9 AI Techniques You Probably Haven't Tried — 文稿

M10:00

如果我告诉你，你一直都把 AI 用错了呢？

What if I told you you were using AI all wrong?

好吧，那我就是在撒谎，而且很明显，我就是想用一个极其荒唐、离谱的开场白来骗你点击。

Well, then I'd be lying, and I'd clearly be trying to get you to click on something by using an absolutely ridiculous and preposterous hook.

但换个说法，如果我告诉你，有九种 AI 技巧已经给一些人带来了非常棒的效果，而你可能还没来得及尝试呢？

But instead, what if I told you that there were 9 AI techniques that were delivering some really awesome results to some people that you might not have had the time to try just yet?

这就真实多了，因为在过去几个月里，我们看到一大批新功能和新工具陆续上线，比如 Claude 的 /design，Codex 的实时语音模式，还有 GrokBot 可以通过观看屏幕，让用户把一整套工作流程训练给它。

That would be a lot more true because over the last couple of months, we've seen a slew of new features and new tools become available, like Claude's /design and Codex's live voice mode and GrokBot's ability for a user to train it on an entire workflow.

就是通过看屏幕来学。

By watching the screen.

AI 最让人兴奋的一点，也正是它最有挑战性的一点——它一直都在变化。

One of the things that makes AI so exciting is also the thing that makes it the most challenging—that it's changing all of the time.

不过今天这一集，会让你很快就跟上最新进展。

But today's episode is going to get you up to speed in no time at all.

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

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

好了，朋友们，在我们正式开始之前，先快速说几个通知。

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

M11:40

哎呀，朋友们，AI 这趟炒作列车现在真是越开越猛了。

Well, my goodness, friends, the AI hype train is a-hypin'.

但在这个故事里，AI 真正扮演的角色是什么呢？

But what is AI's real role in this story?

最近，Anthropic CEO Dario Amodei 和一些批评他的人之间，围绕他传递信息的语气展开了一整场讨论。

Recently, there was a whole discussion between Anthropic CEO Dario Amodei and some of his critics about the tone of his messaging.

有人批评 Dario 过于负面，而他对此的一个回应基本上是说，改变人们对 AI 看法的不会是营销，而会是 AI 真正拿出结果。

One of Dario's responses to the critique that he had been overly negative was basically to say that it wasn't going to be marketing that changed people's opinions about AI, it was going to be AI actually delivering results.

具体来说，他发帖说：“我不认为一场包装得光鲜亮丽、带着正面叙事的营销活动，是重新赢回信任的办法。”

Specifically, he posted, "I don't think that a glitzy marketing campaign with a positive spin is the way to win back trust."

“到了现在，说 AI 会治愈癌症，与其说是鼓舞人心，不如说已经成了陈词滥调，而且大多数人会觉得这是在骗人。”

"At this point, saying that AI will cure cancer is more a cliché than it is inspiring, and most people think it is deceptive."

“真正有效的办法，是实际治愈癌症。”

"The thing that will work is actually curing cancer."

也正因为如此，昨天有很多人基本上都在说，这件事已经发生了。

Which is why there were a lot of folks basically saying that that's what had happened yesterday.

周三，Moderna 和 Merck 宣布，一项个性化癌症疫苗的三期临床试验取得成功，这是第一个显示出前景的后期临床试验。

On Wednesday, Moderna and Merck announced a successful phase 3 trial for a personalized cancer vaccine, the first late-stage trial that's shown promise.

这种治疗方式和基于化疗的方法非常不一样。

The treatment functions very differently from chemotherapy-based approaches.

它不是用放疗去轰击癌细胞，而是分析癌细胞，找出导致癌症的 DNA 突变，然后制造一种个性化的 mRNA 疫苗，来纠正这个突变。

Instead of blasting the cancer with radiation, the new method involves analyzing the cancerous cells, identifying the DNA mutation causing the cancer, and creating a personalized mRNA vaccine that can correct the mutation.

这项试验发现，在一千一百多名晚期黑色素瘤患者中，这种治疗通过延长他们的缓解期，取得了成功。

The trial found the treatment was successful in more than 1,100 patients with advanced melanoma by extending their time in remission.

Moderna 和 Merck 也在针对肺癌进行类似试验，科学家们希望这种治疗方法能应用到很多不同类型的癌症上。

Moderna and Merck have similar trials underway for lung cancer, and scientists hope the treatment can be applied to many different cancers.

Mass General 黑色素瘤中心主任 Ryan Sullivan 医生说，这对整个领域来说都是一件大事。

Dr. Ryan Sullivan, director of the Center for Melanoma at Mass General, said, it's a big deal for the field in general.

有了这项积极的研究结果，人们有理由期待，接下来也很可能会有投资跟进，而这些方法或许会更广泛地改变我们治疗癌症的方式。

With this positive study, there is a hope and likely investment to follow that these approaches may change the way we treat cancer more broadly.

这也引发了 AI 社区里大量非常兴奋的帖子，比如 Chubby 发的这条。

This gave rise to a lot of very excited posts from the AI community, like this one from Chubby.

他们写道：“这也太重大了。”

This is freaking huge, they wrote.

“这是第一次，由 AI 辅助的个性化 mRNA 癌症治疗在三期临床试验中取得成功。”

For the first time, an AI-assisted personalized mRNA cancer treatment has succeeded in a phase 3 trial.

“Moderna 和 Merck 会对每位患者的肿瘤进行测序，并把它和他们的健康 DNA 进行比较。”

Moderna and Merck sequence each patient's tumor and compare it to their healthy DNA.

“然后 AI 会帮助识别，肿瘤中的哪些突变最有可能触发免疫反应。”

AI then helps identify which of the tumor's mutations are most likely to trigger an immune response.

“简直太不可思议了。”

Absolutely incredible.

Dario 是对的。

Dario was right.

癌症会在短短几年内被治愈。

Cancer will be cured in just a few years.

当然，也有一些人对把这件事称为 AI 提出了异议。

Some, of course, took issue with the labeling of this as AI.

Antibody42 总结了这种批评，他说，这太蠢了。

Antibody42 summed up the criticism, saying, this is so dumb.

他们用 AI 这个词来替代 machine learning，因为对那些不在乎现实的人来说，这两者的区别并不重要。

They are using AI in place of machine learning because the distinction doesn't matter to people who don't care about reality.

用于 mRNA 设计的 AI 已经存在一段时间了。

AI for mRNA design has been around for a while.

它不是 LLM。

It's not an LLM.

先不谈语义上的区别，重要的是要说明，这并不是科学家把一堆结果塞进 ChatGPT，然后让它想出一个个性化的癌症疗法，也就是说，治愈癌症，而且不要出错。

And putting aside semantics, it is important to note that this is not scientists just plugging a bunch of results into ChatGPT and asking it to come up with a personalized cancer cure, i.e., cure cancer, make no mistakes.

