企业目前真正关注的AI挑战

过去几周，社会上关于AI的讨论真是热闹非凡，但在这所有的喧嚣中，有一群人还在琢磨，对他们来说，情况到底有没有实质性的改变，还是说他们只是在走以前的老路，这些人就是那些多年来一直想方设法利用AI为自己创造价值的企业和机构。

It has been a heck of a last couple of weeks when it comes to the AI discussion in society, and yet in all of that, one group that's left trying to figure out if anything has actually changed for them or if they are just on the same path that they were before, is the businesses and enterprises that have been trying to figure out how to maximize AI for their own value for, at this point, a number of years.

今天，我们要讨论的是，企业如何看待这个AI安全的新时代（如果他们确实在考虑的话），同时我们也会更深入地挖掘，看看企业真正担心的是什么，以及他们现在面临的真实AI挑战。

Today, we're discussing both how enterprises are thinking about, if at all, this new era of AI safety, and also digging a little bit deeper to find out what the real concerns that businesses have and the real AI challenges they're facing right now.

和往常一样，我们正处在一个新挑战带来新机遇的时刻。

As always, we're in a moment where new challenges are creating new opportunities as well.

《AI每日简报》是一档每日更新的播客和视频节目，内容涵盖AI领域最重要的新闻和讨论。

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

好了，朋友们，在开始之前，先快速说几个通知。

All right, friends, quick announcements before we dive in.

首先，感谢今天的赞助商：毕马威、Blitzy、Robots & Pencils 和 Hyperagent。

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

想要收听无广告版本，请访问 patreon.com/aidailybrief，或者也可以在Apple Podcasts上订阅。

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

如果想了解如何赞助本节目，请发送邮件至 sponsors@aidailybrief.ai。

And to learn more about sponsoring the show, send us a note at sponsors@aidailybrief.ai.

现在，AI安全话题已经连续两周占据了绝对主导的讨论。

We are now two weeks into the AI safety discourse absolutely dominating the conversation.

对于那些觉得这些问题一直缺乏关注的人来说，这段时间确实是个好消息。

For those who think that awareness of these issues had been sorely lacking, it has been a very good period.

不过，现在看到民调显示大约17%的美国人完全相信AI会终结人类，这确实让人担心我们可能有点反应过度了。

Although now seeing polling numbers that suggest that something like 17% of Americans are completely convinced that AI is going to end humanity, there is certainly some reasonable concern that we might have over-calibrated.

先把这个讨论放一边，我觉得几乎所有人都在期待更多的具体性。

Holding that discussion aside for a moment, one thing that I think almost everyone is looking for is increasing specificity.

也就是说，政策要具体，监控也要具体。Anthropic正是针对这一点，提出了新的测量方案，帮助公众了解先进AI的发展速度到底有多快。

That is specificity of policy, but also specificity of monitoring, and it's to that, that Anthropic speaks with their new proposed set of measurements to help the public understand just how quickly advanced AI is developing.

抛开那些最吓人的头条不谈，这个月的安全辩论主要围绕递归自我改进，以及AI发展不仅速度快，而且即将变得更快这一观点。

Once you move past the scariest headlines, this month's safety debate has been largely about recursive self-improvement, and the idea that AI development is, A, moving too fast, but B, really about to move much more quickly.

社会毁灭的威胁确实能吸引眼球，但发出警告的AI研究人员很难准确描述他们在实验室里看到了什么。

The threat of societal destruction makes for a good headline, but the AI researchers issuing the warnings have had a difficult time describing exactly what they've seen inside the labs.

为了让公众更了解情况，Anthropic提出了一个三方面的测量方法，用来评估当前AI发展的速度。

To better inform the public, Anthropic has proposed a three-access measurement to understand the current pace of AI development.

第一个指标是AI构建下一代自身的能力。

The first access is AI's ability to build the next version of itself.

第二个是Anthropic监督和干预智能体行动的能力，第三个是投入AI模型开发的资源规模。

The second is Anthropic's ability to oversee and intervene in the actions of agents, and the third is the scale of resources that go into AI model development.

Anthropic明确指出，这些指标只关注模型的构建方式。

Now, Anthropic explicitly notes that these measurements only look at how models are built.

换句话说，他们只衡量模型开发的输入，但他们认为这些输入与模型能力的增长相关。

In other words, they are measuring only the inputs to model development, but they argue that these inputs are correlated to the growth in model capabilities.

关于第一个指标，即在未来模型训练中使用AI模型的情况，Anthropic提出了一个名为“研发自动化指数”的衡量标准。

On the first measure, the use of AI models in future model training, Anthropic put together a measure they called the R&D Automation Index.

它旨在衡量模型训练的每个方面，评估每个任务当前的自动化程度，并将这些评级汇总为一个数字。

It seeks to measure every aspect of model training, rate how automated each task currently is, and aggregate those rankings to express them as a single number.

根据Anthropic的衡量，Claude现在“主导”了他们26%的研发工作，并在超过90%的工作中“协作”。

By Anthropic's measure, Claude now "leads" 26% of their R&D work and "collaborates" on more than 90%.

Anthropic对AI主导研发的定义是：人类提供高层目标，AI在人类监督下端到端完成这些任务。

Anthropic's definition of AI-led R&D was tasks where a human provides a high-level goal and the AI completes it end to end with human oversight.

这里你就能看到挑战所在，即使他们试图做到精确。

And here you see the challenge even as they are trying to be precise.

他们说的是AI主导研发，但那个“主导”的定义里，仍然包括人类提供高层目标，而不是AI自己决定目标。我觉得有些人听到“AI主导研发”时，可能会以为是后者。

They are talking about AI's leading R&D, but that definition of leading still includes the human providing a high-level goal rather than the AI determining the goal, which is what I think some might assume when they hear AI-led R&D.

Anthropic方面也谨慎地指出，“Claude在任何被测量的AI研发工作子集中，都不是完全自主运行的。”

For their part, Anthropic was careful to note that, "Claude is not operating fully autonomously for any measured subset of AI R&D work."

不过，他们提到，自从Mythos完成训练以来，研究自动化水平大幅提升，并且在最近几个月里还在持续加速。

Still, they noted that research automation has massively increased since Mythos finished its training run and has continued to ramp in recent months.

三月份，在Mythos之前，Anthropic的研发中只有1%是由AI主导的。

In March, prior to Mythos, only 1% of Anthropic's R&D was led by AI.

Mythos在五月份把这个比例提升到了12%，之后又翻了一番。

Mythos boosted this to 12% in May, and it has doubled since then.

关于对智能体的监督，Anthropic衡量了其智能体监督系统中的覆盖率、审查延迟和升级率。

Regarding oversight of agents, Anthropic measured coverage, review latency, and escalation rates within their agent oversight systems.

他们表示，目前大约有3万个智能体在任意时刻从事研究和工程工作。

They said that around 30,000 agents are currently doing research and engineering work at any particular time.

