← 返回任务列表

What the Best Business AI Users Are Doing Different

221 段 · 1 位说话人 · 原片 22:38
M1
M10:00

那些最好地使用AI的企业,其实做法有点不一样。

The businesses that are using AI the best are really doing things a little bit differently.

根据最近的一项调查,他们正在构建模型路由器,制定组织架构和数据主权策略,并且总体上让他们的AI管理层变得更加健壮。

According to a recent survey, they are building model routers, building organizational and data sovereignty strategies, and generally making their AI management layer much more robust.

与此同时,他们使用AI的领域以及从中获得的价值也在发生变化。

Along with that, what they are using AI for and the value that they are seeing from it is changing.

在所有这些中,他们正在打造一个其他企业可以效仿的模板。

And in all of this, they are building a template that other businesses can follow.

这也是我们今天要讨论的内容。

And that's what we'll be discussing today.

《AI每日简报》是一个关于AI领域最重要新闻和讨论的每日播客和视频节目。

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

好了朋友们,在我们深入之前,先说几个快速通知。

All right, friends, quick announcements before we dive in.
M1
M10:54

我们最近讨论了很多关于个人AI代理的新兴竞争,而在那些代理大战中的一个早期胜利,让Meta的股票迎来了多年来最好的一个月。

A lot of our discussion recently has been around the new emerging competition for personal AI agents, and an early win in those agent wars has delivered Meta stock its best month in years.

Meta的股票在九月份上涨了27%,尽管本周因为OpenAI发布了一个Muse的竞争对手的消息而下跌了5%。

Meta stock was up 27% in September, even with a 5% slide this week on news that OpenAI was launching a competitor to Muse.

这使得Meta自2022年11月以来的单月表现最佳,当时公司开始了他们的效率之年,进行了大规模裁员、招聘冻结,并在元宇宙计划上有所节制。

That made it Meta's best single-month performance since November of 2022, when the company began their year of efficiency with mass layoffs, hiring freezes, and a bit of temperance on their metaverse plans.

Muse早期的成功到目前为止已经增加了5000亿美元的市值。

Muse's early success has so far added half a trillion dollars in market cap.

但更重要的是,它向市场表明Meta有一套可行的AI战略。

But even more important than that, it has given the market an indication that Meta has a viable AI strategy.

华尔街共识已将Meta推至强力买入,但并非所有人都信服。

The Wall Street consensus has pushed Meta to a strong buy, but not everyone is convinced.

Needham的分析师Laura Martin是少数在周四的一份报告中坚持持有评级的人之一。

Needham analyst Laura Martin is one of the few sticking with a hold rating in a Thursday note.

在肯定他们的同时,她写道Meta显然已经从元宇宙转向了以Muse为中心的个性化代理AI。

Giving them credit, she wrote that Meta has clearly pivoted away from the metaverse and towards personal agentic AI with Muse at the center.

不过,她对回报持怀疑态度,指出据她所说,Meta在以新产品的商业化方面是出了名的慢。

However, she's skeptical of the payoff, noting that Meta is, in her words, notoriously slow at monetizing new products.

事实上,Muse目前除了最重度用户外对所有用户都是免费的,而且Meta表示他们计划将用户数据与广告业务隔离开来。

And indeed, Muse is currently free for all but the biggest power users, and Meta has said they plan to keep user data segregated from their ad business.

即使在本周早些时候推出了企业平台,Morningstar也认为Meta运营企业业务的能力是“未经证实的”,因为消费者产品仍然是公司的核心。

Even with the launch of an enterprise platform earlier this week, Morningstar argued that Meta's ability to run an enterprise business is, quote, unproven, as consumer products remain at the center of the company.

不过,如果你是Meta,你肯定感觉不错。

Still, if you're Meta, you gotta be feeling pretty good.

六个月前,投资者还在质疑该公司在AI方面的基本能力,而现在他们质疑Meta能否将成功的赌注变现,这是一个巨大的转变。

6 Months ago, investors were questioning the company's basic competence in AI, so the fact that they're now questioning Meta's ability to monetize their successful bets is a huge shift.

现在,继续市场话题,Anthropic正力争在感恩节前完成IPO,投资者会议定于本月晚些时候举行。

Now, staying on the markets theme, Anthropic is pushing to get their IPO out before Thanksgiving, with investor meetings set for later this month.

彭博社报道称,Anthropic计划在11月9日那一周开始IPO营销。

Bloomberg reports that Anthropic aims to begin marketing the IPO in the week of November 9th.

假设首个交易日定在感恩节那一周的早些时候,那么他们将有大约两周的时间进行路演。

That would give them around 2 weeks for the roadshow, assuming the first day of trading somewhere early on the week of Thanksgiving.

消息人士称,时间表仍有变动的可能,但Anthropic正在大力推动在年底前完成交易。

Sources said that the timeline is still subject to change, but Anthropic is pushing hard to get the deal out by the end of the year.