这些是先进的 machine learning 方法，和支撑 AlphaFold 的技术更相似。

These are advanced machine learning methods that have more similarities to the technology underpinning AlphaFold.

说实话，同样重要的是，不能因为简单地把突破归功于 AI，就忽视那些正在和 machine learning 技术一起工作、开发这些新疗法的杰出人类科学家。

It's also frankly important not to lose sight of the brilliant human scientists that are working with machine learning techniques to come up with these new treatments by simply attributing the breakthrough to AI.

话虽如此，我也认为，如果说这只是 AI 炒作列车，那也太轻描淡写了。

That said, I do also think it would be dismissive to say that this is just the AI hype train.

虽然现在还处在早期阶段，但它确实有真正的潜力，成为迈向治愈癌症的重要一步。

Although it's still early days, this has genuine potential to be a major step towards curing cancer.

而且，虽然它和 LLM 以及聊天机器人没有直接关系，但这项技术仍然极大受益于过去几年在算力、研究和人才上的大量投入。

And while it isn't directly related to LLMs and chatbots, this technology has still benefited greatly from all of the investment into compute research and talent over recent years.

AI 时代，不管有没有 LLM，都正在为一些真正了不起的突破铺路，而这件事值得庆祝。

The AI era, LLMs or not, is setting the stage for some truly remarkable breakthroughs, and that is something worth celebrating.

市场当然也在庆祝。

Certainly the market's celebrating it.

Shabir 指出，Moderna 的股价在公告发布后上涨了百分之七，仅仅几个小时后又飙升到百分之十二点五的涨幅。

Shabir pointed out that Moderna stock was up 7% after the announcement, rocketing up to 12.5% growth just a few hours later.

Elon Musk 说，接下来会有太多突破。

Said Elon Musk, so many breakthroughs are coming.

接下来这个话题，和治愈癌症相比看起来非常普通，但对很多企业来说极其重要。

Next up, a topic which seems extremely pedestrian next to curing cancer, but is extremely important for lots of enterprises.

OpenAI 提出了一种新的安全技术，可以让他们提供 frontier AI，同时不需要持续监控用户。

OpenAI has come up with a new safety technique that will allow them to offer frontier AI without constantly monitoring their users.

他们把这个流程叫作 Private Safety Processing，并会把它应用到符合条件、签有 zero data retention 协议的 API 客户身上。

They're calling the process Private Safety Processing and will be applying it to eligible API customers with zero data retention agreements.

OpenAI 表示，这将让他们能够兑现承诺，也就是在一次请求完成后不保留 prompts 或 outputs，并确保敏感的企业数据不会被 OpenAI 员工接触到。

OpenAI says this will allow them to fulfill their promise of not retaining prompts or outputs after a request has been completed and ensure that sensitive corporate data is not available to OpenAI staff.

自从 long-horizon agents 兴起以来，安全监控变得困难得多。

Since the rise of long-horizon agents, safety monitoring has become much more difficult.

旧的系统设计只能评估单次交互，而不能评估整个 agentic work 的完整过程。

Older system designs can only evaluate individual interactions rather than assessing the entire arc of agentic work.

Anthropic 对这个问题的解决方案，就是直接对 Claude Code 关闭 zero data retention，收集每个 session 的完整数据，用来扫描有害活动。

Anthropic's solution to this problem has been to simply disable zero data retention for Claude Code, gathering full data from each session to scan for harmful activity.

对很多企业客户来说，这完全不可接受，也反映在 Claude Code 在企业中的采用率相当惨淡。

For many enterprise customers, this is a complete non-starter and has shown up in fairly dismal adoption of Claude Code in the enterprise.

OpenAI 的新系统把自动安全扫描扩展到了整个 session，包括客户控制的数据存储，这些数据存储会被用于多个 agentic steps 之间的上下文。

OpenAI's new system extends automated safety scanning across an entire session, including customer-controlled data storage used for context across multiple agentic steps.

这种扫描仍然是完全自动化并且加密的，也就是说，绝不会有任何人类看到敏感数据。

This scanning is still fully automated and encrypted, meaning that no human ever lays eyes on sensitive data.

如果某个问题被标记出来，OpenAI 员工会收到一份活动摘要，里面有类别和严重程度评级，但会去掉所有客户数据。

If an issue is flagged, an OpenAI employee gets a summary of the activity with a category and severity rating, but stripped of any customer data.

这意味着，误报将不再导致数据暴露，同时也让 OpenAI 能够开发更强大的模型，而不需要进行完全透明的人工审查。

This means that false positives will no longer cause data exposure, as well as allowing OpenAI to develop more powerful models without the need for fully transparent review.

简单来说，这项改变可能会帮助企业里的 AI 少一些围绕风险和安全的取舍。

The TL;DR is that this is a change that could help make AI in the enterprise less of a trade-off around risk and security.

OpenAI 的产品政策负责人 Aaliyah House 说，我们一直在和很多企业客户交流，他们真的、真的非常在意自己企业数据的隐私和安全。

OpenAI's head of product policy, Aaliyah House, said, we've been talking to a bunch of enterprise customers and they really, really care about their enterprise data privacy and security.

我们已经非常清楚地从企业那里听到，这件事很重要。

We've heard very loud and clear from businesses that this is important.

他们自己通常也对客户作出过一些承诺。

They often have their own commitments that they have made to their customers.

总结一下，OpenAI 员工 AdamGPT 补充说，这是那种看起来很小、但实际上非常大的事情。

Summing it up, OpenAI staff member AdamGPT added, this is one of those small things that is actually a huge thing.

而根据我自己持续和企业围绕 AI 进行的交流来看，这一点绝对是真的。

And based on my ongoing conversations with enterprises around AI as well, that is absolutely true.

OpenAI 还宣布了一项新的合作，为 Replit 新推出的 free mode 提供模型支持。

OpenAI also announced a new partnership providing the model for Replit's newly launched free mode.

不过，尽管名字叫 free mode，它并不是字面意义上的免费。

Now, despite the name, free mode isn't literally free.

相反，它让每月二十美元套餐的用户，可以在不消耗使用额度的情况下使用 Replit。

Instead, it gives users on the $20 a month plan the ability to use Replit without spending usage credits.

free mode 会把所有查询都路由到 GPT-5.6 Luna，Replit 表示，这能让用户在正常订阅下多创作三十倍的内容。

Free mode routes all queries through GPT-5.6 Luna, which Replit says will allow users to create 30 times more on their normal subscription.

当 Luna 不足以处理某个复杂任务时，用户仍然可以切换到另一种模式，选择更强大的模型。

When Luna isn't enough for a complex task, users can still shift to another mode to select a more powerful model.

但对于日常任务，比如头脑风暴，或者快速做出一套 slide deck，很多人发现 Luna 已经绰绰有余。

But for everyday tasks like ideating or spinning up a slide deck, many are finding Luna to be more than enough.