Anthropic声称，对于这类工作，他们对智能体行为有100%的覆盖，任何被标记的行为都会立即由AI审查，而需要实时审查的升级率大约为0.002%，也就是说，大约每47,000个行为中就有1个会被监控系统拦截。

Anthropic claims to have 100% coverage of agentic actions for this work, instant AI review of any flagged actions, and around a 0.002% escalation rate for real-time review, meaning around 1 in 47,000 actions are blocked by the monitoring system.

Anthropic还维护了一个事后审查系统，覆盖率达到100%。

Anthropic also maintains an after-the-fact review system with 100% coverage.

这个系统每周会标记大约10万个智能体记录，然后对这些记录进行误报筛选，其中大约有50个会升级到人工审查。

This system flags around 100,000 agentic transcripts per week that are then parsed for false positives with around 50 escalated to human review.

换句话说，每千份记录中大约有一到两份会引起实质性关注。

In other words, around one or two transcripts per thousand cause any material concern.

最后，在资源分配方面，Anthropic用分配给安全系统的总计算资源的百分比来衡量。

Finally, on resource allocation, Anthropic expressed their measure in terms of a percentage of overall compute dedicated to safety systems.

在用于AI辅助研发的计算资源中，大约有6%用于安全；而在AI主导研发的计算资源中，大约有12%用于安全。

Around 6% of the compute allocated to AI-assisted R&D went towards safety, and around 12% of the compute allocated to AI-led R&D went towards safety.

Anthropic指出，这个衡量标准并不包括持续监控智能体活动的分类器，而且它只是“衡量一家公司对安全重视程度的一个不完美的指标”。

Anthropic noted that this measure doesn't include classifiers that constantly monitor agentic activity and is a "imperfect proxy for how much a company focuses on safety.

这是因为安全研究是由个别研究人员设计实验组成的，这很耗时，尽管运行实验本身并不特别消耗计算资源。”

This is because safety research consists of individual researchers designing experiments, which is time-consuming even though running the experiments is not particularly compute intensive."

总的来说，Anthropic 的目标并不是提供完美的衡量标准。

Overall, Anthropic's goal was not to provide perfect measures.

相反，它是为了定义并提出一套任何 AI 实验室都可以在监管体系下报告的指标。

Instead, it was to define and propose a set of metrics that any AI lab could report as part of a regulatory system.

他们总结道：“当世界在考虑如何为前沿技术设定节奏时，我们应该尽一切可能缩小前沿实验室所知道的与公众所知道的之间的差距。”

They conclude, "As the world considers pacing the frontier, we should do everything possible to minimize the gap between what frontier labs know and what the public knows.

这意味着要更好地衡量 AI 的发展，公开报告这些信息，并让社会有机会决定如何使用这些信息。

This means better measuring the development of AI, reporting on it publicly, and giving society an opportunity to decide how to use this information.

我们希望以身作则，通过发布这些测量数据来展示透明度，并且我们会继续这样做。”

We hope to model that transparency by releasing these measurements and will continue to do so."

现在，对这一公告的反应，我认为，展现了当前这个时刻的好、坏和困难之处。

Now, the response to this announcement, I think, shows the good, bad, and difficult of this particular moment.

一方面，这是一个罕见的举动，AI 安全辩论中不同立场的人大多认为，有这些措施比没有要好。

On the one hand, it is the rare move where people on all different sides of the AI safety debate largely think that this existing is better than it not existing.

AI 安全研究员 Jeffrey Ladish 写道：“很高兴看到 Anthropic 这样做。”

AI safety researcher Jeffrey Ladish writes, "Great to see this from Anthropic.

几周前，我写过文章说这家公司可以更加透明，我很感激他们这次真的做到了，无论是这份报告，还是最近的 incident 调查报告和滥用报告。

A few weeks ago, I wrote that the company could be a lot more transparent, and I appreciate how they've stepped up, both with this and the recent incident investigation report and misuse report."

与此同时，也有很多人希望看到更多内容。

At the same time, there are plenty of people who also wanted to see more.

Prime Intellect 的 Elie Bakoush 写道：“有意思的是，OpenAI 给了我们每个研发任务的用量细分，而 Anthropic 给了我们自动化水平。”

Prime Intellect's Elie Bakoush writes, "Interesting that OpenAI gives us the breakdown of usage per R&D task and Anthropic gives us automation levels.

“如果能两者结合，看到每个研发任务自动化的演变，那就太好了。”

Would be great to combine both and get the evolution of automation per R&D task."

也有人认为，这种自我报告虽然是个好的开始，但还不够。

There are also some who argue that the self-reporting, while a fine start, isn't enough.

Arun Rao 写道：“我们需要所有实验室采用统一标准来衡量 RSI 的进展，并每周或每月公开报告。”

Arun Rao writes, "We need standard measures across all labs to measure progress to RSI and publicly report on it weekly to monthly."

现在，随着市场消化这些信息，我们会看到更多这样的讨论：把安全监控的增加看作研发的昂贵开销，这或许是一种愤世嫉俗的看法，也可能是现实主义的看法，也就是说，这些安全支出占比的指标，本质上就是在衡量这种开销。

Now, one discussion that we'll see a lot more of, especially as markets digest all of this, is that one way, perhaps a cynical way, perhaps a realistic way, to look at an increase in safety monitoring is to view it as really expensive overhead for R&D, meaning that these measures of percentage of spend on safety is basically a measurement of that overhead.

市场是会奖励那些在安全上投入更多的公司，还是会因为它们削减自身利润而惩罚它们呢？

Will markets reward companies that spend more on safety, or will they punish them for cutting into their own margins?

这又说明了这些公司在公开市场环境下运营的另一个复杂之处。

This shows another complication for having these companies operate in a public market environment.

顺便说一句，这周也提醒了我们，递归自我改进并不只是美国实验室的专利。

For what it's worth, this week also reminded that recursive self-improvement is not just the province of the US labs.

智谱AI周四发布了一篇博客文章，标题是《迈向递归自我改进：GLM如何构建自己的推理基础设施》。

ZAI released a blog post on Thursday called "Towards Recursive Self-Improvement: How GLM Built Its Own Inference Infrastructure."

文章第一句话写道：“在开发GLM的过程中，模型有时会展现出让我们惊讶甚至不安的能力。”

The first sentence reads, "As we develop GLM, the model sometimes exhibits capabilities that surprise us and even unsettle us.

最让我们震撼的最近一次是，GLM越来越能帮助构建AI本身。

The most recent moment that shook us, GLM is increasingly helping build AI itself.

我们看着模型完成了一项基础设施任务，而这项任务以前需要一支经验丰富的基础设施工程师团队花上几周时间。

We watched the model complete an infrastructure task that would have previously taken a team of experienced infrastructure engineers weeks.

当我们意识到这项工作将直接改变下一代模型的训练方式时，我们更加确信了。

When we realized that this work would directly change how the next generation of models is trained, we became even more convinced.

我们的继承者就是我们自己创造的AI系统。”

Our successors are the AI systems we are creating ourselves."