在营销活动之前,Anthropic计划于10月14日举办投资者日。

Ahead of the marketing campaign, Anthropic is planning to host an Investor Day on October 14th.

可能参与IPO的机构投资者已受邀参加此次活动,届时他们将有机会与高级管理层会面。

Institutional investors who may participate in the IPO have been invited to the event, which will give them an opportunity to meet with senior management.

时间表趋于明确的消息,是在路透社本周早些时候泄露了Anthropic的S-1招股说明书部分内容之后传来的。

News that the timeline is firming up comes after Reuters leaked portions of Anthropic's S-1 prospectus earlier in the week.

S-1文件显示,Anthropic在2025年的营业收入为46亿美元,运营亏损为80亿美元。

The S-1 revealed that Anthropic had an operating loss of $8 billion on revenue of $4.6 billion in 2025.

围绕这些数字有很多悲观论调。

There has been a lot of doomsaying around these numbers.

但我非常非常坚定地认为,如果真到了紧要关头,投资者会关心Anthropic 2025年的数据吗?Anthropic在2026年有超过10倍的增长,我觉得这就像是把Anthropic当成一家前AI时代的公司来对待了,而它显然不是。

But my position, which I feel very, very strongly about, is that the idea that when push comes to shove, investors are going to care about 2025 numbers when Anthropic has more than 10x growth in 2026, I just think is treating Anthropic like they are a company from the pre-AI days, which they are very decidedly not.

这绝不是说投资者会喜欢他们在完整招股说明书中看到的一切,只是说如果IPO表现不佳,那不会是2025年的运营亏损造成的。

Now, that is not at all to say that investors are going to love everything they find in the complete prospectus, just that to the extent the IPO underperforms, it won't be their operating loss from 2025 that did it.

看来我并不是唯一这么想的人,彭博社报道说,潜在的IPO投资者认为公司能达到1.8万亿到2万亿美元的目标估值。

And I appear to be not alone in this, with Bloomberg reporting that potential IPO investors believe the company can achieve its target valuation of between $1.8 and $2 trillion.

顺便说一下,这预计将是历史上最大的IPO,募资额超过SpaceX的750亿美元。

This is, by the way, expected to be the largest IPO in history, taking in more than the $75 billion raised by SpaceX.

有人拿这个高得离谱的目标估值和之后不可避免的下跌开玩笑,Liquidity发帖说,如果Anthropic在感恩节前上市,我宁愿等到他们的黑色星期五/网络星期一打75折时再投资。

Joking about that insanely high target valuation and the inevitable dip that comes after, Liquidity posted, if Anthropic goes public before Thanksgiving, I'd rather just wait to invest during their Black Friday/Cyber Monday 25% discount.

与此同时,还是说Anthropic,也许出乎很多人的意料,特朗普总统似乎对Dario产生了好感。

Meanwhile, staying on Anthropic, perhaps against many expectations, President Trump seems to have taken a liking to Dario.

周四,《时代》杂志发表了与总统的广泛访谈,其中一句引人注目的引用是特朗普对与Anthropic CEO会面的积极印象。

On Thursday, Time magazine published a wide-ranging interview with the president, and one of the quotes that stood out was Trump's positive impression on meeting the Anthropic CEO.

特朗普说:“我非常喜欢他和他的妻子。”

I liked him and his wife a lot, Trump said.

“非常聪明的人。”

Very smart guy.

“也许和我想象的有点不一样。”

Maybe different than I thought a little bit.

“确实有点不一样。”

Really a little bit different.

“但不会,他懂。”

But no, he understands.

特朗普在周日与Amodei及其妻子共进了两小时的晚餐,赶在本周科技领袖聚会之前。

Trump met with Amodei and his wife for a 2-hour dinner on Sunday ahead of this week's gathering of tech leaders.

看来总统离开时对Dario的世界观有了更细致的理解。

It seems the president came away with a more nuanced understanding of Dario's worldview.

特朗普评论说:“我和他谈过他的观点,我认为这些观点与媒体描绘的非常不同。”

Trump commented, I spoke to him about his views, and they're much different, I think, than what is portrayed in the media.

这些评论表明,白宫与Anthropic之间的某些敌意至少是由工作人员推动的,而非特朗普本人。

The comments suggest that at least some amount of the acrimony between the White House and Anthropic has been driven by staffers rather than Trump himself.

六月,你可能记得《连线》杂志报道说,工作人员对Anthropic联合创始人Tom Brown接管了Fable发布的相关谈判感到宽慰,因为他没有“像Dario那样古怪”。

In June, you might remember Wired reported that staffers were relieved that Anthropic co-founder Tom Brown had taken over negotiations around Fable's release because he was not, quote, being a weirdo like Dario.

《时代》那篇文章还详细说明了总统本人对使用AI的投入程度。

The Time article also unpacked just how much the president has gotten into using AI himself.

一位工作人员说,特朗普在12月与Elon Musk会面后,花了几个小时与Grok聊天,询问这个聊天机器人关于他的总统遗产。

A staffer said Trump had spent hours talking to Grok after a meeting with Elon Musk in December, asking the chatbot about his presidential legacy.