在发布博客文章里，Replit 写道，AI 模型现在已经足够强、也足够便宜，可以让过去遥不可及的结果变得实际可行。

In their launch blog post, Replit wrote, AI models are now capable and affordable enough to make once-unreachable outcomes practical.

而我觉得有意思的是，相比这次 Replit 合作的具体细节，更多是 OpenAI 对 5.6 Luna 的营销和推广重点。

And I think what's interesting to me is more OpenAI's focus on marketing and promoting 5.6 Luna as opposed to the specifics of this Replit partnership.

如果你一直在密切关注，就会很明显地看到，OpenAI 认为自己是在两条非常不同的轴线上同时竞争。

If you've been watching closely, it's clear that OpenAI sees themselves as competing on two very different axes at once.

当然，他们在用 5.6 Sol 以及即将到来的 Astra model，在最前沿的技术水平上竞争；但尤其是自从他们推出 Luna 之后，也非常明显，他们同时在回头盯着中国的 open-weight models，并且完全不愿意把那块阵地也让出去。

They are, of course, competing at the state of the art with 5.6 Sol and the Astra model to come, but especially since they launched Luna, it's very clear that they are also looking behind them to the Chinese open-weight models and are completely unwilling to surrender that ground as well.

我预计，随着用户越来越成熟，越来越理解不同类型的任务分别需要什么级别的能力，我们会看到更多围绕效率前沿的竞争，而不只是围绕能力前沿的竞争。

I would expect that as we see more maturation among users in terms of their understanding of which power levels are required for which different types of tasks, we're gonna see a lot more of this competition on the efficiency frontier, not just the capability frontier.

现在我们先不说 Replit，但还是继续讲 AI 编程创业公司。Bloomberg 本周报道称，SpaceX 曾接触 coding agent 创业公司 Cognition，讨论一项潜在交易。

Now, moving off of Replit, but staying around AI coding startups, Bloomberg reported this week that SpaceX had approached coding agent startup Cognition about a potential deal.

Bloomberg 援引熟悉此事的匿名消息人士写道，这些交易谈判目前并不活跃，但两家公司仍在继续讨论合作，包括可能安排 Cognition 使用 SpaceX 的计算能力，这些人士说。

Citing anonymous sources familiar with the matter, Bloomberg wrote, the deal talks are not currently active, but the companies continue to hold discussions about working together, including potentially arranging for Cognition to use SpaceX's computing capacity, said the people.

Cognition 最近一次融资是在五月，估值达到二百六十亿美元，据报道他们正在寻求新一轮融资，目标估值至少四百亿美元，而这一轮目前仍处于早期阶段。

Cognition most recently raised at $26 billion in May and is reportedly seeking a valuation of at least $40 billion in a new round that's still in the early stages.

重要的是，这篇报道暗示，SpaceX 是把 Cognition 看作对 Cursor 的补充，而不是替代品。

Importantly, the report suggests that SpaceX is looking at Cognition as an addition to Cursor rather than an alternative.

当然，对那些一直密切关注的人来说，Elon Musk 愿意花大约一千亿美元去收购正在构建 AI 应用层的创业公司，这并不令人意外。

And certainly it's not surprising to people who have been watching closely that Elon Musk would be willing to spend around $100 billion on acquiring startups that are building the AI app layer.

不过，在报道发布后不久，Cognition CEO Scott Wu 就否认了整件事。

However, shortly after the report was published, Cognition CEO Scott Wu disavowed the entire thing.

他在 X 上发帖写道，这不是真的。

In an X post, he wrote, this is not true.

非常尊重 SpaceX 团队，但 Cognition 不出售，而且我们也没有在谈。

Huge respect for the SpaceX team, but Cognition is not for sale and we haven't been talking.

Bloomberg 记者 Rebecca Torrence 坚持自己的报道，并感叹媒体素养的消亡，她评论说，说清楚一点，根据我们的消息来源，SpaceX 曾带着出价接触 Cognition。

Bloomberg journalist Rebecca Torrence stood by her reporting and decried the death of media literacy, commenting, to be clear, according to our sources, SpaceX approached Cognition with a bid.

报道并没有说 Cognition 参与了那次尝试。

The story does not say Cognition engaged with that attempt.

当天晚些时候，Elon 给这件事画上了句号，他说，Scott 说的是准确的。

Later in the evening, Elon put the matter to bed, saying, what Scott says is accurate.

我们没有和 Cognition 谈过任何事情，除了确保 Grok 能很好地满足他们的需求。

We haven't talked with Cognition about anything except making sure Grok works well for their needs.

这是那种少见的情况，我觉得你其实不需要怀疑任何一方，尤其是因为“接触某人讨论潜在交易”这件事本身到底意味着什么，边界就很模糊。

This is one of those rare cases where I don't think you actually have to doubt anyone, specifically because of the blurriness of what it even means to actually approach someone about a potential deal.

如果某天下午 Elon 和 Scott 在聊 Grok 4.6 在 Devin 里面运行得怎么样，然后 Elon 轻轻碰他一下肩膀说，嘿，听着，兄弟，如果你以后有想过希望 Cognition 走向哪里，那这算不算是在接触他们讨论一项潜在交易呢？

If Elon and Scott were talking one afternoon about how Grok 4.6 was working inside Devin, and Elon gives him a little shoulder nudge and says, hey, listen, man, if you ever think about where you want to see Cognition in the future, does that constitute approaching them about a potential deal?

很多这类报价一开始真的就是这样的，而且我也确实认为，Elon 对收购的胃口不会只是因为 Cursor 这笔交易完成了就得到满足。

That is genuinely how a lot of these offers start, and I certainly think that Elon's appetite for acquisitions has not been sated just because the Cursor deal is done.

今天最后一个话题，是围绕政府安全测试框架细节的持续疑问。

Lastly today, continued questions around the details of the administration's safety testing framework.

这个框架在几周前推出的时候，很多人最大的担忧之一就是细节不会公开，这会让主要实验室之外的所有人都一头雾水。

When the framework was rolled out a couple of weeks ago, one of the big concerns for many was that the details wouldn't be made public, leaving everyone outside of the major labs in the dark.

White House 在一次闭门会议上公布了这个新框架，参会的只有 OpenAI、Anthropic 和 Google 的技术人员。不过看起来，就连受邀参加会议的这些公司，也仍然很难完整了解这个新框架。

The White House unveiled the new framework in a closed-door meeting attended only by technical staff from OpenAI, Anthropic, and Google, but it seems that even the companies invited to the meeting are still struggling to get a full picture of the new framework.

消息人士说，在会议期间，他们拿到了这个框架的纸质副本，但只被允许做笔记。

Sources said that during the meeting, they were handed paper copies of the framework but were only allowed to take notes.

这些消息人士还说，White House 仍然没有发放这个框架的书面版本，也没有分享任何关于本月计划举行的公开 AI 活动的更多细节。

Those sources also said that the White House still hasn't distributed written versions of the framework or shared any further details about a public AI event intended to be held this month.

