← 返回任务列表

What Anthropic’s IPO Filing Reveals

275 段 · 1 位说话人 · 原片 28:32
M1
M10:00

嗯,事情真是越来越精彩了。

Well, it just keeps getting more exciting.

Anthropic的IPO越来越近了,这周我们拿到了完整的S-1招股说明书。

The Anthropic IPO is inching ever closer, and this week we got the full S-1 prospectus.

过去24小时,所有人都在仔细研读这份400页文件的每一个细节,看看有没有什么意外情况。

Everyone has spent the last 24 hours poring over every detail of the 400-page document, looking to see if there were any surprises coming their way.

总的来说,我们看到的是一个彻底重新定义公司扩张速度的商业图景。

Overall, what we got is a picture of a business that completely redefines how fast a company can scale.

不过最终大家对它的看法不光取决于它现在的表现,还取决于它未来几年的承诺。

Although ultimately what people think of it isn't just going to be about how it's doing now, but also the commitments it's made for the next few years.

我们会深入探讨所有这些,以及更多,全面分析Anthropic S-1中的重要内容。

We're going to get into all that and more as we look at everything important in the Anthropic S-1.

《AI每日简报》是一个每日更新的播客和视频节目,专注于人工智能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.
M1
M11:34

在上个月发布了100多个数学证明之后,OpenAI又在数学、计算机科学和物理学领域增加了三项重大发现。

After releasing more than 100 mathematical proofs last month, OpenAI has added 3 new significant discoveries across mathematics, computer science, and physics.

在数学领域,AI模型证明了一个1991年提出的局部平滑猜想,并将其推广到了所有维度的一般情况。

In mathematics, AI models proved a local smoothing conjecture proposed in 1991 as a general case across all dimensions.

这个猜想认为,聚焦波形会产生一个单一的、短暂的峰值。

This conjecture proposes that focusing waveforms results in a single, short-lived peak.

Guth、Wang和Zhang在2019年证明了二维情况下的猜想,而Wang和Zahl在2025年证明了三维情况下的相关Kakeya猜想。

Guth, Wang, and Zhang proved the conjecture in 2 dimensions in 2019, while Wang and Zahl proved the associated Kakeya conjecture in 3 dimensions in 2025.

OpenAI的结果将证明推广到了所有维度。

OpenAI's result generalizes the proof to every dimension.

在计算机科学领域,OpenAI推翻了指数时间假说,也就是复杂性理论中常说的ETH。

In computer science, OpenAI disproved the Exponential Time Hypothesis, known in the field of complexity theory as ETH.

这个假说与更著名的P与NP问题密切相关,都涉及计算复杂性。

This hypothesis is deeply related to the more well-known P versus NP problem, dealing with computational complexity.

该假说认为,对于一种叫3SAT的算法,增加变量必然会导致计算时间呈指数级增长。

It suggests that for an algorithm called 3SAT, adding variables necessarily increases time to compute exponentially.

这意味着对某些类型的问题,不存在计算捷径。

That would mean that for certain classes of problems, no computational shortcut exists.

这个假说几十年来一直被认为是正确的,所以推翻它打破了计算科学的核心假设之一。

This hypothesis was assumed to be true for decades, so disproving it breaks one of the core assumptions in computational science.

最后,在物理学领域,OpenAI部分证明了在四维空间中存在某些包含规范场和费米子的相互作用量子场论。

Finally, in physics, OpenAI is partway to an existence proof of certain interacting quantum field theories containing gauge fields and fermions in 4 dimensions.

这项工作旨在构建一个统一标准模型的简化版本,该模型将引力、电磁力和亚原子力结合在一起。

This work aims to produce a simplified version of a unified standard model, which combines gravitational forces with electromagnetism and subatomic forces.

其中包含了对纯杨-米尔斯理论存在性的证明,这是七大千禧年难题之一。

It contains a proof of the existence of pure Yang-Mills theory, which is one of the 7 Millennium Prize problems.

这些结果都是由OpenAI的内部模型发现的。

The results were each found by OpenAI's internal models.

对于我们这些可能觉得这是《AI每日简报》有史以来最难理解的一段内容的人来说,我们只能听听旁观者的意见,看看这些领域的专家觉得哪些结果最重要。

And for those of us for whom that was maybe the least understandable segment on the AI Daily Brief ever, all we can do is turn to the peanut gallery to get a sense from people who are actually in these fields which of them matter most.

简而言之,虽然每个结果在其领域都极其重要,但推翻ETH在X平台上引发的反响最为强烈。

The short is that while each of those results is incredibly important in their field, disproving ETH drew the strongest response on X.

数学教授、菲尔兹奖得主Timothy Gowers写道,对于不了解的人来说,ETH就是那个认为3SAT需要指数时间的假说。

Mathematics professor and Fields Medalist Timothy Gowers wrote, for those who don't know, ETH is the hypothesis that 3SAT needs exponential time.

它竟然是错误的,这非常令人惊讶且影响深远。

Very surprising and consequential that it is false.

A16z的Martin Casado评论说,哇,ETH是错的,这个结果太惊人了。

A16z's Martin Casado commented, wow, ETH being false is a stunning result.

很多人记得在大学里学到ETH时,它被当作一个可以信赖的坚实假设,所以看到它被推翻,影响非常大。

Many remembered learning about ETH in college as a rock-solid assumption they could rely on, so to see it disproved is very impactful.

实际上,这意味着一些加密算法可能存在比暴力破解更快的捷径,这让我对该领域的一个关键假设产生了质疑。

Practically, this means that some cryptographic algorithms may have shortcuts that are faster than brute-force guessing, and that calls into question one of the key assumptions in the field.

顺便说个有意思的事,我本来想说它是“承重墙式的假设”,但考虑到“承重墙”这个词在AI领域被用得太滥了,我就换成了不那么夸张的“关键假设”。

As a fun aside, I was literally about to call it a load-bearing assumption, but given how overused load-bearing is in AI right now, I went instead for the much less impressive key assumption.

我们真是活在一个奇怪的时代。

Weird times we're living in.

总之,Anthropic 研究员 Levendopogee 指出,Anthropic 大约在两周前也独立得出了关于 ETH 的相同结论。

In any case, Anthropic researcher Levendopogee noted that Anthropic had independently come to the same conclusion about ETH about 2 weeks ago.

数学家兼创始人 Conrad Wolfram 思考了更深远的影响,写道:这个复杂性理论的结果其实非常巨大。

Mathematician and founder Conrad Wolfram considered the further implications, writing, the complexity theory result is actually huge.

如果能通过同行评审,这可能会是几十年来计算机科学领域最大的突破之一。

If it survives peer review, this could be one of the biggest computer science breakthroughs in decades.