当时,特朗普正在权衡是否升级在委内瑞拉的行动。

At the time, Trump was weighing up an escalation in Venezuela.

他问Grok,《时代》写道,“如果美国抓获马杜罗,委内瑞拉人会如何反应。”

He asked Grok, writes Time, quote, how Venezuelans would react if the US captured Maduro.

《时代》继续写道,据在场的官员说,聊天机器人回答说,马杜罗是一个压制性且极不受欢迎的独裁者,许多委内瑞拉人可能会庆祝他的倒台。

Time continued, according to the official present, the chatbot responded that Maduro was a repressive and deeply unpopular dictator and that many Venezuelans would likely celebrate his downfall.

在特朗普下令于次月执行抓捕马杜罗的任务后,街头爆发了庆祝活动。

After Trump ordered the mission to seize Maduro the following month, celebrations broke out in the streets.

据官员称,特朗普离开时觉得Grok很有智慧。

Trump, according to officials, came away thinking Grok was ingenious.

这让很多人得出了下一个结论,Hakio在X上总结道:“等等,难道伊朗的惨败是因为某个大语言模型告诉他斩首行动也会迅速导致政权垮台吗?”

That led a lot of folks to jump to the next conclusion, summed up by Hakio on X: Wait, so there's a chance the Iran fiasco is happening because some LLM told him that decapitation is going to quickly lead to regime's downfall too?

真是活久见。

What a timeline.

接下来,对于那些认为太空数据中心只是营销噱头的人来说,SpaceX已经正式将一枚AI芯片发射到太空,作为构建轨道数据中心网络的第一步。

Next up, for those who thought that data centers in space were just a marketing ploy, SpaceX has officially launched an AI chip into space as a first step towards building an orbital data center network.

周四一次成功的试飞使用了猎鹰9号火箭,将一颗包含4块谷歌TPU的卫星送入轨道。

A successful test flight on Thursday used a Falcon 9 rocket to place a satellite containing 4 Google TPUs into orbit.

这是谷歌Suncatcher项目下的首次发射,该项目是去年11月宣布的SpaceX与Planet Labs的合作。

This is the first launch under Google's Suncatcher project, a collaboration between SpaceX and Planet Labs announced last November.

Suncatcher旨在确定在当前技术下,在太空中运营太阳能数据中心是否可行。

Suncatcher aims to determine whether operating solar-powered data centers in space is feasible with current technology.

当然,4块TPU对于任何实际用途来说都是微不足道的算力,所以这纯粹是对芯片和其他组件的压力测试。

Now, 4 TPUs is of course an insignificant amount of compute for any real purpose, So this is purely a stress test for the chips and other components.

到目前为止,测试看起来不错。

So far, the test looks good.

Project Suncatcher 的高级总监 Travis Beals 报告说,他们的团队已经与卫星通信,一切运行正常。

Travis Beals, the senior director of Project Suncatcher, reported that their team has communicated with the satellite and everything is operating as expected.

Beals 写道,这是长期研究“登月计划”的第一步,旨在探索太空未来能否承载可扩展的机器学习基础设施。

Beals wrote, this is the first step in a long-term research moonshot exploring whether space could one day host scalable machine learning infrastructure.

在接下来的几周里,我们将收集在轨数据,了解我们的 TPU 如何应对太空飞行的物理压力,以及太空中的辐射和极端温度。

Over the coming weeks, we'll gather in-orbit data on how our TPUs handle the physical stress of spaceflight and the radiation and thermal extremes of space.

有些事情只能在太空中测试。

Some things can only be tested in space.

随着实验的开始,我们会利用学到的东西来优化设计,也很期待随着任务推进分享更多进展。

As we begin our experiments, we'll use what we learn to refine our designs, and we're excited to share more as the mission unfolds.

就测试范围而言,TPU 每次只会运行 15 分钟,以免给卫星的电源和热管理系统造成过大负担。

Now, in terms of the scope of this testing, the TPUs will only operate in 15-minute bursts to avoid straining the satellite's power and thermal management systems.

如果芯片能正常工作,下一步就是发射两颗更大的卫星,测试更重的工作负载和基于激光的网络技术。

If the chips function, the next step will be to launch a pair of larger satellites to test heavier workloads and laser-based networking technology.

记住,这是一个可能需要好几年甚至几十年的项目,依赖多项技术突破。

Keep in mind, this is a multi-year, if not multi-decade, endeavor that relies on multiple technological breakthroughs.

中期来看,Google 计划发射一个由 80 颗 TPU 卫星组成的网络,并且已经在开会讨论设计一个足球场那么长的巨型卫星。

In the medium term, Google plans to launch a network of 80 TPU satellites and is taking meetings to design a mega-satellite the length of a soccer field.