当然，到目前为止，这种不明确显然还没有成为前沿 AI 发展的障碍。比如 OpenAI 最近自愿放慢研究节奏这样的决定，就可以看作是行业在自我控制步伐。

Now, obviously, this lack of clarity hasn't been a roadblock for frontier AI development so far, given recent decisions like OpenAI's voluntary slowdown in research as an example of the industry pacing itself.

不过，随着这个框架逐步推出，要求提高透明度的呼声是很明确的。

Still, there is a clear call for more transparency as this framework is rolled out.

上周，自由意志主义智库 Cato Institute 的 Juan Londoño 批评了这种保密做法。他写道，White House 实际上是在把 AI 测试制度放进一个黑箱里。

Last week, Juan Londoño of the libertarian Cato Institute criticized the secretive approach, writing, the White House is essentially putting the AI testing regime in a black box.

这种做法违背法治原则，有可能变得像许可制度一样具有强制规定性，而且最重要的是，它没能实现自己最基本的目标：在公众中建立信任。

This approach contravenes the rule of law, risks becoming as prescriptive as a licensing regime, and most importantly, fails to fulfill its most basic objective: to build trust in the population.

一个提出的问题比它回答的问题还多的框架，比完全没有框架还糟。

A framework that raises more questions than the one it answers is worse than no framework at all.

尤其是结合我们这周关于 data centers 和信任的所有讨论，我完全同意这一点。

Especially in light of all of our discussions around data centers and trust this week, I fully agree.

不过今天的 AI Daily Brief 头条版，到这里就要结束了。

For now though, that is going to do it for the AI Daily Brief Headlines Edition.

接下来，是今天的主要节目。

Next up, the main episode.

M114:37

欢迎回到 AI Daily Brief。

Welcome back to the AI Daily Brief.

今天我们会稍微偏一点操作流。

Today we are getting a little bit operator-y.

有一件事真的很有意思，而且我相信你们很多人也都有这种感觉：就算你觉得自己是 AI 的高阶用户，或者至少是一个会关注 AI 如何变化的人——如果你在听这个节目，大概率你就是这样的人——人们和 AI 互动的方式变化太快了，所以在大家实际怎么用这些工具这件事上，稍微落后一点其实一点也不难。

One of the things that's really interesting, and I'm sure that a lot of you have felt, is that even if you consider yourself a power user of AI, or at least someone who pays attention to how AI changes, which presumably if you're listening to this show, you are, the ways that people interact with AI change so much that it's not hard at all to fall a little bit behind in how people are actually using these tools.

我们会习惯并安于自己一直以来做事的方式。

We get settled and comfortable into the way that we've been doing things.

而且说实话，有时候我们还真的会变得很擅长用那种新方式工作。

And honestly, sometimes we get really good at working in that new way.

所以要把已经成型的一套做法推翻，去尝试新的 AI 工作方式，并不总是我们会优先去做的事情。

And so to upturn that apple cart and try new ways of working with AI is something that we don't always prioritize.

考虑到夏天快要结束，返校、返工等等也要开始了，我想分享一些我最近看到的讨论，关于人们正在使用、正在尝试的一些新技巧、小窍门和功能，可能会对你有用。

Given that we are coming towards the end of the summer and the beginning of back to school, back to work, et cetera, I wanted to share a bunch of the chatter that I've seen recently about new techniques and tricks and features that people are using and doing that might be useful for you.

当然，很可能这里面有一些事情你已经在做了。

Now, probably it is the case that some number of these are things that you are already doing.

但也许其中会有某一件事，适合你在下一次实验日拿来试一试。

But maybe there's something that on your next experimental day might be the thing to try.

在进入这九个技巧、建议、小窍门等等之前，我想先给一个贯穿整期节目的元建议。

Now, before we get into these 9 techniques, tips, tricks, et cetera, I want to provide one meta tip that really runs across this entire show.

在这一期里，我提供给你的不一定是我自己对某些具体技巧的亲身经验，而更多是我把其他人分享的、他们如何学习的内容做了一个汇总。

In this particular episode, what I'm providing you is not my experience with specific techniques necessarily, but instead an aggregation of what other people are sharing about how they're learning.

AI 特别酷的一点就在于，人们是在公开地学习和实验。

One of the things that makes AI so cool is that people are learning and experimenting in public.

说实话，虽然这些平台有时候确实很像粪坑，但如果你不在 LinkedIn，尤其是不在 X 上看那里的 AI 讨论——好吧，我想你也算有人照顾，因为你在这里听我说——但真的可以从其他人以及他们正在做的实验里学到很多东西。

And honestly, for as cesspool-y as they can be, if you're not on LinkedIn or especially X and watching the AI conversation there, well, I guess you're in good hands because you're here, but there is a lot to learn from other people and the experiments that they're running.

我想讨论的第一个技巧，是 voice mode。

Now, technique number 1 that I want to discuss is voice mode.

一方面，voice mode 并不是什么新东西。

On the one hand, voice mode is nothing new.

如果你参加过任何 AIDB 或者 Superintelligent 的培训项目，你应该听过或者读过我一直念叨，让你在电脑上设置类似 Whisper 这样的东西，这样你就能更有效地使用语音。

If you have ever done any of the AIDB or Superintelligent training programs, you will have heard or read me haranguing you to set up something like Whisper on your computer so you can use your voice more effectively.

但最近我们有了 ChatGPT 新的实时语音模式，对很多人来说，这完全是个改变游戏规则的东西。

But recently we have gotten ChatGPT's new live voice mode, and for a lot of people it has been a complete game changer.

早在七月底，Every 的 Dan Shipper 在 Slack 里分享了他们关于这个功能的讨论，并补充说，Every 几乎每一个人都在为 ChatGPT for Work 里的语音模式有多好用而疯狂。

Back at the end of July, Dan Shipper from Every shared their discussion in Slack about the feature and added, almost every single person at Every is freaking out about how good voice mode in ChatGPT for Work is.

已经有一阵子没见过大家情绪这么高涨了。

Have not seen vibes this high in a while.

如果你还没试过，你真的应该试试。

If you haven't tried it yet, you should.

Allie K. Miller 在这期节目里会频繁出现，如果你还没关注她，她非常值得关注；她一直在不断强调这件事。

Allie K. Miller, who is going to feature prominently across this show and who is a great follow if you're not yet, has been beating this drum consistently.

差不多就在 Every 那篇帖子发布的同一时间，她另外也说，已经快凌晨两点了，而我之所以能完成所有紧急工作，唯一的原因就是 ChatGPT Voice Mode。

Right around the same time of that Every post, she separately said, It's nearly 2 AM and the only reason I finished all of my urgent work is ChatGPT Voice Mode.

天哪，为什么不是所有人都在为这个疯狂？

Good God, how is everyone not freaking out about this?