OpenAI 的 Greg Brockman 呼吁大家先冷静一下,评论说:P 仍然很可能不等于 NP。

OpenAI's Greg Brockman urged everyone to take a breath, commenting, P still probably doesn't equal NP.

接下来,故事的另一个重要部分,就是 OpenAI 是怎么推进这些发布的。

Now, the other big part of the story was how OpenAI went about these releases.

上个月大规模公开数学成果之后,OpenAI 遭受了大量批评,让很多数学家感觉自己像被外部力量剥夺了专业主导权。

The company caught a lot of criticism after last month's giant math dump, leaving many mathematicians feeling like they were becoming dispossessed by an outside force.

更实际地看,这些结果带来了巨大的验证难题,数学家们很难推进对这些生成证明的确认。

More practically, the results produced a giant verification problem, with mathematicians struggling to make progress on confirming the generated proofs.

这次,OpenAI 决定与专家科学家更紧密地合作。

This time, OpenAI decided to partner more closely with expert scientists.

数学结果已完整发布,包括完整的提示词和 AI 输出,这是按照现有数学家理事会的建议来做的。

The mathematics result has been released in full, with complete prompts and AI outputs, as advised by the existing Council of Mathematicians.

对于计算机科学的结果,OpenAI 选择公开结果,但同时与一组科学家合作,对其进行进一步解释。

For the computer science result, OpenAI is making the results public but is working with a group of scientists to explain it further.

对于物理结果,OpenAI 则保持不公开,因为这些结果还不完整。

For the physics result, OpenAI is keeping the results private as they are incomplete.

不过,他们正与诺贝尔奖得主 David Gross 领导的物理学家团队密切合作,继续推进这项工作。

However, they are working closely with a group of physicists led by Nobel laureate David Gross to continue the work.

OpenAI 表示,他们不追求学术署名,并坚持人类科学家应因推进结果所做出的工作而获得唯一的认可。

OpenAI noted that they are seeking no academic credit and insist that the human scientists get sole recognition for the work they do to advance the result.

这并不是 OpenAI 昨天唯一的新闻。

That wasn't the only news out of OpenAI yesterday.

该公司发布了新模型 GPT-6.1 Terra,但很多人都在想,这个中等大小的模型在系列里到底有没有位置。

The company has released a new model, GPT-6.1 Terra, but a lot of folks are wondering if this mid-sized model actually has a place in the lineup.

在 GPT-5.6 系列推出时,Terra 在有些人眼里就是那个被冷落的模型,旁边还有 Sol 和 Luna。

During the rollout of the GPT-5.6 family, Terra felt to some like the orphaned model alongside Sol and Luna.

Sol 是处理难题的高性能模型,而 Luna 则是处理简单问题的更廉价模型。

Sol was the high-performance model for difficult problems, while Luna was the cheaper model for simpler problems.

很少有人需要一个中间地带的模型——智能只是还行,价格也只是还行。

Few had a use case for a middle-ground model that was only kind of intelligent and kind of cheap.

更糟的是,用户只要调整 SOL 和 LUNA 的 effort 设置,就能在 effort 曲线的每一个点上,以更低的成本获得比 Terra 更强的性能。

To make matters worse, by adjusting the effort settings for SOL and LUNA, users could beat the performance of Terra at a cheaper cost for every point along the effort curve.

到了 6.1 Terra,OpenAI 将其定位为均衡的选择。

With 6.1 Terra, OpenAI is presenting the model as the balanced option.

他们写道:Terra 继续推动我们模型系列中成本与智能的前沿,为速度、能力和成本提供了新的平衡。

They wrote, Terra continues to advance the cost-intelligence frontier across our model families, providing a new balance of speed, capability, and cost.

而 OpenAI 让这个模型变得有用的主要方式,就是把它变得更便宜。

And primarily, OpenAI has made the model useful by making it cheaper.

6.1 Tera 的定价是 5.6 Tera 的 50%,这恰好也是 SOL 定价的一半。

6.1 Tera is available at 50% the cost of 5.6 Tera, which also happens to be half the cost of SOL.

Tera 在计算机使用方面似乎也极其擅长。

Tera also seems to have extremely high proficiency in computer use.

在 Agents Last Exam 上——这个测试包含一系列专业任务——Tera 的得分实际上比包括 Astra 在内的其他任何 OpenAI 模型都要略高一点。

On Agents Last Exam, which features a range of professional tasks, Tera actually scored slightly higher than any other OpenAI model, including Astra.

它完成任务的花费大约是 SOL 的一半、Astra 的十分之一,而且速度比这两者都快三倍。

It completed tasks at around half the cost of SOL and one-tenth the cost of Astra, while being 3 times faster than both.

Tera 在这个特定基准上甚至接近 Opus-5.5 的得分,但 token 效率是它的三倍。

Tera even approached Opus-5.5's score on this particular benchmark with 3x token efficiency.

这使得它以九分之一的价格完成基准测试任务。

Allowing it to complete the benchmark tasks at 1/9th the price.

不过,这并不是所有基准测试上的普遍结果。

That wasn't the general result across all benchmarks, however.

在大多数类别中,包括 agentic coding,Tera 的得分略低于 SOL,成本也略低一些。

For most categories, including agentic coding, Tera scored slightly lower than SOL at a slightly reduced cost.

OpenAI表示,该模型在高重复性的自动化任务中表现最好,比如读取收到的发票,然后把详细信息录入会计软件。

OpenAI said that the model performs best in high-volume automated tasks like reading incoming invoices and entering their details into accounting software.

换句话说,这些任务价值相对较低,不值得动用SOL,但又需要较强的电脑操作能力,以及比Luna能提供的稍高一点的智能。

In other words, relatively low-value tasks that don't justify the use of SOL, but still require strong computer use skills and a little more intelligence than Luna can offer.

来自Artificial Analysis的测试得分为47分,比GPT-6.1 SOL低4分,比Grok 4.7高2分。

Testing from artificial analysis resulted in a score of 47, 4 points shy of GPT-6.1 SOL and 2 points ahead of Grok 4.7.

在最高设置下,他们发现Tera的单任务成本与GPT-6.1 SOL在超高设置下的成本接近,所以可能不值得使用Tera。

On max settings, they found that Tera's cost per task was similar to GPT-6.1 SOL on extra-high settings, so likely didn't justify using Tera.

他们评论说,低强度和中强度设置在成本效率上是帕累托最优的,而其他强度设置则落在前沿之后。

They commented, the low and medium effort levels are Pareto optimal for cost efficiency, while the other effort levels sit behind the frontier.

其中一个有趣的结果是,在新的AA Coding Agent Index上表现非常扎实。

One of the interesting results was a very solid performance on the new AA Coding Agent Index.