要达到这种规模,Falcon 9 是不可能的,所以这个项目需要依赖 SpaceX 的 Starship 成功——它在九月份首次进入轨道。

Hitting this kind of scale is impossible with the Falcon 9, so the project hinges on the success of SpaceX's Starship, which reached orbit for the first time in September.

Beals 在解释这个项目的雄心和规模时说,如果五年后我们做的一切都完美成功,那很可能意味着我们承担的风险不够多,学到的东西也不够。

Explaining the ambition and scope of the project, Beals said, if 5 years from now everything we've done has worked perfectly, it probably means we've not taken enough risk and we've not learned as much as we could.

如果我们长期来看真的成功了,这最终会变得很普通,人们不会意识到他们的 Gemini 查询可能是在太空里处理的。

If we're really successful with this in the long run, this will ultimately be boring and people won't think anything of the fact that their Gemini query might be getting served in space.

转到 AI 安全方面的戏剧性事件:OpenAI 解雇了安全团队的三名员工,原因是他们不当处理公司信息。

Moving over to AI safety drama, OpenAI has fired 3 employees on their safety team for mishandling corporate information.

OpenAI 在一份声明中说,我们与三名个人分道扬镳,因为他们违反了关于访问和处理敏感公司信息的规定。

In a statement, OpenAI said, we have parted ways with 3 individuals for violating our policies on accessing and handling sensitive company information.

我们的调查确认,这些人在既定公司流程之外处理了敏感信息,违反了我们的政策,也破坏了工作中必要的信任。

Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.

与 The Information 交流的消息人士还补充说,这件事涉及向一个做 AI 评估的外部组织泄露敏感信息。

Sources speaking with The Information added that the matter involved sharing sensitive information with an outside organization that does AI evaluations.

考虑到目前围绕 AI 安全的所有争议,以及我们对细节知之甚少,针对这些解雇事件出现大量不同的解读,一点也不奇怪。

Now, given how contentious everything surrounding AI safety is right now, and given how little we know of the details, there are unsurprisingly a huge range of interpretations when it comes to these particular dismissals.

有些人怀疑这是否应该被视为对告密者的报复。

Some wondered whether this should be read as retribution against whistleblowers.

CodeAI 的总法律顾问 Nathan Calvin 则担心 OpenAI 放弃了熟练的安全研究员。

And CodeAI General Counsel Nathan Calvin was concerned that OpenAI had let go of skilled safety researchers.

他们的名字泄露后,他指出这些人是关于思维链监控的学术研究的主要作者,并评论说,他们在安全可监控性崩塌的时候离开 OpenAI,真是太糟糕了。

After their names were leaked, he noted that they were lead authors on academic work around chain-of-thought monitoring, commenting, them leaving OpenAI right when safety monitorability is collapsing is terrible.

另一些人则不太同情违反公司规定的行为,Mark Kretschmann 写道,做 AI 安全不代表你就可以随意泄露机密信息。

Others had less sympathy for breaking corporate policy, with Mark Kretschmann writing, working on AI safety doesn't give you a free pass to leak confidential information.

这个标签并不能让你在其他人必须遵守的规则之外获得道德豁免。

The label isn't a moral exemption from the rules everyone else has to follow.

如果指控属实,解雇他们似乎完全合理。

If the allegations are accurate, firing them seems entirely reasonable.

从外部来看,很难判断 OpenAI 解雇这些员工是否合理,很大程度上是因为我们不知道泄露了什么信息。

From the outside, it is extremely hard to know how justified OpenAI was in removing these employees, in large part because we don't know what information was leaked.

我猜,对这件事的解读大多取决于一个人对 AI 安全的看法。

My guess is that mostly the interpretations of this are going to be based on what one thinks about AI safety.

如果某人极其担忧安全问题,那么 OpenAI 因为员工与独立第三方分享担忧而解雇他们,感觉就站不住脚。

If one is extremely concerned about safety issues, then OpenAI firing people for sharing their concerns with independent third parties feels unjustifiable.

另一方面,也有很多人同意 OpenAI 的看法:关心 AI 安全不代表你就有权在现有流程之外,把公司机密透露给独立组织。

On the other hand, there are going to be lots of people who agree with OpenAI that being concerned about AI safety doesn't give you carte blanche to share confidential company matters with an independent organization outside the existing pathways for doing so.

我想,不管我们在这个光谱上站在哪一边,大家可能都能同意的是,这件事很好地提醒了我们,尽快建立正式的举报和第三方评估渠道有多重要。

I think what perhaps many of us could agree on, regardless of where we sit on that spectrum, is that this is a good reminder about why it is important to get the formal channels for reporting and third-party evaluation up and running as soon as possible.

最后,如果你最近试新模型还没试够的话,似乎 Fable 5.5 的首批踪迹已经开始出现了。

Lastly today, in case you haven't had enough to do trying new models recently, it appears that the first sightings of Fable 5.5 are starting to show up.

小道消息说,有些Anthropic的用户被引导到了一个版本,知识截止日期更新了,还测试了一些很酷的新设计。

The rumor mill is suggesting that some Anthropic users are getting routed to a model with an updated knowledge cutoff and some slick new design tests.