我感觉我现在工作的方式，跟一个月前完全不一样了。

I feel like the way I'm working looks completely different than a month ago.

所以我会对它说一些很基础的话，比如：嘿，帮我看一下这段话，你觉得怎么样？

So basic examples of what I say to it: Hey, take a look at this paragraph, what do you think?

我的清单上还有什么？

What else is on my list?

我来改这个，你去找那个文档。

I'm going to edit this while you find that doc.

下载我最新的 Instagram Reel，给它做一个好看的缩略图。

Download my latest Instagram Reel, make a good thumbnail for it.

我去接点水，你需要我就喊一声。

I'm gonna go fill up my water, holler if you need me.

刚发完十五封客户邮件，我开始写脚本的时候，你对照清单检查一下。

Just sent the 15 client emails, check against the list while I start on the script.

不过 Allie 澄清说，重点其实不太在具体用例，而在于这种环境式的互动。

Allie clarifies though, it's less about use case and more about ambient interaction.

跟一个一直在环境里待命的助手说话，和点击按钮再说话，听起来差别可能微不足道，但在实际使用中，这中间差得太远了。

It probably sounds like a negligible difference to speak to an ambient assistant versus click to speak, but it is a world of distance in practice.

最近 Allie 在一条推文里谈到她觉得 AI 里哪些东西没有得到足够关注时，她说，有些人会说 Codex Live Voice Mode 是一个环境式的 chief of staff。

More recently in a tweet where Allie was talking about what she thinks isn't getting enough attention in AI, she said, some will say Codex Live Voice Mode is an ambient chief of staff.

我觉得这还是低估了它。

I think that undersells it.

它是一个环境式的 workforce，一个环境式的语音控制操作系统。

It's an ambient workforce, an ambient voice-controlled operating system.

我去散步，跟 AI 说话，发起多个电脑任务，梳理我一天的安排，还改进我的系统架构，全都通过自然语言和不停地聊天完成。

I go on walks, talk to AI, fire off multiple computer tasks, triage my day, and improve my system architecture all via natural language and yapping.

而在另一篇帖子里，她还讲到，她怎么在路上使用它，或者周末把它当作后台进程，让事情持续推进，同时她还能出门享受生活、散步、徒步，等等。

And in yet another post, she discussed how she uses this on the go or as a background process during the weekend to keep things moving even as she's out enjoying her life, taking walks, hiking, et cetera.

现在，关于 Allie K. Miller 的 Codex 经典内容，最后再补充一条：她其实在 X 上写了一份完整的设置指南并发布了，我会把它放进节目说明里，这样你就可以自己去试试。

Now, one final entry to the Allie K. Miller Codex canon, she actually wrote and published a full setup guide on X, which I will include in the show notes so you can go try it out for yourself.

我自己已经被语音模式种草有一阵子了，而我想说的是，这类东西光靠描述，你其实很难真正体会到。

Now I've been voice-pilled for a while, and what I will say is that this is the type of thing that you're not really going to get from just the descriptions.

值得给自己设定一段时间，用一种新的方式去和它互动，哪怕一开始觉得有点怪，也看看它是否会改变你的工作方式、怎样改变，以及你喜不喜欢这种变化。

It's worth committing to some period of time where you're interacting in a new way, even if it feels weird at first, to see if and how it changes how you work and whether you like that shift.

下一个技巧，是教 AI 你的工作流程。

Next technique is teaching AI your workflows.

这是我们上周一期节目里讨论过的内容，但随着 computer use 方面的进展越来越多，让 AI 承担你工作中更高级的部分，也变得比以往更容易，因为你可以直接让它观察你现在是怎么做事的。

This is something we discussed in an episode last week, but increasingly because of advances in computer use, it's getting easier than ever to get AI to do more advanced parts of your work because you can simply let it watch how you currently do things.

有两个很重要的例子，说明这件事已经在哪里变得可用了。

Two big examples of where this has become available.

第一个是 ChatGPT 的 computer history 功能。

The first is ChatGPT's computer history feature.

Computer history 会保留一条时间线，并记录和回顾你具体做了什么、你具体是怎么工作的，这样你就可以把它转化成持续的、可重复的流程。

Computer history keeps a timeline and a review of the specific things that you do and the specific ways that you work so that you can turn that into ongoing and repeatable processes.

这也是新 GrokBot 里最让人兴奋的功能之一。

This was also one of the most exciting features from the new GrokBot.

而且我们几分钟后会看到，人们其实还在真正探索，怎么才能把它的价值最大化。

Which, as we will see in a few minutes, people are still really just exploring how to get the most out of.

不过，GrokBot 最明显有用的地方之一，是你可以按一个小按钮来教它完成一项任务，方式就是让它专门观察你正在做的那一整套特定操作。

One of the most obviously useful parts of GrokBot, however, is the fact that you can press a little button to teach it a task by having it specifically watch that particular set of activities that you're doing as you're doing them.

所以，computer history 算是一种更偏环境式的任务教学方法，而 GrokBot 则是一种更有意识、更主动的方法，你会明确告诉它什么时候看、看多久。

Computer history then is sort of an ambient approach to teaching a task, while GrokBot is a much more deliberative approach where you specifically tell it when to watch and how long to watch for.

不过在这两种情况下，如果有些复杂的事情，以前你会觉得 AI 做不了，因为太难解释到底该怎么做，那么简单说，这件事可能已经发生了很大的变化。

In both cases though, if there are complex things that previously you would've assumed that AI couldn't do because it was too hard to explain how to do it, the TL;DR is that that might've changed significantly.

第三，创建一个 skill，让你的 AI 写作变得更好。

Number three, create a skill to make your AI writing better.

那说到 AI 写作，我这里不打算卷入一个争论，就是写作作为思考本身的价值，以及用写作来理清自己对某件事的想法到底有多重要。

Now, when it comes to AI writing, I am not wading into the debate here about the value of writing as thinking and the importance of using writing as a way to figure out your own thoughts about something.

我说的是那种数量巨大、而且我们都很愿意外包给 AI 的写作。

I'm talking about the sheer voluminous writing that we do that we are all very comfortable outsourcing to AI.

比如 emails、memos、summaries，或者更重要一点的内容，像 website copy。

Emails, memos, summaries, or more important areas like website copy.

虽然 AI 的写作在很多情况下已经变好了，但还是有太多套路，会让 AI 写出来的东西一眼就能被认出来。

Even though AI's writing has in many cases gotten better, there are still so many tropes that make AI's writing instantly recognizable.

比如那种特别短、特别压缩的句子，只有一两个词，然后用句号隔开，好像这样就显得很有戏剧感。

Tiny compressed sentences of just a word or two separated by periods to sound dramatic.

不是这个，而是那个。

It's not this, it's that.

只要出现所谓的“诚实的保留意见”或者“担忧”，你就知道又来了。

Anytime there's a quote-unquote honest caveat or concern, you know the drill.