Tera得分48,刚好排在GLM-53之后,略领先于DeepSeek V4.1 Flash。

Tera scored 48, placing it just behind GLM-53 and just ahead of DeepSeek V4.1 Flash.

所以我们试着理解一下,虽然最初的Tera版本同时存在不够好也不够便宜的问题,但这一版本确实改进了不少,至少对于一部分日常主力类型的任务来说,已经足够好了。

So trying to wrap our heads around this, while the original version of Tera suffered from being not quite good enough and not quite cheap enough at the same time, this iteration does seem to have improved enough to become good enough for at least some volume of daily driver type tasks.

不用看评测了,你需要知道的都在这里。

No need to watch a review, here's everything you need to know.

它很快,写得好,现在能全天操作我的电脑了。

It's fast, Writes well, runs my computer all day now.

对于Codex来说,是很棒的默认选择。

Great default for codecs.

事实上,“现在能全天操作我的电脑”这一点,正是大多数人认为这个模型原本就应该做的事。

And in fact, that runs my computer all day now is the thing that most people assume that this model is meant for.

个人智能体的兴起,为Tera提供了一个潜在的细分市场,那就是常驻型智能体,而这个市场在最初发布这类模型时并不真正存在。

The rise of personal agents gives Tera a potential niche for always-on agents that didn't really exist when the model class was first released.

换句话说,现在有大量任务,你可能需要比Luna更强大一点,但比Sol更划算、更快的模型。

In other words, there's now a large category of tasks where you might want something a little more powerful than Luna, but more cost-efficient and faster than Sol.

Peter Yang问他的观众,现在Tera应该取代SOL成为Codex的默认模型吗?

Peter Yang asked his audience, Should Tera be the default for Codex now instead of SOL?

开发者Rohan Hyman似乎总结了早期的共识,他评论说:“如果你是在写代码,那可能不应该。”

Developer Rohan Hyman seems to have summed up the early consensus, commenting, "if you're producing code, probably not."

如果是操作电脑的话,那可能应该。

For computer use, probably yes.

在我们最近发布的这些模型中,这个模型最清楚地印证了一个观点:新模型发布在很大程度上不再是简单的一对一替代,而是要思考它们在哪个位置最合适。

Of all the model releases we've had recently, this is the one that most clearly reaffirms the idea that new model releases are no longer about 1-to-1 replacements in most cases, but instead about asking where they fit in.

有些类型的任务,尤其是长期运行的电脑操作类任务,现在这个模型已经足够好了,能给你提供更高效、更具成本效益的选择。

There are types of tasks, especially long-running computer use types of tasks, where this is now good enough and is going to give you more efficient and cost-effective options.

这可能不令人兴奋,但它可能非常有用。

That might not be exciting, but it could be very useful.

我肯定不会说结论已经出来了,因为你能看到各种各样的观点,所以比平时更建议你实际在自己的使用场景或基准测试中试试,看看它是否能为你带来价值,以及在哪些方面有价值。

I would certainly not say that the verdict is in on this one yet, as you can find opinions all across the spectrum, so even more than normal I would recommend you actually put this one through your own personal use cases or benchmarks to see if and where it can be valuable for you.

最后今天,一个引发很多讨论的消息——Cohere在测试一个流行的中国开源模型时,发现了一些令人担忧的行为。

Lastly today, a story that's generating a lot of chatter—Cohere has discovered some concerning behavior while testing a popular Chinese open model.

该公司写道,在最近的一次网络能力测试中,研究团队发现一个阿里巴巴的Qwen 3.8模型做出了绝不应该发生的事情。

During a recent cyber capabilities test, writes the company, the research team caught an Alibaba Qwen 3.8 model doing something that should never happen.

他们接着说,在被要求编写一个防御性检测规则时,Qwen 3.8 FlashNext无视指令,清点了运行它的机器信息,然后试图把一个探测文件上传到北京的一个阿里云存储桶中。

They continued, asked to write a defensive detection rule, Qwen 3.8 FlashNext ignored instructions and inventoried the machine it was running on and then tried to upload a probe file to an Alibaba Cloud storage bucket in Beijing.

Cohere表示他们的沙箱阻止了这一行为,但也许最令人担忧的是,模型的推理轨迹显示不会调用任何工具,而且完全没有提示正在进行的操作。

Now Cohere said that their sandbox blocked the action, but perhaps most concerningly, the model's reasoning trace said that no tools would be called and gave no indication of the actions being taken.

许多安全团队把推理轨迹作为其可观测性模型的核心部分,所以这种行为很容易逃避检测。

Many security teams use reasoning traces as a core part of their observability model, so this action could have easily evaded detection.

Cohere利用这一事件来挑战一种想法,即本地托管模型就能消除所有数据泄露的风险。

Cohere used the incident to challenge the idea that hosting a model locally removes all risk of data exfiltration.

他们写道,一个令人安心的想法影响了公司和政府承包商采用开放权重AI模型的方式——任何人都可以下载并运行这些模型。

They wrote, one comforting idea has shaped how companies and government contractors adopt open-weight AI models that anyone can download and run.

我们的想法是,如果我们自己在服务器上运行这些权重,数据就无法离开大楼,系统就是安全的。

If we run the weights on our own servers, the thinking goes, our data can't leave the building and our systems are secure.

然而,Cohere认为,既然AI模型现在是在采取行动,而不仅仅是在回答问题,威胁态势已经改变了。

Now that AI models are taking actions rather than just answering questions, however, Cohere argues that the threat profile has shifted.

他们写道,一旦模型能够输入命令行,训练期间嵌入其权重中的任何内容,就等于在你的网络内部获得了一个键盘。

They wrote, once a model can type into a command line, whatever was baked into its weights during training gets a keyboard inside your network.

一些安全专家并不同意这个结论——事实上,他们甚至称这份报告完全是胡说八道。

Some security professionals didn't agree with that conclusion—in fact, going so far as to call this report total BS.

他们指出,模型可能是在通过 ping 北京的服务器来测试自己编写的检测规则,这很可能是权重中嵌入的默认设置。

They pointed out that the model could have been testing the detection rule it had written by pinging the Beijing servers, likely a default setting embedded in the weights.

此外,他们指出,Cohere 在运行评估时观察到这种行为的概率低于十万分之一,这表明它并非恶意行为。

And furthermore, they pointed out that Cohere had only been able to observe this behavior less than 1 out of 100,000 times running the evaluation, suggesting that that means it isn't a malicious action.

但话又说回来,Cohere 指出,如果你每周运行一千万次 agent 调用,就可能出现数百次这样的工具调用,而只要一次就足以识别你的网络和你运行的 agent。

Then again, Cohere pointed out, if you're running 10 million agent turns a week, hundreds of these tool calls could occur, and just one would be enough to identify your network and the agent you're running.