Chubby在X上发帖说:来了来了!

Chubby on X posted, here we go!

很多用户反映,他们的查询被路由到了Fable 5.5。

Numerous users are reporting that their queries are being routed to Fable 5.5.

这只是时间问题。

It was only a matter of time.

准备好了,全球最厉害的模型就要发布了。

Get ready, the best model in the world is about to be released.

Token Gremlin补充说,第一批Fable 5.5的结果看起来简直不可思议。

Adds Token Gremlin, the first Fable 5.5 results look incredible.

期待一个非常强大的模型,特别是在设计、3D工作和空间智能方面。

Expect a seriously powerful model, especially for design, 3D work, and spatial intelligence.

有趣的是,Opus 5.5本身就已经这么猛了,所以升级到Fable可能反而没大家想象中那么震撼。

The funny part is that Opus 5.5 is already such a monster that it may actually make the jump to Fable feel a little less dramatic than it really is.

这是接下来一周值得期待的,不过今天就到这里,新闻部分结束。

Something to look forward to for the week to come, but for now, that is going to do it for today's headlines.

接下来是主节目。

Next up, the main episode.

毕马威和德州大学奥斯汀分校的一项新研究发现,当人和AI合作时,光有类似的技能并不能保证相似的成果。

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.

研究人员调查了500多名职场新人,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.

这些顶尖高手,被称为AI放大器,他们并非单凭知识取胜,而是靠和AI协作的方式。

These top performers, called AI amplifiers, weren't defined by what they knew alone, but by how they worked with AI.
M1
M112:28

每期节目我都会聊OpenAI、Anthropic、xAI、Google和Meta之间的竞争。

Every episode, I talk about the competition between OpenAI, Anthropic, xAI, Google, and Meta.

如果你听了一段时间,可能心里已经有了偏爱。

And if you've been listening for a while, you might have a favorite.

也许你觉得OpenAI和Anthropic能保持领先,或者Meta的开源策略最终会胜出。

Maybe you think OpenAI and Anthropic can stay ahead, or perhaps Meta's open-source strategy can win out.

无论你怎么看,每家AI实验室都创造了不同的投资机会。

Whatever your view, every AI lab creates a different investment opportunity.
M1
M113:23

开会的时候,我会全神贯注。

When I'm in a meeting, I'm fully in it.

我思考的是推动目标所需的反复迭代和创意碰撞。

I'm thinking about the iteration and creative back and forth it takes to actually push a goal forward.

我不会去想记要点、跟踪待办事项这类事。

What I'm not thinking about is capturing takeaways, tracking to-dos, or any of that.
M1
M114:10

欢迎回到《AI每日简报》。

Welcome back to the AI Daily Brief.

最近出了太多新品,新模型、新形态,比如各种个人助手,搞得大家都没空喘口气,看看AI在企业里的实际应用进展。

There have been so many releases recently between new models, new form factors like all these personal agents, that we haven't had a chance to catch our breath and get the latest read on how AI continues to be adopted in the enterprise.

现在,我想说说为什么这对你很重要。

Now, let me make a pitch for why this should matter to you.

如果你在企业工作,那当然不用说,了解其他公司的情况、什么方法有效、他们遇到了什么障碍,都是很关键的。

It's obvious if you work inside the enterprise, getting a sense of where other companies are, what's working for them, what barriers they're facing for adoption.

这些信息能帮你判断你的组织处于采纳曲线的哪个位置,以及需要做哪些调整。

All of that can be critical insight that's valuable in helping you understand where your organization sits on the adoption spectrum and what you might need to do differently.

但如果你不在大企业,比如你是独立创业者或自由职业者,我觉得有两点值得你关注。

But for those who aren't in a big enterprise, for those solopreneurs among you or freelancers, 2 reasons why I think this is worth paying attention to.

第一,你们很多人可能和这些公司有交集。

The first is that likely many of you interact with these companies.

比如作为服务提供商。

Perhaps as a service provider.

所以理解他们目前的状况和面临的挑战,对你来说就很有价值。

And so understanding where they actually are and what they're going through becomes valuable in that way.

但除此之外,企业级AI的采纳会在很多方面深刻影响AI的发展,而这些影响远远不止于企业本身。

But for everyone else, enterprise adoption is going to have a dramatic impact on many parts of how AI evolves that matter far beyond just the enterprise itself.

举个例子,Ramp 最近公布的最新数据。

Take for example, the latest data published by Ramp.

在本周的 AI 指数中,他们发现代币消费比上周下降了 5.2%。

In this week's AI Index, they found that token spend fell 5.2% from last week.

有意思的是,代币数量其实是上升的,这意味着企业能够在降低使用 AI 成本的同时,增加 AI 的使用量。

Now interestingly, token volume was up, meaning that companies were able to grow their use of AI While still decreasing the cost to use that AI.