这当然超出了头条新闻的范畴，但它提醒我们，任何不包含中国的放缓讨论基本上都不算真正的讨论。

It's certainly beyond the scope of the headlines, but it is a reminder that any slowdown discourse that does not include China is basically no discourse at all.

与此同时，如果我没判断错的话，我们现在已经进入了下一代AI的新谈判期，不再只是AI实验室决定我们的节奏，很多人现在都在提出建议，讨论新的整体系统可能是什么样子。

Meanwhile, if I am correct that we have now entered this new negotiation period for the next generation of AI, where it is no longer just the AI lab that determine our pacing, lots of people are now stepping in with proposals for what the new overall system might look like.

桥水基金的首席投资官格雷格·詹森这周上了头条，他提出权力集中是AI面临的主要风险，应该受到监管。

Bridgewater CIO Greg Jensen made headlines this week by suggesting that concentration of power is the major AI risk to be regulated.

桥水是全球最成功的对冲基金之一，近年来大量使用AI。

Bridgewater is one of the world's most successful hedge funds and has used AI extensively over recent years.

在接受《信息》杂志采访时，詹森谈到了他对AI风险真正所在之处的看法，以及他会如何应对监管。

In an interview with "The Information," Jensen discussed his views on where AI risk actually resides and how he would approach regulation.

他说：“开源模型存在一个大问题，因为你可以训练它们。”

He said, "There's a huge problem with open source models because you can train them.

我们在桥水就是这么做的。

We do this at Bridgewater.

在强大的开源模型上进行强化学习训练，效果非常显著。

It's extremely effective, reinforcement learning training on powerful open source models.

它极其强大，是一项伟大的技术，但同时也显然非常危险。

It's extremely powerful, a great technology, and at the same time, clearly a very dangerous one.

你不知道人们拿它们做什么。

You don't know what people are doing with them.

没有办法追踪，所以从这个角度看很危险。

There's no way to track that, so it's dangerous from that perspective.

我们甚至还没开始讨论如何处理这个问题。

We haven't even begun the conversation on how to deal with that issue."

然而，在詹森看来，这个风险与前沿实验室本身带来的风险相比，简直是小巫见大巫。

And yet, in Jensen's view, that risk pales in comparison to the risk posed by the frontier labs themselves.

他接着说：“我们必须应对权力集中问题。

He continued, "We have to deal with the concentration of power.

两年内，OpenAI和Anthropic将控制全球35%到50%的计算能力。

In two years, OpenAI and Anthropic are going to control 35% to 50% of the world's compute.

社会竟然允许如此强大的计算能力出现这种结果，这太疯狂了。”

That's a crazy outcome for a society to allow on something as powerful as compute.

我们会让一个实体控制那么多其他形式的能源或商品吗？

Would we let one entity control that much of some other form of energy or commodity?"

过去一个月里，大家经常提到的一个观察是，Hugging Face事件以及其他类似的安全漏洞，都需要大量的资源来驱动智能体集群。

Now, one of the common observations over the past month has been that the Hugging Face incident and other security breaches like it require immense amounts of resources to power agent swarms.

目前还不清楚，除了前沿实验室之外，其他威胁行为者能否复制这些攻击，至少在当前模型下，因为他们在计算资源获取上有限制。

It's unclear that they could be replicated by threat actors outside of the frontier labs, at least with current models, because of limits in their access to compute.

Jensen认为，解决方案的一部分需要明确AI行为责任的法律指导，这样公司和个体就能清楚他们承担的风险。

Jensen argued that part of the solution needs to be clear guidance on liability for actions taken by AI so the companies and individuals understand the risks they're taking.

当被问及如何处理权力集中问题时，Jensen回应说：“我们应该把任何拥有全球或美国计算资源超过某个百分比，比如5%的实体，都视为系统重要性机构。”

When asked how he would deal with the concentration of power, Jensen responded, "We should take anybody that has more than x%, let's say 5%, of the world or US's compute resources.

然后说：‘好的，这些是系统重要性机构。’

Say, 'Okay, those are systemically important institutions.'

我们会像对待系统重要性银行那样，把它们纳入某种监管框架，并宣布会有相应的监管措施。

We're going to put them into some sort of thing like we do with systemically important banks and say there's going to be regulation.

我也认为，设定持有上限是合理的。

I also think it's fine to say we're going to have caps on how much you can own.

商品期货的持有量是有上限的。

There's a cap on how many commodity futures you can own.

比这重要得多的事情都有上限。

There's a cap on much less important things.

现有的法律可能就能做到这一点。

We may be able to do that with current law.

也许需要新的法律。

You may need new laws.

不管怎样，我们得赶紧在法律上把这件事理清楚。

Either way, let's get going on sorting that out legally.

两党应该都认识到，你不应该让某个实体垄断我认为大多数人都会同意是世界上最重要的资源之一的东西。

There should be a bipartisan recognition that you shouldn't want monopolistic control on what I think most people will agree is one of the most important resources in the world."

现在，我可以花接下来一周的节目来讨论对现有计算资源应用类似系统重要性银行监管的影响和挑战。

Now, I could spend the next week's worth of episodes discussing the implications and challenges of trying to apply GSIB-style regulation to existing compute.

但我喜欢这个讨论的地方在于，它不仅摆脱了过于简单的二元对立，还开始追问权力的真正所在。

But what I like about this discourse is that not only does it leave behind overly simplistic binaries, it starts to ask questions about where the actual locus of power is.

问题出在模型本身，还是出在运行模型的算力上？

Is the problem in the models, or is the problem in the power to run the models?

如果我们要防止不好的结果，这两者的干预点差别很大，而这正是我们需要讨论的话题。

Those have very different intervention points if we're trying to stop bad outcomes, and that's exactly the sort of conversation we need to be having.

现在，安全问题的另一个动态是法律层面的问题：实验室能否真正在安全方面进行协调，还是这涉及反垄断问题。

Now, one additional dynamic of the safety question that has been playing out is a legal question of whether the labs can actually coordinate on safety or whether that implicates antitrust issues.

据报道，特朗普政府正在考虑给予反垄断豁免，允许前沿实验室在安全方面进行合作。

The Trump administration is reportedly considering an antitrust carve-out to allow frontier labs to collude on safety.

这个理论的基本逻辑是，AI放缓可能意味着协调减少研究支出，从而以牺牲消费者利益为代价提高盈利能力。

Now, the theory that this deals with is basically that an AI slowdown could mean a coordinated reduction in research spending, and therefore an increase in profitability at the expense of the consumer.

抽象来说，打个不太严谨的比方，这就像苹果和三星同意停止研发新手机一样。

In the abstract, and in very loose analogy, it would not be dissimilar to if Apple and Samsung agreed to stop working on new phones.

美国副司法部长斯坦利·伍德沃德表示，政府正在考虑更新跨机构反垄断指南，为AI安全协调提供豁免。

Associate US Attorney General Stanley Woodward said the administration is considering an update to interagency antitrust guidance to provide a carve-out for AI safety coordination.