基本上，只要你曾经对一个 LLM 说过，把这段重写一下，不要有这些 AI 味儿，我觉得其实有一个稍微更好的做法。

Basically, if ever you found yourself saying to an LLM, rewrite this without all the AI-isms, I think there's a slightly better way to do that.

Ruben Hassid 最近在 X 上发了一条帖子，讲了一些使用 AI 的明显破绽。

Ruben Hassid recently posted on X about some of the dead giveaways of using AI.

其中很多就是我刚才提到的那些，另外他还补充了，比如自我鼓掌。

A bunch of them are the ones that I just mentioned, and he also adds, for example, clapping for itself.

而这很重要。

And that matters.

这就是大家都漏掉的部分，而这恰恰就是重点。

That's the part everyone misses, which is exactly the point.

你完全可以把这条帖子打印成 PDF，然后直接丢进 Claude，让它把这个做成一个 skill，这样以后就不要再出现这些问题。

You can genuinely print this post as a PDF and just drop it into Claude and ask it to make it a skill so that that doesn't happen anymore.

其他人也在尝试不同版本的做法，比如给 AI 一些具体的 style guides，像飞机手册用的 ASD-STE100 writing standard，然后把这些也转成一个 skill。

Other people are experimenting with different versions of this, like giving the AI specific style guides like the ASD-STE100 writing standard for plane manuals and turning those into a skill.

简单说，把你不希望 AI 做的事情集中整理到一个地方，并且确保它每次写作的时候都把这些记在心里。

TL;DR, capture the things you don't want AI to do in a consolidated place and make sure it has that in mind as it's writing every time.

第四个技巧其实是一个新功能，也就是 Claude 里的 /design command。

Technique number 4 is actually a new feature, which is the /design command in Claude.

当你使用 Claude Code 的时候，现在可以输入 /design，它会调出一个 artboard workflow，给你一种不同类型的界面，来处理设计问题。

When you're using Claude Code, you can now type /design and it's going to bring up this artboard workflow that gives you a different type of interface for interacting with a design question.

你可以给它留下具体备注，高亮你想修改的特定区域，还可以预览不同风格。

You can leave it specific notes, highlight particular areas you want to change, preview different styles.

有意思的是，以我的经验来看，它在宏观和微观两个层面上，都改善了设计体验。

And what's interesting about it is that in my experience, this is an improved design experience on both the macro and the micro level.

在宏观层面，如果你不知道一个东西应该怎么运作，让它设计出一堆模板，让你可以一次性看完，并给出相当高层次的反馈，再一起迭代，这是一种非常有价值的方法。

On the macro level, if you don't know how you want a thing to work, having it design a bunch of templates that you can look at all at once and give pretty high-level feedback on to iterate with is a really valuable approach.

但到了微观层面，能够编辑一个设计里非常具体的部分，而不是只能写一个很大的 prompt，让它又把整个东西重新改一遍，这一点也极其有价值。

But then also when you get down to the micro, being able to edit very specific parts of a design rather than having to just give a big prompt that changes the whole thing all over again is incredibly valuable.

现在 Claude 和 ChatGPT 一直在不断加入这种新 command，所以始终值得关注一下，但 /design 是我已经觉得非常有用的一个功能；如果你是 Claude 用户，我想你可能也会觉得有用。

Now Claude and ChatGPT are constantly adding new commands like this, and so it's always worth keeping an eye on that, but /design is one that I've found really useful already, and I think you might too if you are a Claude user.

说到 skills，agent skills 简单来说就是一个新的学科。

Now speaking of skills, agent skills are simply put a new discipline.

随着我们给 agents 分配越来越复杂的任务，知道应该在哪里插入 skills，给它哪些 skills，什么时候退一步，让它用原生能力自己工作。

As we give our agents more and more complex tasks, knowing where to insert skills, which skills to give it, where to back off and just let it work natively.

所有这些，都会变成非常重要的人类能力，或者说 agent management 能力。

All of those become really important human capabilities or agent management capabilities.

你能找到的最好的 skills 资源库之一，包括怎么使用这些 skills 的资源，来自 Matt Pocock，他的网站是 aihero.dev，具体来说是 aihero.dev/skills。

One of the best repositories of skills that you can find, as well as resources for how to use them, comes from Matt Pocock over at his site aihero.dev, specifically aihero.dev/skills.

他会按照你实际会在什么时候用到这些 skills 来分组：比如入门、主流程、塑形、维护，等等。

He groups the skills there based on when you would actually use them: getting started, the main flow, shaping, upkeep, etc.

所以举个主流程里的 skill 的例子，他有一个 skill 叫 Grill with Docs。

So as an example of a skill in the main flow, he's got a skill called Grill with Docs.

这个 skill 会围绕一个计划或者一个设计来采访你，一直到你和 agent，引用一下，‘对它有了同一份理解’，并且在这个过程中，把词汇表和那些艰难的决策写进你的 repo 里。

It's a skill that interviews you about a plan or a design until you and the agent, quote, share one understanding of it, and writes the vocabulary and the hard decisions into your repo while it does.

重要的是，它不只是一次采访，而是一次会留下交付物的采访。

Importantly, it is not just an interview, but an interview that comes with a leave-behind.

而且在每一个案例里，Matt 不只是解释这个 skill 是做什么的，他还给你一个简单的、可以直接复制的安装方式，让你把这个 skill 直接装上。

And in each of these cases, Matt is not just explaining what the skill does, he's giving you a simple, copyable way to install the skill directly.

skill hygiene，skill practice，我觉得这类事情不应该被看成一个二元选择，说我们到底会还是不会，而应该把它看成一种持续的训练，我们一直都在努力变得更好。

Skill hygiene, skill practice is one of those things that I think we should not view as a binary between whether we can or whether we can't, but instead just an ongoing discipline that we are always trying to get better at.

而且这个领域里，你能做的最好的事，绝对就是站在那些已经花了大量时间搞清楚什么有用、什么没用的人的肩膀上。

And this is definitely an area where the best thing that you can do is stand on the shoulders of the people who have already spent a ton of time figuring out what is and isn't working.

下一个你可能还没试过的 AI 技术，是我今年秋天会花更多时间去研究的一个方向。

The next AI technique that you might not have tried yet is one that I'm gonna be spending a lot more time on this fall.

有些人把它叫做 multiplayer AI，或者你也可以把它理解成 team agents。

Which some people are referring to as multiplayer AI, or which you could think about as team agents.

这个想法其实很简单。

And the idea of this is pretty simple.

随着今年 agents 真正变得可用了，我们大多数时候还是在单人模式下和它们互动。

As agents have become real this year, we've mostly interacted with them in single-player mode.

我的意思是，Claude Code 出来之后，你买了一台 Mac Mini，你搭了一个个人 chief of staff，你搭了 research agents，或者 developer agents，或者你搭了其他任何 agents。

What I mean by that is that when Claude Code came out, you got a Mac Mini, you built a personal chief of staff, you built research agents or developer agents or whatever agents you built.