在质疑之余,也有许多人呼吁 Cohere 公布完整的提示词和推理追踪,以便进一步调查。

Now, alongside the skepticism, there were many calls for Cohere to release the full prompts and reasoning traces for further investigation.

在 Cohere 公布之后,AI 开发者 Mia 分析了这些内容并写道:TLDR,确实是异常,但影响被夸大了。

After they did, AI developer Mia analyzed them and wrote, TLDR: real anomaly, overstated implications.

关于意图的说法没有任何依据。

The claims about intent have nothing behind them.

意外且可能不安全的工具行为确实存在,但这并非蓄意意图、植入的后门,或者由阿里巴巴指使的泄露机制。

Unexpected and potentially unsafe tool behavior, yes, but not a deliberate intent, a planted backdoor, or an Alibaba-directed exfiltration mechanism.

尽管如此,无论意图为何,一个中国模型能够未经请求就“回传”数据,这对企业用户来说显然是一个巨大的问题。

Still, whatever the intent may be, the idea that a Chinese model could call home without being asked is obviously a massive problem for enterprise users.

其他人也在进行无关的网络研究时报告了这种行为,因此在一些人看来,这有可能是一种低发生率的监控技术。

Others have reported the behavior while doing unrelated web research, so to some, there's a possibility that this is a low-occurrence surveillance technique.

由于主权方面的考虑,各国政府也在考虑采用开放模型,因此这种行为在这些场景中同样可能是个大问题。

Open models are also being considered by governments for sovereignty reasons, so this behavior could be a huge problem in those settings as well.

有人指出,Cohere 正在与中国的模型竞争这些政府合同,因此在散布安全疑虑方面有其自身利益。

Some noted that Cohere is competing with Chinese models for these government contracts, so has a vested interest in spreading security doubts.

尽管如此,AI 新闻聚合者 Andrew Curran 评论说,这类事情有潜在可能推动新法律的通过。

Still, AI news aggregator Andrew Curran commented, this is the type of thing that potentially gets new laws passed.

为了强调这一点,美国国务院副国务卿 Sarah B. Rogers 转推了 Curran 的推文,并补充说,构建安全可信的技术栈是一个至关重要的政策优先事项,这并非没有道理。

Reinforcing that point, State Department Undersecretary Sarah B. Rogers retweeted Curran, adding, a secure, trusted tech stack is a vital policy priority with good reason.

生活从不无聊,但今天的热点新闻就先到这里。

Never a dull moment, but for now that is going to do it for the headlines.

接下来,进入本期主节目。

Next up, the main episode.

毕马威(KPMG)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,发现表现最出色的人通过引导、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,也就是所谓的AI amplifiers,他们的特色不在于单靠知识储备,而在于他们如何与AI协同工作。

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AI amplifiers与其他人有何不同,请访问kpmg.com/us/ai。

Learn more about what separates AI amplifiers from everyone else at kpmg.com/us/ai.
M1
M112:52

正在部署 AI 的公司不断遭遇同样的障碍。

Companies rolling out AI keep hitting the same wall.

Token 开销不断攀升,安全问题在拖 agent 的后腿,而且没有人真正知道 AI 的使用情况。

Token spend is climbing, security is holding agents back, and no one knows how AI is truly being used.

根本原因在于:AI 与你的业务所依赖的系统和上下文是割裂的。

The root cause: AI is cut off from the systems and context your business runs on.
M1
M113:34

你整天都在一个接一个地开会。

You're in back-to-back meetings all day.

你努力保持专注,但同时也担心自己会忘记那个决定、待办事项,以及重要的下一步。

You're trying to stay present, but you're also worried you'll forget the decision, the action item, the important next step.

这正是我们在《AI Daily Brief》中经常遇到的情况,因为一切发展变化太快,我们也在迅速调整我们所做的每一件事。

This is a scenario that we absolutely live here at the AI Daily Brief, given how fast things move and change and how quickly we're evolving everything that we do.

而这就是 Granola 的用武之地。

That's where Granola comes in.

Granola 是一款专为会议打造的 AI 记事本。

Granola is an AI-powered notepad for meetings.

你像往常一样随手记下粗略的笔记,而在后台,Granola 会进行转录,并在会议结束时把它们变成清晰、有用的笔记。

You jot down rough notes like you always do, and, in the background, Granola transcribes and turns them into clear, useful notes when the meeting ends.

没有机器人加入你的通话,没有干扰,就是一个简洁的记事本,帮你保持专注。

No bots joining your calls, no distractions, just a clean notepad that helps you focus.

通话期间或结束后,你可以和笔记对话,让 Granola 帮你提取行动项、协助谈判、写跟进邮件,甚至用预设提示的模板来指导你。

During or after the call, you can chat with your notes, ask Granola to pull out action items, help you negotiate, write a follow-up email, or even coach you using recipes which are pre-made prompts.

一旦你在第一次会议上试用过,就很难再离开它了。

Once you try it on a first meeting, it is very hard to go without.
M1
M114:30

Blitzie 在写代码之前,会先深入理解你的代码库。

Blitzie deeply understands your codebase before it writes code.

这是第一个体现价值的地方:AI 时代的安全。

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

漏洞不是孤立存在的。

Vulnerabilities don't live in isolation.

它们深藏在数百万行相互关联的代码里,补一个东西,可能悄悄弄坏另外三个。

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

这就是为什么表面扫描会失效。

That's why surface-level scans fail.

Blitzie 从整个应用的知识图谱出发,识别并呈现整个体系中的 CVE,主动推荐补丁,还能执行 PR。

Blitzie 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.

每项修复都基于系统之间的连接和验证方式,确保不会引入新问题,知识图谱还会动态更新,让你始终领先于不断加速的威胁环境。

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

有一位 Blitzie 客户在 4 天内解决了 6 个核心微服务中的 21 个活跃 CVE。

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

零编译错误,所有验证扫描干净通过,计划几个月的工作,不到一周就修完了。

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

基于真实架构背景的安全修复,以计算速度推进。

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

访问 blitzy.com,加固你的代码库。

Harden your codebase at blitzy.com.

网址是 B-L-I-T-Z-Y.com。

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

欢迎回到 AI Daily Brief。

Welcome back to the AI Daily Brief.

经过数月的等待,以及陆续向媒体泄露的消息,Anthropic 终于公开了他们的 S-1 文件,正在做上市前的最后准备。

After months of waiting and a steady trickle of leaks to the press, Anthropic has made their S-1 public as they make final preparations to go public.

我们现在有了所有细节、所有数字、所有你能想到的风险披露。

We now have all the details, all the numbers, all the risk disclosures you could possibly want.