Ramp 的首席经济学家 Eric Karazian 也指出,这跟开源模型没关系,开源模型在企业支出中的占比仍然不到 5%。

Ramp lead economist Eric Karazian also pointed out that this was not about open-source models, which remain less than 5% of business spend.

他说,这个下降几乎完全是由 OpenAI 和 Anthropic 之间的竞争推动的。

The decline, he said, is driven almost exclusively by competition between OpenAI and Anthropic.

与此同时,这周 OpenAI Dev Day 上最被低估的公告之一,就是 OpenAI 现在为企业提供了一项市场功能,企业可以用他们的消费承诺,不是花在 OpenAI 的代币上,而是花在通过 OpenAI 销售的其他开放模型上。

Meanwhile, one of the most underappreciated announcements from this week's OpenAI Dev Day was that OpenAI now has a marketplace feature for enterprises where companies can use their spend commitments, not on OpenAI tokens, but on other open models sold through OpenAI.

所以企业得到的是他们越来越想要的多种模型灵活切换的能力,以及利用更便宜的开放权重模型的机会。

So what enterprises get is that multi-model flexibility that they are increasingly looking for and the ability to take advantage of cheaper open weight models.

而 OpenAI 继续得到的是客户锁定,以及即使资金流向其他代币商家,也能让消费留在自己生态内的能力。

What OpenAI continue to get is customer lock-in and the ability to keep spend in their ecosystem even if it's flowing to other token merchants.

所有这些都会影响这些不同模型提供商之间的竞争方式,当然也会影响市场如何看待它们的前景。

All of that impacts how these different model providers compete with one another and of course how the market views their prospects.

如果华尔街的投资者开始相信,用量上升但支出下降是长期趋势,你最好相信,这会影响他们愿意为基础设施的建设提供多大程度的资金支持。

If Wall Street investors become convinced that volumes up but spend down is the permanent trend, you better believe there's gonna be implications for how much they're willing to backstop and fund infrastructure buildout.

所以,说完了这些,我们再回到一些最近的数据,看看现在企业采用 AI 的情况到底如何。

So with all of that said, let's come back to some recent data from where enterprise adoption is right now.

为此,我们来看一个我们一直使用的可靠来源,就是 KPMG 第三季度的 AI Pulse 调查。

And for this, we're turning to a consistent source that we use, which is KPMG's AI Pulse survey for Q3.

第三季度的 Pulse 是基于对 20 个不同国家超过 2100 名高级领导人的调查得出的。

The Q3 Pulse was based on a survey of more than 2,100 senior leaders across 20 different countries.

结果显示,企业 AI 战略已经相当显著地走向成熟。

And shows a fairly significant maturation in enterprise AI strategy.

现在,KPMG 开始做的一件非常有意思的事情,就是根据各个组织在 AI 旅程中所处的阶段来细分结果。

Now, one of the things that's really interesting that KPMG has started doing is breaking out results based on where organizations are in their AI journey.

我的意思是,他们对比了仍处于试验阶段的组织与已经从 AI 投资中获得回报的组织的回答。

What I mean by that is that they're comparing the responses of organizations that are still in an experimentation phase compared to those who have already established ROI from their AI investments.

那些仍在实验的组织和那些已经获得回报的组织之间的差距,读起来就像一张未来几个月企业 AI 用户整体走向的地图。

And the gap between those organizations that are still experimenting and those with established ROI reads like a map of where business AI users will go in general over the next several months.

而那些已经获得回报的组织正在做的和关注的事情,很多都是我们节目里经常讨论的内容。

And a lot of the types of things that those established ROI organizations are doing and paying attention to are the things that we discuss on the show all the time.

举个简单的例子,在已经获得回报的组织中,有 48% 表示员工正在大量采用 AI 代理。

To take an easy example, among those organizations with established ROI, 48% are seeing significant employee adoption of AI agents.

这并不令人惊讶。

This should come as no surprise.

2026 年,就像我说过很多次的,是代理真正成为现实的一年。

2026, As I have said numerous times, was the year that agents became real.

如果参加我们免费和付费项目的人能说明什么的话,那就是这确实是一个正在影响企业的现象。

And if the people participating in both our free and paid programs are any indication, This is very much a phenomenon that is impacting the enterprise.

然而,走得更远的组织和仍处于实验阶段的组织之间仍然存在巨大差距。那些实验者中,只有 15% 表示员工大量采用了 AI 代理。

And yet there remains a huge gap between the organizations that are farther along and the organizations that are still in experimentation phase, with those experimenters seeing only 15% with significant employee adoption of AI agents.

现在,在这个代理时代,网络防御显然变得越来越重要。那些成熟且已获得回报的组织中,有 58% 正在部署 AI 辅助的网络防御。

Now, in this era of agents, obviously cyber defense is becoming more important, and 58% of those mature established ROI organizations are putting into place AI-assisted cyber defense right now.

这与实验组织之间相差 50 个百分点,后者只有 8% 在进行任何形式的 AI 辅助网络防御。

That's a 50-point gap with the experimenting organizations where just 8% are doing any sort of AI-assisted cyber defense.

