Daily A.I. Brief · September 8th

来自StoryFlow。

From StoryFlow.

这是你的每日音频简报。

This is your daily audio brief.

嘿。

Hey.

我是西奥。

It's Theo.

9月8日。

September 8.

如果你错过了昨天的收盘，这里有五个我值得提的故事。

Here are five stories I'd flag if you missed yesterday's end of day.

我们开始吧。

Let's get into it.

首先来自IEE Spectrum AI。

First from IEE Spectrum AI.

谷歌DeepMind绘制了90亿种可能的DNA变异图谱。

Google DeepMind maps 9,000,000,000 possible DNA variants.

DNA常被比作一本密码书或一套制造蛋白质、最终形成生命的指令。

DNA is often explained as a code book or set of instructions for producing proteins and ultimately life.

DNA中有些片段叫基因，负责编码蛋白质，但绝大多数DNA被认为是非编码的。

Some stretches of DNA called genes code for proteins, but the vast majority of DNA is considered non coding.

其中一些没有已知功能，而另一些片段则对调控基因活动至关重要。

Some of it has no known function, while other segments are critical to regulating gene activity.

这些调控元件可能以复杂的方式相互作用，其效果在不同细胞和组织中也会有所不同。

These regulatory elements can interact in complicated ways, and their effects can vary across different cells and tissues.

有些还会影响基因组中距离很远的基因。

Some also influence genes located far away in the genome.

下一条。

Next.

第二条来自《数据科学之路》。

Second from towards data science.

上下文窗口不知道什么仍然是真的。

Context windows don't know what's still true.

我构建了一个有效性层，它能做到。

I built a validity layer that does.

你知道上下文窗口，有时候感觉它们像是停留在过去。

You know how context windows, you know, I can sometimes feel like they're stuck in the past.

就像它们试图理解一个已经前进的世界。

It's like they're trying to make sense of a world that's already moved on.

所以这篇文章深入探讨了这个问题，指出这些窗口可能在技术上准确，但在相关性上却失之偏颇。

So this article dives into the issue highlighting how these windows can be technically accurate but still missed the mark when it comes to relevance.

作者创建了一个有效性层，它就像上下文窗口的看门狗。

The author created a validity layer, which acts like a watchdog for context windows.

它会检查正在处理的信息是否仍然有效，基本上过滤掉任何过时的数据。

It checks if the information being processed is still valid, essentially filtering out any stale data.

这很重要，因为基于过时信息行动可能导致代价高昂的错误。

This is important because acting on outdated information can lead to costly mistakes.

他们还开发了一个确定性的基准测试，来衡量依赖旧上下文的影响。

They also developed a deterministic benchmark to measure the impact of relying on old context.

有趣的是，这个层能帮助我们的回答更贴近当前现实，让它们更可靠。

It's fascinating how this layer can help keep our responses grounded in the current reality, making them more reliable.

总的来说，这是一种深思熟虑的方法，能提升我们AI系统的准确性，确保它们与当下保持同步。

Overall, it's a thoughtful approach to enhancing the accuracy of our AI systems, ensuring they stay in tune with the present.

接下来，是来自解码器的第三条新闻。

Up next, third from the decoder.

MistralOut Eye筹集了30亿欧元，这是欧洲有史以来最大的一轮科技融资，尽管它落后于竞争对手。

MistralOut Eye raises €3,000,000,000 in Europe's largest ever tech funding round despite lagging behind rivals.

MistralOut Eye刚刚完成了令人瞩目的30亿欧元D轮融资，这实际上是欧洲历史上最大的一轮科技融资。

MistralOut Eye just wraps up an impressive €3,000,000,000 series d funding round, which is actually the largest tech funding round in Europe's history.

令人惊叹的是，他们自三年前成立以来发展如此迅速，现在估值已经超过210亿欧元。

It's fascinating to see how quickly they've grown since their launch just three years ago, now boast in evaluation that's crossed €21,000,000,000.

特别有意思的是，尽管有这么多资金涌入，Mistral似乎仍在努力追赶一些在航空航天领域的竞争对手。

What's particularly interesting is that despite this massive influx of cash, Mistral seems to be playing catch up with some of its competitors in the aerospace.

这提醒我们，即使有巨额资金支持，科技竞赛也不仅仅是钱的问题。

It's a reminder that even with significant financial backing, the race in tech isn't just about the money.

还关乎创新和市场定位。

It's also about innovation and market positioning.

这轮融资可能会给Mistral提供所需的资源，来加强技术，追赶那些领先的对手。

This funding round could provide Mistral with the resources it needs to bolster its technology and catch up to those ahead of them.