今天,我们要来聊聊这些内容,看看解读中最主要的共识和分歧在哪里,以及这一切对这家公司未来几周的前景意味着什么。

Today, we are going to talk through that, look at where the biggest areas of consensus and disagreement are in the interpretation, and see what it all says for the likely prospects of the company heading into the next couple of weeks.

Every 的 Dan Shipper 开玩笑写道:“大新闻:Anthropic 的 S-1 并没有在第一段说‘它可能会杀死/让所有人失业’。”

Every's Dan Shipper jokingly wrote, “Big news: the Anthropic S-1 doesn't say, ‘or it might kill/unemploy everyone,’ in the first paragraph.”

这是人类的胜利。

A win for humanity.

这倒不是说它没有一些大胆而重磅的表述。

Which is not to say that it doesn't have some big and bold statements.

如果你听过我那期 How We Got Here,讨论过 Facebook 和 Google 的 IPO 公开信,那么 Anthropic 也为这一传统增添了新内容:“我们相信,AI 是人类历史上最具影响力的技术之一。”

For anyone who listened to my How We Got Here episode that discussed the Facebook and Google IPO letters, Anthropic added to that canon: “We believe that AI is one of the most consequential technologies in human history.”

如果管理得当,它能够迅速加速科学创新,大幅提高经济增长率,让重大疾病的治愈方法成为可能,并解决那些几代人都未能攻克的难题。

If managed well, it could rapidly accelerate scientific innovation, greatly increase economic growth rates, bring cures for major diseases within reach, and solve problems that have resisted generations of human effort.

Anthropic 的目标是在实现这些益处的同时,应对并预防这场如此迅猛、如此强大的技术革命所带来的风险。

Anthropic aims to realize these benefits while navigating and preventing the risks that come with a technological revolution this fast and this powerful.

我们希望每一位投资 Anthropic 的人都能理解 4 件事。

There are 4 things that we would like any investor in Anthropic to understand.

AI 的巨大潜力、我们必须应对的风险、Anthropic 在建立公司和运营方式上的独特方法,以及我们的行为可能不同于纯粹追求利润最大化的公司的地方。

The incredible potential of AI, the risks we must address, Anthropic's unique approach to building a company and operating in the world, and the ways in which our behavior may differ from that of a purely profit-maximizing company.

Anthropic 当然给了自己足够的空间去做这一切。

Anthropic certainly gives themselves space to do all that.

这份文件厚达 400 页,其中超过 90 页专门用来陈述公司可能面临的风险。

The document weighs in at a meaty 400 pages, with over 90 pages just dedicated to the risks that could face the company.

对于那些不熟悉 S-1 的人来说,它也叫招股说明书,这份文件的目的就是向潜在投资者全面披露新股票的相关信息。

For those unfamiliar with an S-1, also known as the prospectus, this is a document that is designed to give potential investors full disclosure about a new stock.

正在看这期节目的朋友们,屏幕上展示的可视化内容,也会在网站配套的 aidailybrief.ai 上同步提供。

For those of you who are watching this episode, the visualization that you see on screen will also be available on the website companion at aidailybrief.ai.

那话不多说,咱们先从数字开始看。

But with that said, let's start with the numbers.

Anthropic 今年上半年的营收是 163 亿美元。

Anthropic earned $16.3 billion in revenue across the first half of the year.

相比之下,2025 年全年只有 46 亿美元。

That's up from $4.6 billion for all of 2025.

据称,Anthropic 告诉投资者说,第三季度他们赚了 168 亿美元,所以这种惊人的增长势头似乎还在延续。

Anthropic reportedly told investors that they earned $16.8 billion in Q3, so that incredible growth seems to be continuing.

不过呢,在 S-1 文件里我们并没有看到第三季度数字的确认。

That said, we didn't get any confirmation of Q3 numbers in the S-1.

那栏就是空白的。

With that line left blank.

Anthropic 声称第二季度实现了 9.74 亿美元的营业利润,这是他们首次实现季度盈利。

Anthropic claimed a $974 million operating profit in Q2, which was their first profitable quarter.

这是按照 GAAP 会计准则计算的,所以我们可以认为这是真实的。

That's under GAAP accounting standards, so we can take that as the real deal.

上半年的毛利率算下来是 85%,而 2025 年全年只有 40%,说明利润率在提高。

Gross operating margins were calculated at 85% for the first half of the year, up from 40% across 2025, meaning margin is increasing.

不过呢,研发、销售和行政费用也在同步上涨,导致营业利润率是 -4.3%。

However, research, sales, and administrative costs are rising alongside, leading to an operating margin of -4.3%.

S-1 文件还证实了上半年那吓人的 800 亿美元净亏损。

The S-1 also confirmed that scary $80 billion net loss for the first half of this year.

但我上周提过,其中 720 亿美元是因为可转换债务的会计处理变更。

However, as I mentioned last week, $72 billion of that was an accounting change related to convertible debt.

基本上,这只是因为股价上涨造成的账面亏损,并不是实际的现金损失。

Essentially, this is just a paper loss caused by stock price appreciation, not a cash loss.

之前没公布的一个数字是 Anthropic 的计算开销。

One number that wasn't previously reported was Anthropic's compute spend.

自 2023 年以来,他们在计算资源上总共花费了 192 亿美元。

They've spent a total of $19.2 billion on compute since 2023.

今年上半年的计算花费是 89 亿美元,而且 Anthropic 大约 70% 的计算预算都用于内部工作,而不是用来服务客户。

$8.9 billion in compute spending came in the first half of this year, and Anthropic is spending around 70% of their compute budget on internal work rather than serving it to customers.

Anthropic 说他们 90% 的代码现在是由 Claude 编写的,按照 API 价格计算,内部使用 Claude 的成本在今年 4 月份已经超过了工资支出。

Anthropic said that 90% of their code is now written by Claude, and internal Claude use valued at API prices surpassed payroll as an expense in April.

大家特别关注的一个数字是,Anthropic 对未来十年的计算资源做出了数千亿美元的承诺支出。

The numbers that are getting a lot of attention is that Anthropic has made hundreds of billions of dollars in future compute spend commitments over the coming decade.

其中大约 4170 亿到 4310 亿美元是不可取消的,包括对 Google、Amazon、Microsoft 和 Broadcom 的主要承诺。

Somewhere between $417 and $431 billion of that is non-cancelable, including major commitments to Google, Amazon, Microsoft, and Broadcom.

签署的总承诺金额超过 6000 亿美元。

The total signed commitments are above $600 billion.

而且估计承诺总额可能达到 8000 亿美元。

And estimates could reach $800 billion in commitments.

现在很多人看到这些大数字都很震惊,但别忘了这些承诺是分布在十年以上的,有些合同长达 20 年。

Now, most people are reacting to those big headline numbers, but remember that these commitments are spread out over more than a decade, with some contracts running 20 years.

