, Ethan Mollick pointed to another tweet from Roon to argue that many of the people inside labs just don't think the way the rest of us think.

他引用的那条 Roon 的推文里,Roon 说的是:如果你是那种完全被 AGI 说服的人,同时又支持 open source,那你的信念体系根本不可能自洽——这一点从二零一五年还是哪年开始就已经很明显了。

The tweet he pointed to from Roon was one in which Roon said, there is no way to hold a consistent belief set where you're AGI-pilled and pro-open source, and this has been obvious since 2015 or whatever.

接下来就是一大波自我安慰。

Enormous cope ensues.

这里面的意思当然是:如果你真的完全相信 AGI,那你会对一个完整 AGI 世界会是什么样子感到太害怕,因此你不会希望它是开放的。

The implication, of course, is that if you are fully AGI-pilled, you are too scared of what a full AGI world looks like to want it to be open.

我觉得大多数人在谈开放权重模型的时候,其实并没有真正相信那些实验室内部人士所相信的 AGI 和 ASI 愿景。

I think most people, when talking about open-weights models, don't deeply believe in the vision of AGI and ASI that lab insiders believe in.

比如说,他们并不觉得 AI 真的会带来那种严重的、半自主的生物安全风险,或者其他类似风险。

Like, they don't expect AI to really present grave semi-autonomous biosecurity or other similar risks.

至于到底对不对,现在谁也不知道。

Whether it's right or not, no one knows.

如果你不相信这些风险是真的,那你大概率就会觉得,实验室里那些在讨论这些风险的人,是出于纯粹的商业目的在故意传播 FUD,是在编造风险,好通过监管来维持他们的高利润率。

If you don't believe these risks are real, then you probably believe that the people in the labs discussing them are intentionally spreading FUD for purely commercial reasons and making up risks to keep their high margins via regulation.

我确实觉得,对很多人来说,Anthropic 这种坚持原则的做法是值得称赞的。

Now, I do think that for many people, Anthropic standing on principle will be something to be lauded.

不过,如果情况真是这样,他们可能就不会希望自己的员工在 Twitter 上那么阴阳怪气。

However, to the extent that is the case, they might not want their employees to be snarky on Twitter.

Anthropic 技术团队成员 Julian Schrittwieser 转发了 Jensen 的帖子,还加了一句:我太激动了,没想到 Jensen Huang 现在也成了 open source 的信徒。

Member of the Anthropic technical staff Julian Schrittwieser retweeted Jensen's post and added, I'm so excited that Jensen Huang is a believer in open source now.

很期待 CUDA 和 GPU driver 的 open source 发布。

Looking forward to the CUDA and GPU driver open source release.

然后他又在同一个话题下面转发了 Satya Nadella 的帖子,接着说:还有 Satya Nadella,现在也站出来支持 open source 了。

He then added in the same thread, retweeting Satya Nadella's post, and Satya Nadella coming out for open source too.

真是迫不及待想看到 Windows 和 MS Office 也 open source。

Can't wait for the open sourcing of Windows and MS Office.

大家对这番话的反应,基本上都是相当震惊。

The responses to this were fairly aghast.

David Sacks 当然也立刻下场了,说:没人说所有软件都必须是 open source。

David Sacks, of course, jumped in saying, nobody is saying that all software has to be open source.

他们的意思是,开放权重 AI 应该被允许存在。

What they're saying is that open-weight AI should be allowed.

换句话说,他们是在拒绝你们公司那种没完没了、想把开放模型生态打残的运作和算计。

By implication, they're rejecting your company's incessant machinations to kneecap the open model ecosystem.

很高兴看到 Anthropic 承认了,所谓 open 的真正问题在于,它会跟他们公司的商业经济策略形成竞争。

Excited to see Anthropic acknowledge the real problem with open is it competes with their corporate economic strategy.

这里表现出来的这种认知,其实很说明问题。

The awareness here is revealing.

当你打造出了史上最大的 private market cap 之一,你就应该预期会有竞争。

When you build the biggest private market cap ever, you should expect competition.

我相信那些二级份额交易肯定让你们赚得挺舒服。

I'm sure the secondaries have been nice.

这观点也太离谱了。

Such a bad take.

认为开源应该存在,并不等于认为所有东西都必须开源。

Belief that open source should exist doesn't equate to belief that literally everything should be open source.

而且 Nvidia 还向开源投入了两百亿美元。

And Nvidia invested $20 billion into open source.

真尴尬。

Embarrassing.

换个角度来看这个 aura 的问题,a16z 的 Justine Moore 写道,想象一下,你是那个在给 Anthropic 写 S-1 的投行人士,现在还得把员工发推带来的巨大 aura 损失,加进风险因素那一栏。

Taking the other side of the aura question, a16z's Justine Moore writes, imagine you're the banker writing Anthropic's S-1 and now have to add massive aura loss from employee tweets to the risk section.