接下来会很有趣，看他们如何利用这笔资金，以及能否从行业追随者转变为领导者。

It'll be intriguing to see how they leverage this capital moving forward and whether they can shift the narrative from being a follower to a leader in the industry.

然后，第四条来自KDnuggets，五种免费访问编码模型的方法。

And then, fourth from k d nuggets, five ways I access coding models for free.

探索五种免费方式，访问他们的iCoding代理、专有编码模型和开放权重模型，无需支付昂贵的订阅费或GPU费用。

Explore five free ways to access their iCoding agents, proprietary coding models, and open weight models without paying for expensive subscriptions or GPUs.

一年前，iCoding意味着让聊天机器人生成一个函数。

A year ago, iCoding meant asking a chatbot to generate a function.

现在情况已经变了。

That has changed.

我们不再局限于让聊天机器人生成函数或解释错误了。

We are no longer limited to asking a chatbot to generate a function or explain an error.

现代编程代理可以检查整个代码库，编辑多个文件，运行终端命令，测试修改，并且几乎自主地处理复杂的开发任务。

Modern coding agents can inspect entire repositories, edit multiple files, run terminal commands, test their changes, and work through complex development tasks almost autonomously.

接下来，来自KD Nuggets的第五条。

Next, fifth from KD Nuggets.

AriaJS真的适合代理式用户界面吗？

Is AriaJS really the right for the agenticurer?

以下是我发现的情况。

Here's what I found.

我们构建界面的方式正在改变。

The way we build interfaces is changing.

随着我们的代理编写越来越多的代码，我们用来渲染代码的工具可能也需要改变。

As our agents write more of our code, the tools we use to render the code may need to change too.

介绍一种过去十年里出现在开发领域的新角色。

Introduction, a new kind of developer in the room for the past decade.

JavaScript生态系统一直基于一个舒适的假设在运作。

The JavaScript ecosystem has operated on a comfortable assumption.

开发者写代码，框架组织代码，浏览器渲染代码。

Developers write the code, frameworks organize it, and browsers render it.

React、Vue、Angular以及它们的生态系统都是为人类在键盘前操作而设计的。

React, Vue, Angular, and their ecosystems were all designed with a human at the keyboard.

这个假设正在瓦解。

That assumption is cracking.

我们的iCoding代理，比如Claude Code、GitHub Copilot和Cursor，现在已经在编写大量生产代码了。

Our iCoding agents such as Claude Code, GitHub Copilot, and Cursor now write meaningful portions of production code.

接下来。

Up next.

来自IEEE新闻的第六条。

Sixth from our IEEE news.

可口可乐在马来西亚使用我们的IEEE来改进零售商的订货流程。

Coca Cola uses our IEEE to improve retailer ordering in Malaysia.

可口可乐正在使用我们的IEEE推荐系统，通过其Coke Buddy平台，告诉马来西亚的零售商应该订购哪些产品以及订购多少。

Coca Cola is using our IEEE recommend which products Malaysian retailers should order and in what quantities through its Coke Buddy platform.

“完美购物篮”功能利用可口可乐的中央推荐引擎，分析过去的订单、订购频率、季节性、天气以及类似企业的购买模式。

The Perfect Basket feature uses Coca Cola's central recommendation engine to analyze previous orders, ordering frequency, seasonality, weather, and purchasing patterns among similar businesses.

可口可乐说“可乐”，但目前它在马来西亚支持大约39,000个零售点。

Coca Cola said Coke, but it currently supports about 39,000 retail outlets across Malaysia.

接下来，第七条新闻：ARM推出了面向物理AI和机器人技术的Total Design框架。

And then seventh from our news, ARM launches Total Design for Physical AI and robotics framework.

ARM推出了ARM Total Design for Physical AI，并附带一个新的机器人框架，旨在为自动化系统建立通用标准。

ARM has launched ARM total design for physical AI alongside a new robotics framework to establish common standards across automated systems.

涵盖采矿、农业、制造业和全球运输的物理行业，贡献了数万亿美元的经济活动，并预计到2030年代每年有2000亿美元的计算机会。

Physical industries spanning mining, agriculture, manufacturing, and global transport account for trillions of dollars in economic activity and an estimated $200,000,000,000 annual compute opportunity by the twenty thirties.

为了解决这些行业中的工程碎片化问题，ARM正在召集超过80个合作伙伴组织，涵盖软件、硬件和ROI领域。

To address engineering fragmentation across these sectors, ARM is convening more than 80 partner organizations spanning software, hardware, and ROI.