如果这么算的话,差不多每年 200 到 400 亿美元,数字虽然还是很高,但更容易接受了,特别是考虑到 Anthropic 现在赚了多少钱。

If you think about it then as closer to $20 to $40 billion a year, the numbers are still high, but a lot more palatable, especially given how much money Anthropic is making.

一个有意思的意外点是,Anthropic 账上有大量现金,确切说是 1320 亿美元的现金及现金等价物。

An interesting curveball is that Anthropic has a ton of cash in the bank, $132 billion to be exact, in cash and cash equivalents.

公司历史上总共融资了 1410 亿美元,所以基本上几乎所有的钱都还躺在银行里赚利息。

The company has raised a total of $141 billion throughout their history, so almost all of it is still sitting in the bank earning interest.

利息收入相当可观,上半年达到了 11 亿美元,约占整体营收的 6.5%。

And that interest income is pretty significant at $1.1 billion for the first half of the year, representing around 6.5% of total revenue.

SemiAnalysis 发了一个帖子,说这很可能是资本主义历史上最高效的业务规模化案例,一下子火了。

SemiAnalysis went viral with a post that called this likely the most efficient scaling of a business in the history of capitalism.

S-1 文件还提供了大量关于 Anthropic 客户群体的细节。

The S-1 also provided a ton of details on what Anthropic's customer base looks like.

收入质量一直是这份文件受质疑的一个大问题,对很多人来说,Anthropic 已经充分回应了这些担忧。

Revenue quality has been one of the big questions hanging over the filing, and for many, Anthropic has adequately addressed those concerns.

公司报告说,截至九月,他们有超过11,900个客户,年消费额在10万美元以上。

The company reported that as of September, they have over 11,900 customers with $100,000 or more in annual spend.

超过1,900个客户年消费超过100万美元,Anthropic大约有250个客户每年消费超过1000万美元。

More than 1,900 customers are spending more than $1 million, and Anthropic has around 250 customers spending $10 million plus annually.

截至九月,净收入留存率是900%,也就是说,平均每个客户的花费比去年同期增长了9倍。

Net revenue retention was at 900% as of September, meaning that the average customer has 9x'd their spend since this time last year.

要理解业务变化,一个更有用的方式是:去年十二月,Anthropic大约有1,500个企业客户年消费超过10万美元,而现在已经跃升到接近12,000个。

A more useful way to understand the shift in the business is that, as of last December, Anthropic had around 1,500 enterprise customers spending over $100,000 annually, and that is the count that has now jumped to just under 12,000.

不过,Anthropic指出,财富500强客户的开支中,只有不到0.1%是花在Anthropic的服务上,这意味着他们才刚刚开始进入企业IT预算,还有更多预算可以争取。

Still, Anthropic noted that less than 0.1% of a Fortune 500 customer's expense goes towards Anthropic services, the implication being that they've only just started to eat into corporate IT budgets, and there's a whole lot more budget to consume.

Fox的Aaron Levy写道:虽然这个数字最终可能会趋于稳定,但没有理由不能比现在高10到50倍,达到开支的1%到5%,而这仍然只是大多数初创公司花费的一小部分。

Fox's Aaron Levy wrote: While that number likely will eventually asymptote somewhere, there's no reason it can't be 10 to 50x higher than this, reaching a mere 1 to 5% of spend, which would still be a fraction of what most startups are spending.

所以,你可以看到,AI的前沿用户和世界其他地方之间,仍然存在巨大的鸿沟。

Thus, you can see how big a divide there is between the frontier users of AI and the rest of the world still.

Anthropic的收入来源也变得更加多元化。

Anthropic has also seen their revenue diversity broaden out.

今年第一季度末,他们大约36%的收入来自前五大客户。

At the end of Q1 this year, around 36% of their revenue came from their top 5 customers.

这个数字现在降到了21%。

That number is now down to 21%.

S-1文件还披露了通过Google和AWS等分销商获得的合作伙伴份额。

The S-1 also disclosed the exact partner share through distributors like Google and AWS.

Anthropic有48%的收入是通过这些合作伙伴流出的,他们会抽取相当可观的分成,这一点之前没有完全披露。

48% of Anthropic's revenue flows through these partners who take a fairly decent cut, which wasn't fully disclosed before.

现在,谈到风险披露部分,灾难性风险都在里面,还有涉及气候变化和数据中心反弹的披露。

Now, when it comes to the risk disclosure section, the catastrophic risks were all in there, as well as disclosures touching on climate change and the data center backlash.

值得注意的是,这些风险体现在Anthropic认为他们基本上是无法投保的。

Notably, these risks manifested in Anthropic's belief that they're basically uninsurable.

他们写道,没有任何一家保险公司,甚至整个保险行业,可能有足够的能力或专业知识来承保与这种风险敞口相匹配的保险。

They wrote, no single insurer, and potentially the insurance industry in the aggregate, may have sufficient capacity or expertise to underwrite insurance commensurate with such exposure.

不过,更具体地说,Anthropic警告说,收入增长正在放缓,尽管之前的速度非常惊人。

Still, much more tangibly, Anthropic warned that revenue is slowing, albeit from a blistering pace.

Anthropic的收入比去年增长了大约1,000%,但他们表示预计这种情况不会持续。

Anthropic grew revenues by around 1,000% since last year, but said that they do not expect that to continue.

第三季度的环比增长大约为45%,所以仍然非常强劲,但Anthropic说他们基本上不知道未来会以多快的速度增长。

Quarter-on-quarter growth for Q3 was around 45%, so still incredibly strong, but Anthropic said they have basically no idea how fast they'll grow in the future.

普通的资产负债表管理是不能忽视的主要风险之一。

Mundane balance sheet management is one of the major risks that can't be overlooked.

Anthropic指出,存在无法获得足够算力或者算力成本大幅上升的风险。

Anthropic noted that there's a risk that they fail to secure enough compute or that the cost of compute could rise significantly.

他们还指出,像CoreWeave、Google和AWS这样的算力合作伙伴可能并不总是与Anthropic的长期利益完全一致。

They also flagged that compute partners like CoreWeave, Google, and AWS may not always be fully aligned — their terminology — with Anthropic's long-term interests.

合同都签了,但Anthropic提醒投资者,随着AI竞争加剧,这些合同可能不会被履行。

The contracts are in place, but Anthropic is warning investors that they may not be honored as AI competition heats up.

Anthropic说,这就是他们开始建设自己数据中心的原因,以减少对那些可能变成无情竞争对手的合作伙伴的依赖。

Anthropic said that this was the reason they were starting to pursue construction of their own data centers, to reduce reliance on partners that could turn into ruthless competitors.