总结一下,Jeremy Howard 写道,真是一直让人想不通,为什么那么多平时看起来脑子正常、也有判断力的人,一到思考那些会影响给他们发工资那家公司的行为时,就好像瞬间失去了全部认知能力。

Summing it up, Jeremy Howard writes, it continues to boggle the mind how many people who otherwise seem to have a functioning intellect appear to lose all cognitive capacity when it comes to thinking about actions that impact the company that pays them.

至少,这能让你感受到这场讨论正在变得有多激烈。

If nothing else, it gives you a sense of just how intense this conversation is getting.

如果你只从这里带走一个结论,那就是,不管美国政府接下来想推行什么政策——无论是中立、支持开源,还是反对开源——影响都会非常大。

If you take away anything from this, it is that there are big implications for whatever policy the U.S. government tries to pursue, whether it's neutral, pro-open source, or against open source.

我想说的是,不管你在这场争论里站哪一边,你都会有很多机会表达自己的看法,因为我觉得,至少眼下,这大概就是我们接下来要讨论的最重要的话题之一了。

I will say that wherever you fall in this debate, you are going to have a lot of room to make your opinions known, because I think that for the moment, this is just about the biggest conversation that we're going to be having.

不过现在,今天这期 AI Daily Brief 就到这里。

For now, though, that is going to do it for today's AI Daily Brief.

一如既往,感谢你的收听或收看,我们下次见,拜拜。

Appreciate you listening or watching as always, and until next time, peace.
已剔除 2 处广告(点击展开查看)
F1
F10:24广告 · 已剔除

First of all, thank you to today's sponsors, KPMG, Robots & Pencils, Blitzy, and Airtable.

To get an ad-free version of the show, go to patreon.com/AIDailyBrief, or you can subscribe on Apple Podcasts.

To learn more about sponsoring the show, send us a note at [email protected].

And if you want to not just hear about the news and discussions in AI, but actually dig in and learn AI, come check out our Summer Adventure.

It is the latest free, self-directed, project-based AIDB learning experience built in conjunction with Superintelligent and our superstar Noubar Gaspar, and is a choose-your-own-adventure for you to do everything in AI, from build a personal context file to actually build an agentic operating system, from doing really simple stuff that can still be incredibly impactful all the way to doing advanced agentic loops.

Again, you can find that for free at summeradventure.ai.

Come join more than 1,000 friends on this AI summer adventure.

F1
F19:23广告 · 已剔除

One of the most important AI questions right now isn't who's using AI, it's who's using it well.

KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising.

The highest impact users aren't better prompt engineers, they treat AI like a reasoning partner.

They frame problems, guide thinking, iterate, and push for better answers.

And the good news?

These behaviors are teachable at scale.

If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time.

Learn more at kpmg.com/us/sophisticated.

That's kpmg.com/us/sophisticated.

One thing I keep seeing in enterprise AI—companies hedging across every cloud, every model, every framework, or paying a GSI for a pilot that never ends.

The teams actually shipping, they've picked a lane and they move fast.

That's one of the reasons I like today's sponsor, Robots & Pencils.

They've gone all in on AWS.

They're an Advanced Tier and AWS Partner partner, and they ship production AI coworkers in 45 days.

That's led to them doing some of the more interesting work I've seen on AI coworkers.

And by that, I'm not talking about chatbots.

I'm talking about actual agentic systems that sit inside a business architecture and do real work.

That kind of focus matters if you're an enterprise leader trying to get something real into production or an AWS rep trying to move a customer from interested to deployed.

Request an AI briefing at robotsandpencils.com.

One conversation with Robots & Pencils and you'll know.

Blitzy deeply understands your codebase before it writes code.

Here's the first place that pays off: security in the age of AI.

Vulnerabilities don't live in isolation.

They live buried inside millions of lines of interconnected code, where patching one thing quietly breaks 3 others.

That's why surface-level scans fail.

Blitzy starts from its knowledge graph of your entire application, identifies and surfaces CVEs across the full estate, proactively recommends patches, and can execute the PR.

Each fix is grounded in how your systems connect and validated so nothing new breaks, and the knowledge graph dynamically updates, keeping you ahead of an ever-accelerating threat landscape.

One Blitzy customer resolved 21 active CVEs across 6 core microservices in 4 days.

Zero compile errors, every validation scan clean, months of planned work fixed in less than a week.

Security remediation grounded in real architectural context at the speed of compute.

Harden your codebase at blitzy.com.

That's B-L-I-T-Z-Y dot com.

This episode of the AI Daily Brief is brought to you by HyperAgent, where you run fleets of agents your team can manage together.

New users get $1,000 in inference.

Forget local agents and chat workflows waiting on your laptop to be prompted.

HyperAgent deploys always-on agents in the cloud, doing real work across the tools your team already uses.

Marketing's agent turns competitor moves into landing pages.

Sales Agent enriches leads, drafts emails, and updates the CRM.

Ops Agent chases the paperwork and tracks the budget.

Every agent has access to shared context and follows your rules about scope and approvals.

It's time you add agents that feel like teammates.

Hire yours at HyperAgent, built by the team at Airtable.

Claim your $1,000 in inference at hyperagent.com/AIDailyBrief.