现有指南已经允许在网络安全风险方面进行协调，所以这一变化将把该指南扩展到AI安全领域。

The guidelines already allow for coordination on cybersecurity risks, so a change would extend that guidance to AI safety.

有意思的是，伍德沃德说，虽然多位AI高管公开呼吁反垄断豁免，但到目前为止还没有人联系过他的办公室。

Interestingly, Woodward said that although multiple AI executives have publicly called for an antitrust carve-out, no one has contacted his office on the matter as of yet.

不过，即使没有相关指导，伍德沃德也表示司法部对协调放缓没有异议，他说：“在我看来，在网络安全或安全方面进行协调并不算反竞争。”

Even without guidance, though, Woodward has indicated that the DOJ has no objections to a coordinated slowdown, commenting, "It doesn't occur to me that coordinating on cybersecurity or security is anti-competitive."

欧洲反垄断负责人特蕾莎·里贝拉也同意这一观点。

Europe's antitrust chief, Theresa Ribera, agreed.

在同一场活动上发言时，她说：“这是一个经典的博弈论问题。

Speaking at the same event, she said, "This is a classic game theory problem.

当风险是共担的，合作符合所有人的利益。”

When the risks are shared, cooperation is in everyone's interest."

与此同时，当然仍有很多人在试图缓和整个讨论的激烈程度。

At the same time all of this is happening, there are certainly still many out there who are trying to tone down the tenor of the conversation in general.

传奇AI研究员、Coursera联合创始人吴恩达在接受彭博电视台采访时，将灭绝风险斥为科幻小说，并警告说，AI末日论者在过去十年中多次推动这种说法，目的是为他们的主张争取关注并帮助塑造监管。

Legendary AI researcher and Coursera co-founder Andrew Ng, in an interview with Bloomberg TV, dismissed extinction risk as science fiction and warned that AI doomers have pushed this narrative numerous times over the past decade in a bid to gain publicity for their cause and help shape regulation.

他说：“过去两周我相当沮丧。

He said, "I was quite dismayed over the past two weeks.

这波公关宣传又开始了，大概也是出于类似的目的。

This wave of PR has kicked up again for probably similar purposes."

Andrew承认AI确实存在真实的风险，主要集中在网络安全方面，但他说：“最近这种关于AI导致人类灭绝的恐惧，更像是科幻小说而不是科学。”

Andrew acknowledged that there are genuine risks associated with AI, largely around cybersecurity, but said, "This recent fear about AI leading to human extinction and so on is much more science fiction than science.

“这非常有害。”

It's very damaging."

在他看来，AI带来的危险是“实际的工程问题”，行业需要继续关注“它能做的许多美妙的事情”。

In his view, the dangers posed by AI are, quote, "practical engineering problems," and the industry needs to continue focusing on, quote, "many of the wonderful things it can do."

说到当下真实的网络安全问题，Mistral已经第二次被黑客入侵，他们的知识产权现在可以在暗网上购买。

And speaking of real here-and-now cybersecurity issues, Mistral has been hacked for the second time, with their intellectual property now available for purchase on the dark web.

五月份，大约5GB的内部源代码和大约450个私有仓库被窃取并出售，现在又发生了同样的事情，黑客提供了完整的源代码、内部开发文件、Web应用代码以及其他专有信息。

In May, around five gigabytes of internal source code and around 450 private repos were exfiltrated and offered for sale, and now it's happened again, with hackers offering full source code, internal development files, web app code, and additional proprietary information.

一位名叫Benny的X用户说，他们联系了卖家，确认文件包含模型权重、训练后流程和数据集构建。

An X user called Benny said that they contacted the seller and confirmed the file dump included model weights, post-training pipelines, and dataset construction.

黑客要价25000美元，但不久后删除了帖子，可能是因为找到了独家买家，或者操作安全出了问题需要消失。

The hacker was asking $25,000 but deleted their post shortly afterwards, likely either because they found an exclusive buyer or got sloppy with their opsec and needed to disappear.

当天晚些时候，Mistral表示经过彻底调查后没有发现未经授权的访问证据，这或许说明这些数据可能还是五月份那次入侵中泄露的，现在又被拿出来卖了。

Later in the day, Mistral said that they found no evidence of unauthorized access after a thorough investigation, suggesting perhaps that this is the same material from the May break-in coming up for sale again.

最后今天要说的，这周有一个模型发布没引起太多关注，就是Gemini 3.8 Live Extended Thinking。

Lastly today, one model release that went a little under the radar this week was Gemini 3.8 Live Extended Thinking.

这个模型是另一个实时语音模型，也就是说它能处理连续的对话，而不是那种一问一答的模式。

The model is another live speech model, meaning it can process continuous conversations rather than using a turn-based structure.

它在Artificial Analysis的语音到语音指数中排名第一，超过了GPT Live 1 Astra和Groq Voice Think Fast 2.0。

It topped the Artificial Analysis Speech-to-Speech Index, beating out GPT Live 1 Astra and Groq Voice Think Fast 2.0.

除了流畅的对话风格，这个模型还支持接近实时的视觉输入、自动识别97种语言，以及工具调用的交接功能，让它在后台完成任务。

Aside from the smooth conversation style, the model also supports near real-time visual inputs, automatic detection for 97 languages, and handoff for tool calls to allow it to complete tasks in the background.

Google开发者关系负责人Tim Messerschmidt发布了一个很酷的技术演示，展示了这个模型驱动REACHY Mini机器人的效果。

Tim Messerschmidt, a developer relations lead at Google, published a cool tech demo showing the model powering a REACHY Mini robot.

Tim演示了模型在英语和德语之间无缝切换对话的能力。

Tim demonstrated the model's ability to keep up a seamless conversation that weaves between English and German.

Greg Isenberg在推测这些影响时写道：“隐形界面要来了吗？

Speculating about the implications, Greg Isenberg wrote, "Are invisible interfaces coming?

谷歌刚刚发布了Gemini 3.8 Live。

Google just announced Gemini 3.8 Live.

它可以和你边聊边处理任务，对话结束后还能继续干活。

It can talk through a task with you and then keep working after the conversation ends.

我觉得超过九成的垂直SaaS都需要一个语音入口。

I think 90%-plus of vertical SaaS will need a voice front door.

我的意思是，你使用软件的方式变成了和它对话，而打字、点击和填表这些操作都在后台自动完成，你完全不用管。

By that I mean the way you use the software becomes talking to it, and the typing, clicking, and form filling happens on the other side without you.

比如，一个承包商站在工地上，直接说出哪里出了问题，等他回到车里的时候，报价已经发出去了，库存也查过了，客户关系管理系统也更新了，客户收到了短信，任何有风险的地方也都给他标出来了。

So a contractor standing on a job site just says what went wrong out loud, and by the time he's back in the truck, the quote is sent, inventory is checked, the CRM is updated, the customer got a text, and anything risky is flagged for him.