总的来说,Anthropic的创始人请求投资者对他们高昂的基础设施承诺抱有善意。

Overall, Anthropic's founders asked investors to give them the benefit of the doubt on their sky-high infrastructure commitments.

在S-1附件的一封信中,他们写道,他们的算力支出“乍一看可能显得不负责任地巨大”,但最终他们将其描述为“大胆但理智”。

In a letter attached to the S-1, they wrote that their compute spending, quote, may seem irresponsibly large at first glance, but they ultimately described it as bold but sane.

算力难题自然引出了关于债务融资的警告。

The compute conundrum leads naturally to a warning about debt financing.

Anthropic警告说,他们可能需要比预期更早地寻求融资,以满足基础设施义务。

Anthropic warned that they may need to seek financing sooner than they anticipated to meet infrastructure obligations.

那1320亿美元可能会支撑他们到年底,而且很可能是历史上最大规模的IPO筹资会有所帮助,但看起来他们不太可能完全用收入来为基础设施支出提供资金。

That $132 billion will probably see them through until the end of the year, and what's likely to be the largest IPO fundraise in history will help, but it seems unlikely that they'll be able to fund their infrastructure spend entirely from income.

在这种情况下,Anthropic指出,他们对通往投资级信用评级的路径很少提及。

In that light, Anthropic notes that Anthropic said little about their pathway to an investment-grade credit rating.

在今年早些时候上市后,SpaceX以极微弱的优势获得了投资级评级,达到BBB,这是垃圾级之上的最后一个等级。

After going public earlier this year, SpaceX managed to secure an investment-grade rating by the narrowest of margins, achieving BBB, which is the last ranking above junk.

当时SpaceX并没有盈利,但靠着在卫星互联网和太空发射领域的垄断地位,再加上巨大的股权估值,还是拿到了投资级评级。

SpaceX wasn't profitable at the time, but achieved investment grade through a combination of monopoly power in satellite internet and space launch, as well as a gigantic equity valuation.

换句话说,评级机构在一定程度上认为,SpaceX不盈利是它们主动的选择,想什么时候盈利就能什么时候盈利。

In other words, in part, the ratings agencies viewed SpaceX's lack of profitability as a choice they could turn off whenever they wanted to.

但这些因素里有多少适用于Anthropic,还不太清楚。

It's unclear how many of those factors hold true for Anthropic.

要说他们靠拉个开关就能盈利,可能有点牵强,因为他们的收入很大程度取决于每年训练多个新模型的情况。

It's probably a stretch to say they could turn a profit by flicking a switch, given how much revenue is contingent on training multiple new models every year.

总结一下,Impax资产管理公司的投资级组合经理Tony Zrinka说,引用原话:“Anthropic的现状是,流动性非常强,但未来的义务也非同寻常。”

Summing it up, Tony Zrinka, an investment-grade portfolio manager at Impax Asset Management, said, quote, the state of Anthropic is that they have very strong liquidity and extraordinary future obligations.

以目前的杠杆水平来看,资产负债表足以支撑投资级评级。

The balance sheet supports an investment-grade rating at today's leverage.

不过,最终评级还是取决于评级机构如何看待他们的承诺安排、短暂运营历史,以及循环的交易对手关系。

However, the rating will come down to how the rating agencies view the commitment schedule, short operating history, and circular counterparty relationships.

一个选项是,Anthropic可以继续依赖合作伙伴提供资金,而不是跑去债券市场融资。

One option is that Anthropic can continue to rely on partners for funding rather than running to the bond market.

目前,Anthropic以可转换票据的形式欠Amazon 62亿美元,而且他们还可以选择以额外本票的形式,从Amazon、Broadcom和Google再借770亿美元。

Currently, Anthropic has $6.2 billion in debt owed to Amazon in the form of convertible notes, and they have the option to borrow another $77 billion from Amazon, Broadcom, and Google in the form of additional promissory notes.

他们还能从Morgan Stanley获得150亿美元的循环信贷额度,这是一种“紧急情况下才砸玻璃”的借款方式。

They also have access to a $15 billion revolving credit facility from Morgan Stanley, which is a break-glass-in-case-of-emergency form of borrowing.

但大多数情况下,Anthropic目前是通过第三方工具,与投资级合作伙伴一起为他们的承诺提供资金。

But mostly, Anthropic is currently funding their commitments through third-party vehicles with an investment-grade partner.

他们六月份和Broadcom设立了一个350亿美元的工具,刚刚又和Broadcom签署了一个新的630亿美元的工具。

They established a $35 billion vehicle with Broadcom in June and have just signed a new vehicle for $63 billion, again with Broadcom.

这两笔交易都将为从Broadcom购买TPU提供资金,而且都是与Apollo和Blackstone共同设计的。

Both deals will finance TPU purchases from Broadcom and were structured with Apollo and Blackstone.

今年早些时候,SpaceX把AI行业的可寻址市场定在了28万亿美元,之后很多人都期待Anthropic能给出类似的评估。

Now, after SpaceX set the total addressable market of the AI industry at $28 trillion earlier in the year, many were looking for a similar assessment from Anthropic.

但Anthropic没有直接给出明确的可寻址市场数字,所以我们并没有得到想要的答案。

We didn't quite get that, as Anthropic shied away from giving a clear TAM.

不过,他们确实估计2025年全球知识工作的价值为40万亿美元。

However, they did estimate the value of global knowledge work at $40 trillion in 2025.

他们引用了IDC的预测,认为到2031年,AI投资可能带来累计22.5万亿美元的全球经济影响,而且他们相信实际影响可能更大。

They pointed to an estimate from IDC that suggested that investment in AI could yield a cumulative global economic impact of $22.5 trillion by 2031, saying that they believe that impact could be even greater.

Anthropic还强调,他们的同行很少,只把OpenAI和Google列为直接竞争对手。

Anthropic also highlighted that they have very few peers, naming only OpenAI and Google as direct competition.

而且,为了稍微怼一下OpenAI,Anthropic的创始人承诺他们的产品不会有广告。

And in a slight jab at OpenAI, Anthropic founders pledged to keep their products free of advertising.

在S-1文件里的一封信中,创始人写道:“一旦广告成为公司收入目标的一部分,优化用户参与度的压力就会越来越大。”

In a letter included in the S-1, the founders wrote, "Once advertising becomes part of a company's revenue targets, the pressure to optimize for engagement tends to grow."

我们选择不让这种压力进入Claude。

We have chosen not to introduce that pressure into Claude.

显然,Anthropic更容易抵制这种诱惑,因为他们没有像OpenAI那样庞大的免费用户群,但这依然凸显出他们把一切都押在了企业采用上。

Now, obviously, it's easier for Anthropic to reject that temptation given that they don't have a large customer base of free users like OpenAI, but it still highlights the extent to which they are betting it all on enterprise adoption.