接下来，第八条来自《麻省理工科技评论》的新闻。

Next, eighth from MIT Technology Reviewer Eye.

一些创业者正在开发能够提前规划、应对突发情况的智能体。

Disarray entrepreneurs developing agents that can plan ahead for the unexpected.

丹尼泽尔·哈夫纳在旧金山索玛区的办公室几乎空无一人。

Denizel Hafner's office in San Francisco, Somar District sits mostly empty.

他全新的初创公司还处于保密模式，门上连名字都没有。

His brand new startup is still in stealth mode and doesn't even have its name on the door.

我去拜访的那天，那里只有另外一个人，家具也寥寥无几。

On the day I visit, there's only one other person there and little in the way of furniture.

虽然装饰不多，但机器人却不少。

But what it lacks in decor, it makes up for in robots.

各种形状和大小的类人机器人像提线木偶一样悬挂在横穿这个开阔空间的架子上。

Humanoids of various shapes and sizes hand like marionettes from wrecks that run down the center of the wide open space.

接下来，是来自《解码器》的第九条新闻。

Up next, ninth from the decoder.

Meta 在 Tokamaxing 计划失败后，将使用情况从工程师绩效评估中移除。

Meta drops our usage from engineer performance reviews after Tokamaxing backfires.

Meta已经决定不再把我们的使用情况纳入工程师的绩效评估中。

Meta has decided to stop incorporating our usage into its engineers performance reviews.

这一变化是在“Tokamaxing”现象之后出现的，员工们为了提升内部排行榜上的排名，过度使用AI工具。

This shift comes after a phenomenon of Tokamaxing, where employees were excessively using AI tools just to boost their standing on internal leaderboards.

公司现在将不再关注他们使用AI的量，而是优先考虑工作的质量、速度和复杂度。

Instead of focusing on how much AI they use, the company will now prioritize the quality, speed, and complexity of their work.

这一变化很可能是对AI使用成本上升的回应，到2026年这些成本可能达到数十亿。

This change is likely a response to the rising costs associated with AI usage, which could reach billions by 2026.

展望未来，Meta还在测试一个名为Hatch的新AI工具，旨在自动化某些任务。

Looking ahead, Meta is also testing a new AI tool called Hatch designed to automate certain tasks.

然而，由于隐私方面的担忧，员工们对于将其与个人账户关联有些犹豫。

However, there's some hesitation among employees about linking it to their personal accounts due to privacy concerns.

看到公司如何在利用技术和回应员工反馈之间找到平衡，确实很有意思。

It's interesting to see how companies are navigating the balance between leveraging the right and addressing employee feedback.

对吧？

Isn't it?

然后是解码器里的第十个。

And then Tenth from the decoder.

阿斯麦、台积电、三星和英特尔，还有华为，都在争相打破它的垄断。

Asmolox and TSMC's Samsung and Intel Huawei races to break its grip.

阿斯麦通过让台积电、三星和英特尔这些大客户转向更大的光罩，已经和他们建立了合作关系，这预计能让它最新的EUV光刻机效率提升大约40%。

Asmol has secured partnerships with major players like TSMC, Samsung, and Intel by transitioning them to larger photo masks, which is expected to enhance the efficiency of their latest EUV machines by about 40%.

这个转变很重要，因为这意味着这些公司能更快、更有效地生产芯片，在竞争激烈的半导体领域里，这非常关键。

The shift is significant because it means these companies can produce chips more quickly and effectively, which is crucial in the competitive semiconductor landscape.

另一方面，华为正在加大力度，减少对阿斯麦技术的依赖。

On the flip side, Huawei is ramping up efforts to reduce its reliance on Osmol technology.

他们正在和玉林盛以及其他国内供应商合作，开发自己的光刻解决方案。

They are working with Yulin Sheng and other domestic suppliers to develop their own lithography solutions.

这一举措反映了中国更广泛的战略，即增强半导体能力，减少对外国技术的依赖，尤其是在当前地缘政治紧张的情况下。

This move reflects a broader strategy within China to strengthen its semiconductor capabilities and lessen dependence on foreign technology, especially amid ongoing geopolitical tensions.

看到这些动态在行业里如何演变，真的很有意思。

It's fascinating to see how these dynamics are playing out in the industry.

今天的内容就到这里。

That's it for today.

明天早上，无论夜晚带来什么。

Tomorrow morning, whatever the night brings.

保持好奇。

Say curious.

新来的朋友。

New here.

在storyflow.com免费获取你的专属内容。

Grab your own free feed at storyflow.com.

只需要几秒钟。

It only takes a few seconds.