这挺理想的，对吧？

Kind of the dream, right?

同样的道理也适用于护士、调度员、招聘人员、经纪人、保险代理人等等。

The same thing works for nurses, dispatchers, recruiters, brokers, insurance agents, et cetera.

人只需要说话，智能助手就把行政工作都搞定了。

The person talks, and the agent finishes the admin.

这里有很多机会可以打造以语音为先的业务。

Lots of opportunities here to build voice-first businesses.

我觉得这就是垂直软件变得隐形的方式。

I think this is how vertical software becomes invisible.

没人登录，没人填表，也没人需要学习你的界面。

Nobody logs in, nobody fills out a form, and nobody learns your interface.

你只要开口说话，工作就在你身后完成了。

You just talk and the work gets done behind you."

这让我们看到了SaaS未来的一个缩影。

This is a glimpse of where SaaS is going.

虽然还没完全实现，但这一天正在到来。

Not fully there yet, but it's coming.

在本周末的深度阅读/深度思考节目中，我会更详细地聊聊这个转变，以及我们在使用AI方面的其他一些变化。

In this weekend's Long Read/Big Think episode, I get a little bit more into this particular shift as well as a bunch of other shifts in how we use AI.

不过现在，头条新闻就先到这里。

But for now, that's going to do it for the headlines.

我们进入正片环节，聊聊经过这几周疯狂的AI安全讨论之后，大型企业接下来在AI方面的优先事项是什么。

Let's move over into the main episode where we talk about what, after this couple weeks of crazy AI safety discussions, big enterprises are thinking about their priorities with AI next.

毕马威和德克萨斯大学奥斯汀分校的一项新研究发现，当人们与AI合作时，相似的技能并不能保证相似的结果。

A new study from KPMG and the University of Texas at Austin found that when people work with AI, similar skills don't guarantee similar outcomes.

研究人员研究了500多名早期职业专业人士，发现表现最好的人通过引导、评估和完善AI的输出，持续放大了AI的价值。

Researchers studied more than 500 early career professionals and found that the best performers consistently amplified the value of AI by guiding, evaluating, and refining its outputs.

这些顶尖表现者被称为“AI放大器”，他们的定义不在于他们知道什么，而在于他们如何与AI合作。

These top performers, called AI amplifiers, weren't defined by what they knew alone, but by how they worked with AI.

想了解更多关于AI放大器与其他人的区别，请访问kpmg.com/us/aiamplifiers。

Learn more about what separates AI amplifiers from everyone else at kpmg.com/us/aiamplifiers.

Blitzy对庞大代码库的理解，能够实现自主安全修复、现代化改造和新功能开发。

Blitzy's understanding of massive code bases unlocks autonomous security fixes, modernization, and new features.

那么，当完全没有遗留代码时会发生什么？

So what happens when there's no legacy code at all?

绿地项目本应是容易的部分。

Greenfield is supposed to be the easy part.

全新的起点，没有技术债务。

Clean slate, no technical debt.

但即便是全新项目，也得按人的速度推进，一次一个冲刺。

But even greenfield moves at human speed, one sprint at a time.

Blitzy 把工作单元从开发者变成了整个项目，自主规划、构建、测试并验证整个应用，从零开始。

Blitzy changes the unit of work from the developer to the project, autonomously planning, building, testing, and validating entire applications from scratch.

数十万行可投入生产的代码。

Hundreds of thousands of lines of production-ready code.

有一位 Blitzy 客户搭建了一个全新应用，共 53.4 万行代码，把原本 65 周的路线图压缩到了两周。

One Blitzy customer stood up a brand-new application, 534,000 lines of code, compressing a 65-week roadmap into two weeks.

另一位客户在没有前端工程师的情况下，交付了完整的应用。

Another shipped an entire application with no front-end engineer.

无论是遗留系统还是全新项目，答案都一样：软件以计算的速度交付。

Legacy or greenfield, the answer is the same: software at the speed of compute.

在 blitzy.com 上构建你的下一个项目。

Build what's next at blitzy.com.

网址是B-L-I-T-Z-Y.com。

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

到了这个阶段，公司是否在用AI已经不是问题了。

At this point, it's no longer a question of whether companies are actively using AI.

但能不能用好AI，那就是另一回事了。

Using it well, on the other hand, is a whole different story.

不过，Robots & Pencils这家公司，我觉得就是为这个时代而生的。

Robots & Pencils, though, is a company that I can point to that is actually built for this time.

他们是一家应用AI工程公司，直接和客户合作，解决对业务真正重要的问题，而不是那些只停留在PPT里的实验。

They're an applied AI engineering firm working directly with clients on problems that matter to the business, not experiments that live in a slide deck.

每个项目都是从客户真正需要的结果出发，倒推着来做的。

Every engagement starts by working backwards from the outcome a client actually needs.

如果你想在这个领域里分辨出真正的AI工程和那些虚头巴脑的东西，这就是关键的区别。

If you're trying to tell real AI engineering apart from noise in this space, that's the difference maker.

去robotsandpencils.com看看吧。

Head to robotsandpencils.com.

本期《AI每日简报》由Hyperagent赞助播出，Hyperagent让你可以管理一群智能体，团队可以一起协作。

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

别再依赖本地智能体和等待你笔记本电脑上提示的聊天工作流了。

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

Hyperagent在云端部署全天候运行的智能体，在你团队已经使用的工具中完成实际工作。

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

营销智能体把竞争对手的动作变成落地页。

Marketing agents turn competitor moves into landing pages.

销售智能体丰富潜在客户信息、起草邮件，并更新CRM。

Sales agents enrich leads, draft emails, and updates the CRM.

运营智能体处理文书工作，跟踪预算。

Ops agent chases the paperwork and tracks the budget.

每个智能体都能访问共享上下文，并遵循你关于范围和审批的规则。

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

是时候拥有像队友一样的智能体了。

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

去Hyperagent招聘你的员工吧。

Hire yours at Hyperagent.

在hyperagent.com/aidailybrief领取100美元积分。

Get $100 in credits at hyperagent.com/aidailybrief.

欢迎回到AI每日简报。

Welcome back to the AI Daily Brief.

今天这期节目的背景，当然就是AI安全辩论，过去几周这个话题已经完全失控了。

The background context for today's episode is, of course, the AI safety debate, which has completely broken containment over the last couple of weeks.

然而，即使我们看到这个问题在政治层面彻底爆发，民调也显示公众态度在快速变化，各种规模的企业和公司仍然在观望，想知道这对他们到底有没有实际影响，还是说他们只需要继续按部就班地做事。

And yet, even as we have seen the political residents of this issue absolutely explode and polling showing very dynamic, fast-shifting attitudes around it, enterprises and businesses of all shapes and sizes are still stuck out here figuring out whether it has any real implications for them or whether they just need to keep on keeping on.