在已经获得 AI 回报的组织中,整整 86% 拥有正式的 AI 管控层。

A full 86% of organizations with established AI ROI have a formal AI harness layer.

也就是说,他们维护着某种接口,员工通过它来与 AI 交互。

In other words, are maintaining some sort of interface through which their people are interacting with AI.

我们再次看到 55 个百分点的差距,不过即使在实验者中,也有 31% 拥有了正式的 AI 管控层。

Once again, we see a 55-point gap, although even among the experimenters, 31% have a formal AI harness layer as well.

在最近我们一直讨论的数据主权问题、组织对将数据交给模型实验室的感受,以及他们是否要构建更复杂的架构来避免其中一些问题方面,53% 已经获得回报的组织表示拥有企业级的数据主权策略,而实验者中只有 8%。

And when it comes to all these questions that we've recently been discussing around data sovereignty and how organizations feel about handing their data over to the model labs, and whether they're going to build more complex architectures that allow them to avoid some of those issues, 53% of the established ROI organizations do report having an enterprise-wide sovereignty strategy compared to just 8% of the experimenters.

KPMG 总结当前最大的差异和转变时指出,我们正在进入一个阶段,最重要的东西是位于 AI 之上的管理层,它确保一切在组织的边界和背景之内正常运行。

The way that KPMG sums up the big difference and the big shift right now is that we're moving into a phase where the most important thing is the management layer that sits on top of AI and makes sure that all of this works inside the confines and context of the organization.

53% 的企业将由AI驱动的决策问责放在C-suite层面。

53% Place accountability for AI-informed decisions at the C-suite level.

有趣的是,我们开始看到一些效率和成本方面的担忧也出现在管理层,23%的企业已经建立了某种模型路由能力,覆盖AI管理的各个维度,从实验、战略规划、规模化、推动采用,到确立投资回报率。

And interestingly, we're starting to see some of the efficiency and cost concerns show up in the management layer as well, with 23% having built some model routing capability across all dimensions of AI management, from experimentation to strategic planning to scaling to driving adoption to establishing ROI.

组织越成熟,就越有可能针对特定问题采取正式的跨职能或全企业范围的方案。

The more mature an organization gets, the more likely it is that they have a formal cross-functional or formal enterprise-wide approach to that particular issue.

但他们在用这个新基础设施做什么呢?

But what are they using this new infrastructure to do?

生产力仍然是AI的一个关键目标,但实际上,这个比例从今年第一季度的42%下降到了今年的37%。

Productivity still remains a key goal of AI, but is actually down from being reported as a key goal by 42% of organizations in Q1 of this year to 37% of organizations in this year.

相反,KPMG所谓的运营优先级正在成为AI要解决的重要举措,越来越受到重视。

Instead, a lot of what KPMG calls operating priorities are gaining ground as the important initiatives that AI is meant to address.

人机协作的比例从Q1到Q3上升了4个百分点。

Human-AI collaboration jumped 4 points from Q1 to Q3.

负责任AI、治理、信任和安全也上升了4个百分点。

Responsible AI and governance and trust and security also jumped 4 points.

适应性和韧性上升了3个百分点。

And adaptability and resilience jumped 3 points.

说到我们一直关心的效率与机会型AI的问题,KPMG写道,四分之一的组织报告正在开发多智能体系统,五分之一的组织在编排多个智能体。

And when it comes to our perpetual question about efficiency versus opportunity AI, KPMG writes, a quarter of organizations report developing multi-agent systems and a fifth are orchestrating multiple agents.

随着他们这样做,目标也在扩大。

The purpose is broadening as they do.

以收入为导向的智能体策略自Q1以来有所上升,而以效率为导向的策略则有所下降,大多数组织现在同时追求两者,而不是在两者之间取舍。

Revenue-focused agent strategies have risen since Q1, while efficiency-focused strategies have declined, and most organizations now pursue the two together rather than trading one off against the other.

因此,随着他们对AI用例思考方式的成熟,成本管理也在成熟,这是有道理的。

And it makes sense then that alongside that maturation of the way that they think about AI use cases, there's also a maturation of cost management.

KPMG将其描述为从成本控制转向从AI经济管理的角度来思考问题。

KPMG characterizes it as moving from cost control to thinking about things in terms of AI economic management.

他们写道,成本可见性已经非常普遍。

Cost visibility, they write, is widespread.

将其与价值挂钩是下一步。

Linking it to value is the next step.

对于那些还在实验的组织来说,他们显然还没有将其与价值挂钩。

For those organizations that are experimenting, they're obviously not linking it to value yet.

这就是为什么他们被归类为尚未确立ROI的原因。

That's the whole point about why they're identified as pre-established ROI.

但对于那些已经确立ROI的组织,48%报告他们持续评估价值与成本的对比。

But for those organizations that have established ROI, 48% report that they consistently assess value against cost.