总体来看,这份S-1似乎让华尔街的一些紧张情绪得到了缓解,很多人都在思考这次IPO的历史意义。

Overall, the S-1 seems to have calmed a few nerves on Wall Street, with many reflecting on the historic nature of the IPO.

Gil Luria是D.A. Davidson的技术研究主管,他说:

Gil Luria, the head of technology research at D.A.

“这真的是投资者第一次有重要机会,直接投资于这家推动AI经济发展的公司。”

Davidson, said, "This will really be investors' first significant opportunity to invest directly in the company that is driving the AI economy."

这具有里程碑式的重要意义。

It's of monumental importance.

NewView Capital的Ravi Viswanathan评论说:“从来没有哪次IPO对整个风险投资行业有如此重大的意义。”

Ravi Viswanathan of NewView Capital commented, "Never have IPOs represented so much to the venture capital industry writ large."

就连《金融时报》在公布前发表了一些非常批评性的文章之后,对这份S-1的态度也变得相当中立。

Even the Financial Times was fairly balanced on the S-1 after publishing some highly critical pieces in the lead-up to the public reveal.

不过,评论作者Lex建议投资者不用真的去读文件本身,他说:“实际上,投资者唯一真正的决定就是买还是不买,而且很可能大多数人在本周之前就已经下定决心了。”

Opinion writer Lex advised investors to skip reading the actual document, however, commenting, "In practice, the only real decision for investors is whether to buy or not, and chances are most people will have already made up their minds before this week."

就像六月的SpaceX和未来OpenAI的IPO一样,投资者要么相信公司是美丽新世界里的赢家,要么不相信。

As with SpaceX in June and the future IPO of OpenAI, they either believe the company is a big winner in a brave new world or they don't.

那么,社交媒体上的‘闲聊圈’都在讨论什么呢?

So where was the discussion among our social media chattering class?

很多人对Anthropic手头有多少现金感到非常惊讶且看好。

A lot of folks were very surprised and bullish on how much cash Anthropic had on hand.

以及他们到目前为止的扩张效率有多高。

And how efficiently they'd scaled so far.

当然,最大的担忧,也是人们分歧的关键,就是有多少账单已经签了但还没付。

The big concern, of course, and really the dividing line between different people, was how much of the bill was signed and yet to come.

另一个有趣的讨论是关于更便宜的token和更多收入之间的关系。

Another interesting conversation was around the relationship between cheaper tokens and more revenue.

如果每张GPU的token吞吐量提升足够大,那么即使token价格下降,每GPU的收入也可能增加。

Revenue per GPU can increase as token prices decline if per-GPU token throughput increases enough.

所以,如果GPU价格上涨幅度小于token价格下降幅度乘以每GPU吞吐量的提升,那么Anthropic的收入和利润率都可以随着token价格下降而上升。

So both Anthropic revenue and margins can go up as token prices go down, if GPU prices go up less than the product of token price decline times per-GPU throughput increase.

Tomasz Tunguz给出了具体的数据。

Tomasz Tunguz put it in specific terms.

当Opus 4.5在2025年11月推出,比Opus 4.1能力更强且每任务成本更低时,收入增长从季度环比48%加速到下一个季度的132%。

When Opus 4.5 launched in November 2025 with more capability and a lower cost per task than Opus 4.1, revenue growth accelerated from 48% quarter over quarter to 132% the following quarter.

现实世界中的杰文斯悖论。

Jevons paradox in the real world.

关于创始人控制权有很多讨论,借鉴了像Mark Zuckerberg等人的一些做法。有一种新的创始人F类股票,这意味着尽管创始人持有的总股本比例很小,但他们的投票权让他们在关键决策(比如谁选董事会)上拥有50.1%的控制权。

There was a lot of discussion around founder control, borrowing some ideas from others like Mark Zuckerberg. There is a new founder Class F stock, which means that even though the founders control a very small portion of the overall equity, their voting shares give them 50.1% control over key decisions like who picks the board.

还有很多细节,但总的来说,我认为Anthropic的形象在完整S-1文件发布后相对而言有所改善。

There is a lot more in here, but I would say that overall, I think Anthropic comes away looking, relatively speaking, better than they did before the full S-1 was released.

我觉得这正是他们想要的,为IPO做好准备、获得势头。

That's exactly, I think, what they would've wanted, giving them momentum heading into the IPO.

不过,呼应一下那种观点——投资者此时基本已经拿定主意了——我仍然认为几乎所有问题都归结于你如何评估未来增长的可能性与非常明确且现实的现有未来承诺。

Still, to echo that sentiment that at this point investors have pretty much made up their mind, I still think that almost everything comes down to how you assess the possibility for future growth versus the very clear and real existing future commitments.

抛开所有细节,这才是唯一重要的数学等式,我不确定S-1对此有什么改变。

Hold aside all the details, that is the one math equation that matters, and I'm not sure that the S-1 changed anything about that.

目前,我们正越来越接近AI行业这一历史性时刻,但今天AI Daily Brief的内容就到这里。

For now, we are getting ever closer to this monumental moment in history for the AI industry, but that is going to do it for today's AI Daily Brief.

一如既往感谢你的收听或观看,下次再见,peace。

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

Peace。

Peace.
已剔除 4 处广告(点击展开查看)
M1
M10:46广告 · 已剔除

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

First of all, thank you to today's sponsors, KPMG, ZData, Granola, and Blitzy.

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].

One more quick note before we get into today's episode, a big thank you to everyone who has checked out my new podcast, How We Got Here.

The feedback on the first episode about how tech lost America has been really helpful, as have all of your follows, ratings, and reviews.

As I mentioned on yesterday's episode, This launch week is where all of that listening and rating is absolutely the most impactful.

So if you haven't checked out How We Got Here yet and you have any time this week to do so, I would so appreciate it.

You can find a link in the show notes or go to howwegothere.com.

M1
M112:50广告 · 已剔除

This episode is brought to you by CData.

M1
M113:06广告 · 已剔除

That's why CData built Connect AI Gateway, one control point between your AI and your systems.

Built on the connectivity layer powering Palantir, Google, Salesforce, and hundreds more, each agent gets exactly the data and tools it needs.

It governs down to the record and lets you finally see your org's complete AI usage in one place.

Accurate answers, a fraction of the token cost, and a gateway that learns your business with every interaction.

See how at cdata.ai.

M1
M114:15广告 · 已剔除

Head to granola.ai/AIDaily and use code AIDAILY to get 100% off for 3 months for new users.

That's granola.ai/AIDaily and code AIDAILY to get 100% off for new users for 3 months.