今天，我们来看看这个问题的具体答案，以及企业和公司目前真正关注的一些更广泛的AI挑战。

Today, we're looking at some specific answers to that question, as well as a few of the broader AI challenges that enterprises and businesses are actually focused on right now.

当然，安全辩论已经进入了商业领袖的讨论范围。

Certainly, the safety debate has found its way to the business leader conversation.

周一，在《华尔街日报》科技委员会峰会上，当《华尔街日报》要求举手示意时，只有少数人表示担心AI可能会毁灭我们所有人。

When The Wall Street Journal asked for a show of hands at the WSJ Technology Council Summit on Monday, only a few people said that they were worried that AI might kill us all.

与此同时，大约一半的与会者表示，他们支持放慢前沿AI研究的速度，并优先考虑更好的防护措施。

At the same time, around half of attendees said that they were in favor of slowing down frontier AI research and prioritizing better guardrails.

这与我上周四在大型AI安全专题中分享的观点相符，我认为虽然市场会将任何放缓视为AI的初期重大风险，但实际上，有一个非常有说服力的经济反论：发展速度放慢可能会增加企业在AI上的支出。

This sort of comports with my thesis that I shared, I think in last Thursday's big AI safety episode, that while I believe that the markets would view any sort of slowdown as initially a big risk for AI, I actually think that there is a very compelling economic counterargument that a slower pace of new development might actually increase the amount that companies were spending on AI.

这个观点基本上是，变化的速度和节奏之快，实际上在某种程度上让企业不愿进行全面的转型，因为逻辑是，他们会花费大量时间和精力去转型，但等到转型完成时，所转型的东西可能已经不再相关了。

The thesis was basically that the incredible speed and pace of change actually, in some ways, creates a disincentive for companies to try to do comprehensive transformation on the logic that they're going to spend all this time and energy on transforming into something that isn't even relevant anymore by the time the transformation is complete.

确实，回到科技委员会峰会，小组讨论的主要结论是，AI放缓实际上对企业目前使用AI的方式影响不大。

Indeed, moving back to the Technology Council Summit, on the panels, the big takeaway was that an AI slowdown actually doesn't really have that many implications for the way enterprises are using AI right now.

联邦快递数据工作区总裁维沙尔·塔尔瓦说：“这掌握在我们手中，由我们来善用AI，我认为它可以为社会带来很多好处。”

Vishal Talwar, the president of FedEx Dataworks, said, "It's in our hands, and it's up to us to apply AI for good, and I think it can do a lot of good for society."

讨论基本上仍然围绕着对正常商业风险进行审慎的AI治理，而不是主导媒体话语的生存风险。

The discussion basically still centered around the need for prudent AI governance around normal business risks, rather than the existential risks that are dominating the media discourse.

在加拿大投资峰会上，贝莱德首席执行官拉里·芬克更担心的是数据中心的反感，而不是生存风险。

Up north at the Canada Investment Summit, BlackRock CEO Larry Fink was far more worried about the data center backlash than xRisk.

他警告说，建设延误可能使AI成为“大公司的专属领域”。

He warned that construction delays could make AI the, quote, "domain of large firms."

芬克补充说：“我们越快建设更多容量，就越能实现民主化，让每个人都能用上。”

Fink added, "The faster we can build out more capacity, the more we can democratize and make it available for everyone."

与此同时，在创业领域，创始人开始思考随着智能体变得更强大，企业将需要的治理和监控工具。

In the startup world, meanwhile, founders are starting to think about the governance and monitoring tools that businesses will need as agents get more powerful.

一位风险投资人告诉《信息报》，他们开始关注那些开发提升模型安全和基础设施技术的初创公司。

One venture investor told the information that they're beginning to focus on startups building tech that improves model security and infrastructure.

换句话说，私人市场正在通过资助能够解决这些具体问题的初创公司来回应所有这些担忧，在我看来，这正是我们希望看到的。

In other words, private markets are responding to all of this concern by funding startups that can address specifics around this concern, which to me is part of exactly what you want to see.

微软也试图将AI安全讨论拉回到他们过去几个月一直在强调的更广泛的AI商业主权问题上。

Microsoft also tried to bring the AI safety discussion back to some of the broader AI business sovereignty points that they'd been trying to make for the past several months.

本周，该公司发布了一份非常详尽的长达15000字的《行为准则》文件，正如他们所说，该文件有一个压倒一切的目标，即人类必须对AI保持有意义的控制，以便它能够帮助人们过上更健康、更幸福、更有成效的生活。

This week, the company dropped a very extensive 15,000-word code of conduct document that has, as they put it, a single overriding objective, that humans must retain meaningful control over AI so that it can help people live healthier, happier, and more productive lives.

围绕这份文件的很多讨论都集中在其对AI意识的彻底否定，以及禁止设计AI甚至模仿意识。

A lot of the discourse around this document was its outright rejection of AI consciousness and a prohibition of designing AI to even imitate consciousness.

但在X上关于此事的帖子中，微软首席执行官萨蒂亚·纳德拉也提到了对企业自身的影响，他写道：“对于企业来说，必须保持对其独特和隐性知识的完全控制。”

But in his post about this on X, Microsoft CEO Satya Nadella also brought up the implications for businesses themselves, writing, "For firms, it's imperative that they retain full control over their unique and tacit knowledge.

每个组织都应该能够建立自己的持续学习循环和爬山优化机制，而不必依赖任何单一的模型提供商，并且能够以自己控制的方式将自身知识嵌入模型中。

Every organization should be able to build its own continuous learning loop and hill climbing machine without becoming dependent on any one model provider and have the ability to embed its own knowledge into models in ways they control."

基本上，这其中的一部分是关于企业中的权力集中，而应对企业权力集中的方式，就是不要以独特和专有数据的形式向企业交出权力。

Basically, part of this is power concentration in the firms, and a way to deal with power concentration in the firms is to not surrender power to the firms in the form of your unique and proprietary data.

与此同时，在AI安全讨论之外，也开始出现一些关于企业支出所反映的AI市场信号的争论。

Now, meanwhile, outside of the AI safety conversation, there had started to be some debate about AI market signals in the form of enterprise spend.

9月9日，Ramp发布了最新的AI指数，发现AI支出在AI支出最高的前1%企业中有所下降。

On September 9th, Ramp released its latest AI index that found that AI spend declined among the top 1% of businesses spending on AI.

Ramp的首席经济学家Era Carasian写道，8月份，前1%的企业每月每名员工的AI支出为7200美元，比7月份8000美元的峰值下降了10%。

In August, wrote Ramp lead economist Era Carasian, the top 1% of businesses spent $7,200 per employee per month, which was down 10% from a July peak of 8,000.

我们已经多次讨论过Ramp数据的局限性。

Now, we've talked a lot about the limits of Ramp data.

Ramp的用户群体非常非常集中于技术前沿的早期采用者。

Ramp has a very, very highly concentrated tech-forward early adopter type of audience.

它同时也在推广一些专门针对成本效率的产品。

It is also pushing products that are specifically about cost efficiency.