你一下子就成了一个 agent 团队的管理者，也许这些 agents 都在为你工作，但它们是在为你一个人工作。

You all of a sudden were the manager of perhaps a team of agents working for you, but they were working for you.

我觉得很多人正在发现，在公司这个场景里，这其实留下了一个很大的机会没有被利用。

I think what a lot of folks are finding is that in the context of their companies, this leaves a big opportunity on the table.

工作并不只是每个人先听完 brief，然后各自散开去做自己的事，最后再回来汇合。

Work is not just a matter of everyone getting briefed and then going off and doing things on their own and then coming back together.

工作往往是高度协作的，有交接，有共享上下文，也有共享空间，工作就是在这些地方发生的。

Work is often highly collaborative with handoffs and shared context and shared spaces where that work happens.

这还处在很早期，但我认为，接下来一个很大的趋势，会是 agentic tools、交互模式和平台，它们关注的是存在于团队交汇处的 agents，而不只是个人各自工作时使用的 agents。

It's nascent, but one of the things that I think is going to be a big trend moving forward is agentic tools and interaction patterns and platforms that focus on agents that live in the intersection where teams come together rather than just where individuals work on their own.

到目前为止，也许最好的例子就是 Claude in Slack。

Maybe the best example of this so far is Claude in Slack.

其实在 Claude in Slack 六月底发布之前，Claude 和 Slack 之间就已经有过一个集成，你可以通过 tag 它，把你的 Claude instances 叫出来。

Now, even before Claude in Slack came out at the end of June, there had been an integration between Claude and Slack where you could summon your Claude instances by tagging it in.

Claude in Slack 的不同之处在于，Claude 是以团队成员的身份加入进来，并且可以访问你把它召唤进来的那个特定 channel 的完整上下文。

The difference with Claude in Slack is that Claude joins as a team member with access to the entire context of a particular channel that you're summoning it in.

存在于那个 channel 里的 Claude instance，可以有特定的权限、特定的工具访问权限、特定的上下文访问权限，而且这些都可以跟其他 channels 里的 Claude 不一样。

The Claude instance that lives in that channel can have specific permissions, specific tool access, specific context access, and that can be different than the Claude in other channels.

但存在于那个团队 channel 里的 Claude，并不只是活在某一个团队成员的电脑上。

But that Claude that exists inside that team channel doesn't just live on one team member's computer.

它是整个团队共享的一项资源。

It is a shared resource across the team.

当然，现在不是每个人都能用上 Claude in Slack，但我认为，这种在团队内部共享 agents 的使用模式，会成为今年秋天 AI 最大、也最重要的趋势之一。

Obviously not everyone has access to Claude in Slack right now, but I think that this pattern of use of agents that are shared across teams is going to be one of the biggest and most important trends for AI this fall.

另一个关于人们如何思考这个方向的例子，来自 Tenex，这是一家位于 New York City 的 AI 构建和实施合作伙伴。

Another example of how people are thinking of this comes from Tenex, which is an AI build and implementation partner out of New York City.

他们最近发布了一篇文章，讲的是他们所说的 Citizen SDLC，这是一个六阶段的生命周期，目的是把非技术员工用 AI 做出来的东西，从个人 prototype，推进到整个公司都能使用的 production。

They recently released a post about the Citizen SDLC, as they call it, which is a six-stage lifecycle that takes what non-technical employees build with AI from a personal prototype to a production that the whole company can use.

换句话说，它是在试图把发生在个人层面的 enterprise vibe coding，变成对整个公司都有用的东西。

In other words, it's trying to take enterprise vibe coding that happens on an individual level and make it useful across the whole company.

这篇文章可以在他们的 blog 上找到，网址是 tenex.co，而且就在页面顶部附近有推荐。

That one you can find on their blog, which is tenex.co, and it's promoted right near the top.

你可能还没试过的第七个技巧或者功能，有点像是偷懒的答案，也有点像是一个总括性的答案，那就是 ChatGPT Agent。

Technique or feature number seven that you might not have tried yet is a bit of both a cop-out and a catch-all, and that is ChatGPT Agent.

现在呢，ChatGPT Agent 目前还是一个单人模式的工具，也就是说，你是在为自己不同的任务启动自己的 ChatGPT agents，但我觉得它不会一直停留在单人模式太久。

Now, right now ChatGPT Agent is still a single-player mode tool in that you are spinning up your own ChatGPT agents for different tasks that you have, but I don't think it's going to stay single-player for long.

现在你启动的 ChatGPT agents 已经可以彼此互动了，这会产生一些非常有意思的模式，比如你可以只跟一个 chief of staff 互动，由它来管理其他所有 ChatGPT agents。

Already the ChatGPT agents that you spin up can interact with one another, which creates some really interesting patterns like being able to interact only with a chief of staff, which manages all the other ChatGPT agents.

我觉得很明显的下一步，就是让一个团队里不同成员的 ChatGPT agents 之间进行共享互动。

And I think that the obvious next step is shared interactions between ChatGPT agents across different members of a team.

现在大概才过了一周，人们已经在大量实验，看看怎么才能把它用到最好。

Now it's been about a week and people are doing a ton of experimenting with how to get the most out of it.

大家反应非常积极的一点是，它有自己的虚拟电脑，可以用来解决某些类型的访问问题，而这些问题以前一直困扰着之前的 agents。

People are responding very positively to the fact that it has its own virtual computer that it can use to solve certain types of access issues that plagued previous agents.

如果你去 X 上随便翻一翻，就会看到人们分享了海量的使用场景。

And if you go poke around on X, there are just a boatload of use cases that people are sharing.

比如申请审核、根据笔记更新销售 deck、Salesforce 报告、每日简报。

Application reviews, sales deck updates from notes, Salesforce reports, daily briefings.

我看到的一个共同点是，只要你会以一种规律、可预测的方式处理大量信息，或者本来可以这样处理，那么像 ChatGPT Agent 这类的 agent，就很值得你去看看。

One common thread that I'm seeing is that anywhere you interact or could be interacting with a lot of information in a regular and predictable way, a ChatGPT Agent-style agent is something that might be interesting to check out.

Lenny's Podcast 的 Lenny Rachitsky 最近提出了一个想法：用一个包含他五百多期播客全部文字稿的 MCP，把 ChatGPT Agent 变成你自己的产品策略、增长和职业顾问。

Lenny Rachitsky from Lenny's Podcast recently gave the idea of using the MCP that has all the transcripts from more than 500 of his podcasts to turn ChatGPT Agent into your own product strategy, growth, and career advisor.

顺便说一句，如果你们当中有人因为各种原因不想用 Grok，我们就这么说吧。

And by the way, for those of you who are not interested in working with Grok for whatever set of reasons, let's call them.