在实验组织中,已经有43%拥有AI成本监控仪表板,而在已确立ROI的组织中,这一比例跃升至77%。

Among the experimenting organizations, already 43% have AI cost monitoring dashboards, a number that jumps to 77% among the established ROI orgs.

在AI审批环节,50%的实验企业有成本审查,而已确立ROI的企业中这一比例为73%。

At the point of AI approvals, 50% of experimenting firms have a cost review compared to 73% with established ROI.

使用量或token预算也开始出现。

And usage or token budgets are also showing up.

31%的实验组织有这样的预算。

31% Of the experimenting organizations have them.

而已确立ROI的组织中,46%也有某种形式的使用量或token预算。

With 46% of the established ROI organizations having some sort of usage or token budget as well.

当然,我一直在强调的一点是,在实验阶段对使用量或token预算过于严格可能会相当限制发展。

One of my big soapboxes is, of course, that I think that being overly restrictive with usage or token budgets when you are in that experimentation phase can be fairly limiting.

但当然,如果不了解具体的项目,很难确切知道人们所说的使用量或token预算到底是什么。

But of course, it's hard to tell exactly what people are considering usage or token budgets without seeing the individual programs.

总的来说,这项调查讲述了企业采用AI的故事,一个组织日益成熟的故事。

Overall, the survey tells the story of AI enterprise adoption as one of increasing organizational maturity.

管理AI的经济性,构建更好的AI监控系统,思考主权和多模态架构等问题。

Managing the economics of AI, building better systems for monitoring AI, thinking about questions like sovereignty and multimodal architectures.

如果你回到一年前,即2025年Q3的脉搏调查,大部分这些考量甚至都不在雷达上。

If you go back a year ago to the Q3 2025 Pulse survey, most of these considerations weren't even on the radar.

KPMG当时还在问诸如有多少组织尝试过使用智能体之类的问题。

KPMG was still asking things back then like what percentage of organizations had even tried an agent.

所有这些都属于关键的管理决策层面,我认为这对整个行业来说是非常积极的。

Now all of these things are mission-critical management decisions, and I think that's extremely positive for the industry as a whole.

另外,对于那些担心注重成本效率会导致投入减少的人,毕马威发现,未来12个月计划的人工智能投资平均值从第一季度的1.86亿美元跃升至第三季度的2.1亿美元。

And by the way, for those who are worried that the focus on cost efficiency is going to lead to decreased spend, KPMG found that the average planned AI investment over the next 12 months jumped from $186 million in Q1 to $210 million in Q3.

我个人认为,我们在智能方面的最终投入,目前还只是刚刚触及表面。

I personally think that we are still barely scratching the surface of what we will ultimately spend on intelligence.

那么,今天就到这里,以上就是今天的AI Daily Brief。

But for now, that's going to do it for today's AI Daily Brief.

一如既往,感谢你的收听或观看。

Appreciate you listening or watching as always.

下次见,保重。

And until next time, peace.
已剔除 4 处广告(点击展开查看)
M1
M10:40广告 · 已剔除

First of all, thank you to today's sponsors, KPMG, Section, Harbor, and Granola.

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

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

M1
M111:29广告 · 已剔除

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

Here's a harsh truth.

Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized.

Half of companies have AI tools, but only 12% use them for business value.

Most employees are still using AI to summarize meeting notes.

If you're the one responsible for AI adoption at your company, you need Section.

Section is a platform that helps you manage AI transformation across your entire organization.

It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value.

The result?

You go from rolling out tools to driving measurable AI value.

Your employees move from meeting summaries to solving actual business problems.

And you can prove the ROI.

Stop guessing if your AI investment is working.

Check out Section at sectionai.com.

That's S-E-C-T-I-O-N-A-I dot com.

M1
M112:46广告 · 已剔除

Harbor Capital Advisors' AI Lab Ecosystem ETF suite lets you invest in the ecosystem behind the AI lab you believe in.

Search Harbor AI Lab Ecosystem ETFs wherever you invest or follow @HarborCapital on X to learn more.

Visit harborcapital.com for a prospectus containing investment objectives, risks, fees, expenses, and other important information.

Read and consider it carefully before investing.

Risks include principal loss and artificial intelligence-related risks.

Harbor ETFs are distributed by Foresight Fund Services, LLC.

Harbor is not affiliated with AI Daily Brief, and the funds are not affiliated with, sponsored by, or endorsed by any AI lab.

This is a paid advertisement and not personalized investment advice.

Investing involves risk, including possible loss of principal.

M1
M113:35广告 · 已剔除

And that's where Granola comes in.

Granola is an AI-powered notepad that captures what happens in your meetings and turns it into clean, structured notes with the decisions and action items pulled out and easy to find.

There's no setup and no configuration.

It just fits into how you already work.

For me, it means I get to stay in idea mode and Granola makes sure those ideas actually become action.

Once you try Granola on a first meeting, it is hard to go without.

You can try it totally free at granola.ai/brief.

That's granola.ai/brief to get your time back.