当然，考虑到所有这些因素，它仍然提供了一个有趣且重要的信号。

But of course, when you take all those caveats, it still provides an interesting and important signal.

关于他们在这一领域的分析，我个人认为他们低估了夏季季节性因素的影响，但他们的观点是，最成熟的AI用户正在变得更擅长使用复杂的模型架构，这些架构并不总是依赖最昂贵的模型。

Now, when it comes to their analysis in this particular area, I personally think that they are underestimating summer seasonality as a driving force, but their take is that this is about the most sophisticated AI users getting more adept at complex model architectures that don't rely on the most expensive models at all times.

Era写道："我的看法是，这不是因为中国的开源模型。

Era wrote, "My take is that it's not because of Chinese open models.

这是模型大战。

It's model wars.

降价加上越来越多的支出份额正在转向已经比前沿模型更便宜的标准和轻量模型。"

Price cuts plus a growing share of spend is shifting to standard and light models which are already cheaper over the frontier."

说到Ramp，虽然他们的数据非常有价值，但他们的分析并不总是完全准确。之前他们曾因显示企业没有采用Fable而引发了一堆恐慌性头条，他们提出了各种理由，却忽略了最明显的一个——数据保留政策。

Now, speaking of Ramp not necessarily having exactly the right analysis all the time, even though their data is really valuable, while they had previously caused a bunch of frantic headlines when they showed that businesses weren't adopting Fable, arguing all sorts of reasoning for that besides the obvious one, which was data retention policies.

但值得肯定的是，在最近的报告中，他们发现取消了数据保留要求的Fable 5.1已开始占企业支出的22.5%，并且增长非常迅速。

But to their credit in more recent reporting, they found that Fable 5.1, which got rid of the data retention requirements, had started to make up 22.5% of enterprise spend and was rising very quickly.

但总体情况如何呢？

But what about overall?

9月10号，Box的Aaron Levie在X上发了一篇长文，讲他从各行各业高管那里听到的问题，包括银行、媒体、信息服务、保险这些行业，他们在考虑企业里的AI和智能体时遇到的麻烦。

On September 10th, Box's Aaron Levie wrote a long post on X about the issues that he was hearing about from executives across industries, including banking, media, information services, and insurance when thinking about AI and agents in the enterprise.

Aaron听到的主要趋势包括网络安全、模型竞争、智能体安全和身份管理、流程再造、架构调整、评估，还有遗留系统这个老大难问题。

The big trends that Aaron heard about were cyber, model battles, agent security and identity, process re-engineering, architecture adjustment, evals, and the hurdle of legacy systems.

关于网络安全，他写道：“大家都对AI带来的漏洞越来越多感到紧张，也担心OpenAI和Hugging Face那个事件的影响。”

On cyber and security, he wrote, "Everyone is nervous about the growing rate of vulnerabilities coming at them from AI and the implications of the OpenAI Hugging Face incident.

“这个讨论不像硅谷那样关乎生死存亡，但大家还是很担心，也很务实，想着在各自的环境里怎么操作上应对。”

The conversation is not as existential as it is in Silicon Valley, but still highly concerned and pragmatic about what to do about it operationally in their environments."

AI带来了很多新发现，但要跟上所有需要执行的变化还是很难。

Lots of new discoveries due to AI and still hard to keep up with all the changes they have to execute now.

接着他谈到智能体安全，继续写道：“和Hugging Face有点关联，大家现在更意识到智能体安全和身份管理的新挑战，因为智能体想进入它们能接触到的所有系统。”

Adding in what he wrote about agent security, he continues, "Somewhat tied to Hugging Face, there's much more awareness to the new challenges around agent security and identity management in a world where agents are trying to get into every system they can.

“理想情况下，企业可以为所有智能体设置身份，控制它们的行为，但当然，有时候智能体也需要完全像用户一样行事。”

In a perfect world, enterprises could set up identities for all their agents and control what they're doing, but of course, sometimes the agent needs to act exactly as the user as well."

这当然是我们之前在播客对话和Superintelligent里经常看到的情况，有意思的是，很多安全担忧不只是针对恶意行为者。

This is certainly something that we've seen a lot in our conversations in the podcast context, as well as Superintelligent, and interestingly, a lot of the security concerns aren't just about malicious actors.

这在某种程度上源于这些系统本身强大的能力。

It is to some extent rooted in just the general power of these systems.

现在，不只是工程师能接触智能体，很多公司发现这些智能体功能强大到超出了非工程师使用者的控制范围，即使这些非工程师并没有想做什么坏事。

Now that it's not just engineers who have access to agents, many companies are finding that the agents are powerful enough that they escape the containment of the non-engineers that are using them, even if those non-engineers aren't trying to do anything problematic.

这一点在数据上也有所体现。

This is also showing up in the numbers.

再看Ramp的数据，首席经济学家Aaron Karaszan写道：“企业增加支出的一个领域是AI安全软件。”

Once again, looking at Ramp data, lead economist Aaron Karaszan writes, "One area companies are increasing their spend is AI security software.

在Hugging Face被黑事件之后，我们排名前三的软件供应商中有三家专门开发用于监控生产环境中智能体的软件。”

In the wake of the Hugging Face hack, three of our trending software vendors make software specifically designed to monitor agents in production."

不过他也指出，这些特定供应商可能并没有采取任何措施来阻止Hugging Face被黑事件，但这显然是商业买家和初创企业建设者关注的焦点。

Now, he did point out that these specific vendors might not have done anything to stop the Hugging Face hack, but it's clearly a category of focus for business buyers as well as startup builders.

Aaron提到的另一个有趣领域是开源和替代模型架构的初步探索。

Another interesting area that Aaron talks about is the nascent exploration of open source and alternative model architectures.

他写道：“大多数公司正在企业内部部署多个前沿模型。”

He wrote, "Most companies are deploying multiple frontier models within their enterprise.

很难统一标准，团队和用例之间的偏好各不相同，但资金仍然集中在少数几家供应商上。

Too hard to standardize on anything and seeing different preferences across their teams and use cases, but the dollars are still concentrated on just a few vendors.

开源权重模型在这些组织中大规模应用还处于早期阶段，通常是因为国内缺乏前沿的开源选项。

Open weight's still in infancy at scale in most of these organizations, often due to lack of domestic frontier open source options.

大家对此有很高的兴趣，但到目前为止，选择很少。

Plenty of appetite for more options here, but so far, few places to go."

我认为，与此相关的是，Aaron观察到架构调整非常果断。

Now, paired with that, I think, is Aaron's observation that there is a ruthless adjusting of architectures.

他写道：“大多数公司，在过去一两年里，都有多次更换系统供应商的例子。”

"Most companies," he wrote, "had examples of changing systems out multiple times just in the past year or two with different vendors.

我可能从未像现在这样频繁听到‘我们试过X，但不行，所以改用Y’这种说法。

I probably haven't heard 'we tried X and it didn't work, so have gone with Y' more than in today's environment.