Nous Research 最近为 Hermes Desktop 推出了 Bot Mode，它和 Grok Bot 非常相似，但它是通过开放的、可控性强得多、可定制性也强得多的 Hermes 系统来实现的。

Nous Research recently introduced Bot Mode for Hermes Desktop, which is very similar to Grok Bot, but through the open, much more controllable, much more customizable Hermes system.

你可能还没试过的第八个技巧，是本地 AI。

Technique number 8 that you might not have tried is local AI.

当然，要深入讲清楚使用本地 AI 到底需要什么，已经远远超出了本期节目的范围，不过你们很幸运，早在六月，我就和 Nufar Gaspar 做过整整一期节目，讲为什么本地 AI 很重要，以及怎么使用它，你可以去听听看。

Now it's too far beyond the scope of the show to go really deep into what using local AI actually requires, but luckily for you, back in June, I did an entire episode about why local AI matters and how to use it with Nufar Gaspar that you can go check out.

之所以现在值得重新思考本地 AI，是因为能在本地硬件上运行的模型，最近出现了一些相当重要的进展。

The reason that it's worth thinking about local AI again is that there are some pretty serious advances in models that are capable of being run on local hardware.

具体来说，我指的是 Qwen3-8B，这是一个可以在常见硬件上运行的本地模型，它在 Artificial Analysis Intelligence Index 上拿到了五十二分，而这个成绩放在几个月前还会是最先进水平。

Specifically, I'm referring to Qwen3-8B, which is a local model that can run on common hardware that is scoring a 52 on the Artificial Analysis Intelligence Index, which would've been state-of-the-art just a few months ago.

换句话说，如果你一直在考虑尝试本地 AI，那么 Qwen3-8B 可能就是一个真正开始上手的好机会。

In other words, if you have been thinking about experimenting with local AI, Qwen3-8B might be the right context to actually dive in.

最后，第九个与其说是技巧，不如说是一个启发思路的东西。

Lastly, technique number 9 is not so much a technique, but a thought starter.

那就是可以试试的两个词提示词。

And these are 2-word prompts to try.

这又来自 Allie Miller，她最近发了一条 tweet，分享了十八个她有点着迷的两个词提示词。

Once again, this comes from Allie Miller, who recently tweeted out 18 2-word prompts she's kind of obsessed with.

你会觉得，其中一些其实也来自她使用 voice mode 的经验，因为它们听起来很像你跟团队成员互动时会说的话。

You have to think that some of these actually come out of her experience using voice mode as well, as they sound a lot like how you would interact with a team member.

事实上，她的第一个两个词提示词是，now what。她说，当你刚完成一个项目或者一轮大冲刺，但你还有精力、想让 AI 再给你更多事情做的时候，这个提示词特别好用。

In fact, her first 2-word prompt, now what, which she says is great for when you've wrapped up a project or big push and you still have energy and want AI to give you more.

这让我很想起 The West Wing 里的 Jed Bartlet 说的那句，what's next？

Reminds me a lot of Jed Bartlet from The West Wing saying, what's next?

她提到的其他一些两个词提示词包括，please fix，通常会配上一张标出问题的截图；simulate it，用来让 AI 跑场景、为边界情况做规划，并且在规划界面里可视化可能会发生什么。

Some of the other 2-word prompts she points to are, please fix, usually accompanied with a screenshot flagging an issue, simulate it to get AI to run scenarios, plan for edge cases, and visualize what might happen in a planning interface.

还有 remember this，也就是当 Allie 遇到一个错误或者关键上下文错误时，她会强制 AI 把它记录到记忆里。

And remember this, where when Allie has run across a mistake or critical context error, she forces the AI to log it in its memory.

她指出，虽然 AI 经常会自动这么做，但并不总是会，所以明确地要求它记住，是有帮助的。

She points out that while AI often does this automatically, it doesn't always, and being specific about it can help.

这也许就是我一开始所说内容的最佳例子：这些东西不一定会给你的工作流带来什么巨大变化，而只是一些简单的小想法，帮你从现在每天都在用的工具里多榨出一点价值。

This is maybe the best example of what I was saying at the beginning, where these aren't necessarily some massive change to your workflow, but just some simple little ideas to get a bit more out of the tools that you now use every day.

好了，这就是九个你可能还没试过的 AI 技巧，至少不太可能全都试过。

So there you have it, 9 AI techniques you probably haven't tried, at least not all of them.

希望这能给你一些有趣的想法，让你在即将到来的周末做点实验。

Hopefully this gives you some fun ideas for experimenting in the weekend to come.

那么今天的 AI Daily Brief 就到这里。

For now, that is gonna do it for today's AI Daily Brief.

一如既往，感谢你的收听或收看，我们下次见，peace。

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

已剔除 2 处广告（点击展开查看）

M10:57广告 · 已剔除

First of all, thank you to today's sponsors.

KPMG, Blitzy, Harvey, and HyperAgent.

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

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

And while you're on aidailybrief.ai, you can check out the link to next week's free webinar about agentic loops for knowledge workers.

If you have heard me or others talk about loops as basically the second coming but aren't exactly sure how to apply them to your work, this free webinar is for you.

I will be there.

Nufar Gaspar will be leading most of it.

You can register for free, and even if you can't make it, we will send you the recording after.

Again, all that information is at aidailybrief.ai.

M111:34广告 · 已剔除

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.

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.

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

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

Blitzy's deep codebase understanding unlocks the thing every roadmap owner cares about: shipping new features.

Here's the truth about building inside a massive enterprise codebase.

Writing code was never the bottleneck—context is.

Which system does this touch, which contracts can't break, which standards apply?

Blitzy already knows because it reverse-engineered your entire codebase into a dynamic knowledge graph before feature work began.

With that complete picture, Blitzy builds features end-to-end—architecture, APIs, UI, and tests all validated against your existing systems.

One Blitzy customer built an AI-native application from scratch with 100% autonomous completion, saving over 2,700 engineering hours.

Features that respect your codebase instead of fighting it.

Stop letting your backlog grow faster than your team.

Accelerate your roadmap at blitzy.com.

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

If you listen to this show, you likely have a thesis.

Maybe it's enterprise adoption, maybe it's compute, maybe it's a specific lab.

Harbor Capital's AI Lab Ecosystem ETFs let you express it via five actively managed ETFs, each seeking exposure to the ecosystem around one major lab.

Anthropic, OpenAI, DeepMind, Meta, or xAI.

Your view of the AI race in ETF form.

Harbor Capital Advisors' AI Lab Ecosystem ETF suite gives investors a way to invest in the AI ecosystem they believe is best positioned for success.

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 Foreside 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.

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

New users get $1,000 in inference.

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

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

Marketing's agent turns competitor moves into landing pages.

Sales' agent enriches leads, drafts emails, and updates the CRM.

Ops' agent chases the paperwork and tracks the budget.

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

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

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

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