这里的教训是，因为创新速度太快，没有人会等着某个供应商把产品做好。

The lesson here is that because innovation is happening so fast, no one hangs around until a vendor gets something right.

他们只会转向下一个供应商。”

They just move on to the next one."

我觉得这为观察当前的竞争动态提供了非常有趣的背景，看看不同类型的参与者是如何试图吸引不同类型的企业。

Now, I think this sets really interesting context for watching some of the competitive dynamics right now around how different types of actors are trying to appeal to different types of businesses.

像OpenAI和Anthropic这样的实验室显然是想把所有东西都整合在自己的环境里。

Labs like OpenAI and Anthropic are clearly trying to keep everything consolidated in their own environment.

其中一部分，尤其是对OpenAI来说，就是真正专注于更便宜的模型，尽可能把模型价格压到最低。

Part of that, especially for OpenAI, has been a real focus on cheaper models and pushing the price of their models down as far as they possibly can.

过去几周我们已经看到了很多这样的情况。

That's something we've seen a lot over the past couple of weeks.

但他们也像今年一直做的那样，专注于垂直解决方案，比如OpenAI新推出的Astra for Law。

But they also are, as they have been all year, focused on vertical solutions, such as the newly launched Astra for Law from OpenAI.

在推出Astra for Law的同时，OpenAI和律师事务所Cooley还联合推出了一款名为Go Public的产品，旨在起草S1文件，这些文件是公司上市前必须向美国证券交易委员会提交的文件。

Alongside the Astra for Law launch, OpenAI and law firm Cooley also co-launched a product that they called Go Public that's designed to draft S1 filings, which are the documents that companies must submit to the SEC before they can go public.

然而，仿佛是为了让我们完美对比不同公司采取的不同选择，另一家律师事务所Latham & Watkins最近被报道购买NVIDIA服务器来建立自己的内部系统，特别是作为OpenAI和Anthropic模型的替代方案。

And yet, as if to give us a perfect comparison of the different types of options that different companies are taking, another law firm, Latham & Watkins, was recently reported to be buying NVIDIA servers to set up their own in-house systems, specifically as an alternative to models from OpenAI and Anthropic.

Latham是美国第二大律师事务所，他们说，引用一下，"有时候我们可能有非常敏感的信息，客户信息我们真的想保护，我们不想把它放在任何云供应商那里。"

Latham, which is the US' second-largest law firm, said, quote, "Sometimes we may have information that is so sensitive, client information that we really want to protect, we don't want to put it on any cloud vendor."

Mistral的首席执行官Arthur Mensch似乎想说明，企业要想避开AI安全争论的漩涡，一个办法就是拥有自己的模型。他说：“企业不要急于构建和拥有自己的AI模型和系统，这样就不会有末日降临。”

Seeming to make the point that one of the ways that enterprises can stay out of the fray of the AI safety discourse is to own their own models, Mistral CEO Arthur Mensch posted, "Don't pace building and owning your own AI models and systems as an enterprise, and there will be no doomsday for you."

Foundation Capital的Jaya Gupta也认为，这个放缓前沿技术发展的时刻对现有软件公司来说可能是个好机会。

Foundation Capital's Jaya Gupta also thinks that this pacing the frontier moment could be a good one for the software incumbents.

她写道：“如果你是任何一家软件公司的CEO，而现在还没有提供开放权重模型作为产品，那你就是在睡觉。”

She wrote, "If you're the CEO of any software company and you're not offering open weight models as a SKU right now, you're asleep.

放缓前沿技术可能是现有软件公司有史以来最大的机遇。

Pace the frontier may be the greatest invitation software incumbents have ever gotten.

当实验室在争论智能应该以多快的速度发展时，软件公司应该竞相将我们已经拥有的智能商品化。

While the labs debate how quickly intelligence should advance, software companies should be racing to commoditize the intelligence we already have.

制药公司和银行已经开始采用开放权重模型，部分是为了利润，部分是因为实验室提供的可撤销API是他们越来越不想要的依赖。

Pharma and banks are already picking up open weight models, partly for margins, partly because a revocable lab API is a dependency they increasingly don't want.

AI原生公司和关心销售成本的科技公司也在这样做。

AI natives and tech companies that care about cost of goods sold are doing the same.

大多数尝试过的软件公司都失败了，因为开放权重模型之前表现不佳。

Most software companies that tried had failed attempts because the open weight models sucked.

但现在开放权重模型已经足够好了。

But now open weight models are good enough.

我认为每家大型软件公司都应该成为自己垂直领域的模型工厂。

I believe that every major software company should become a model factory for its own vertical.

掌握评估，针对自己独特的工作负载对开放权重模型进行后训练，将这些模型提供给客户，并利用生产反馈不断改进它们。

Own the evals, post train open weights on the workload it uniquely sees, serve those models to its customers, and use production feedback to continuously improve them."

所以一方面，从某种程度上说，企业对AI安全讨论的回应似乎还是照常营业。

And so on the one hand, to some extent, the response from businesses to the AI safety discussion seems to be business as usual.

企业内部的采用速度和挑战一直都有其独特性，而且伴随着大型机构固有的惯性挑战。

The pace and challenges of adoption within the enterprise were always unique and distinct to them, and the challenges of big institutional inertia that comes with them.

但另一方面，这场对话也在强化一些已经开始的趋势。

But on the other, there are some ways that the conversation is reinforcing trend lines that were already starting.

公司需要花更多时间和资源在网络与安全问题上。

Companies will need to spend more time and resources on cyber and security issues.

这本来就会发生，但现在这个观点被表达得更加清晰了。

That was always coming, but the point has been made even more crisply now.

虽然之前已经有一些令人信服的理由去探索和考虑开放权重或更自主的模型方案，而不是完全依赖大型供应商，但现在有更多的理由愿意走这条路，尤其是监管干预的可能性越来越大。

And while already there were some compelling reasons to explore and consider open weights or more owned model alternatives to just getting in bed with the big vendors, there are now even more reasons to be willing to walk down that path, not least of which is the increasing likelihood of regulatory disruption.

我觉得如果要把我的建议总结成一句话，那就是虽然整体上AI安全讨论的变化对企业短期影响不大，但它确实强化了一个事实：那些愿意尝试最困难的事情的公司，比如真正投资于自己的自主架构，比以往更有潜力在同行和竞争对手中脱颖而出。

I think that if I had to summarize my advice in a single thought, it's that while overall the changing AI safety discourse doesn't really impact the short term for enterprises, it certainly reinforces the fact that the companies that are willing to try the hardest things, like actually investing in their own owned architectures, have even more potential to differentiate from their peers and competitors than they did before.

我当然会继续关注这些趋势的发展，但今天就到这里，这是今天的AI每日简报。

I will, of course, continue to watch these trends as they evolve, but for now, that's going to do it for today's AI Daily Brief.

感谢你们的收听或观看，一如既往，下次再见，祝平安。

Appreciate you listening or watching, as always, and until next time, peace.