马克·安德森如何实际运用人工智能

这可能是迄今为止最具民主性的技术（小写的民主），因为世界上最好的人工智能完全可以通过任何人都能下载的应用程序获取。

This is already probably the most democratic, you know, small d technology of all time in the sense of the very best AI in the world is fully available on the apps that anybody can download.

这完全是一种不同类型的计算机，具有说实话更像人类的特性，即它在大多数情况下都是正确的。

This is just a completely different kind of computer that has these characteristics that are frankly more like a person, which is it's right most of the time.

它偶尔会出错。

It occasionally gets things wrong.

当它出错时，它能够进行自我批评。

When it gets things wrong, it's able to self critique.

你必须以与人合作的方式来与它合作。

And you have to kind of work with it the way that you work with a person.

你要利用它富有创造力这一优势，同时也要容忍它并不总是正确这一事实。

You wanna take advantage of the fact that it's creative, and then you wanna be tolerant of the fact that it's not always correct.

人工智能基本上将所有事物都重新拉回到了我所在位置20英里半径范围内，程度令人难以置信。

AI basically has snapped everything right back into the 20 mile square radius around where I sit to just an incredible degree.

所以我想说，西方几乎所有真正有趣的人工智能公司现在都集中在硅谷这个原点。

So I would say, like, almost 100 of the actually interesting AI companies in the West are happening at sort of ground zero right here in Silicon Valley.

某地的面包店老板和谷歌CEO使用着同款人工智能，而据马克·安德森所言，这位面包店老板正占据上风。

There's a bakery owner somewhere using the same AI as Google's CEO, and according to Marc Andreessen, the bakery owner is winning.

这位发明了现代网页浏览器并创建了价值数十亿美元公司的男人，刚刚揭示了一个非凡的现象。

The man who invented the modern web browser and built multibillion dollar companies just revealed something remarkable.

人工智能正在社会结构中逆向渗透。

AI is spreading backwards through society.

先是个人用户，再是小微企业，接着是财富500强公司，政府机构最后才跟进。

Individuals first, small businesses second, Fortune 500 companies third, government last.

这与计算机从大型机发展到智能手机的演进路径完全相反。

The exact opposite of how computers evolve from mainframes to smartphones.

马克表示，已有五亿人在手机上用着全球最尖端的人工智能，问题在于：为何多数人仅用来写邮件，而只有少数人用来建立商业帝国？

Mark says half a billion people already have the world's most sophisticated AI on their phones, So the question is, why are most using it to write emails while only some are using it to build empires?

今天我们将分享马克与布鲁斯在马克·哈尔彭的节目《NextUp》中的对话。

Today, we're sharing a conversation Mark and Bruce had with Mark Halperin on his show NextUp.

他们探讨了将AI转化为世界级顾问的具体指令技巧，为何硅谷在经历五年分散后又重新收缩回20英里半径范围，以及中美人工智能竞赛中令人不安的真相。

They talk about the specific prompts that transform AI into a world class adviser, why Silicon Valley just snapped back into a 20 mile radius after five years of dispersion, and the uncomfortable truth about America's AI race with China.

希望您喜欢。

We hope you enjoy.

好的。

Alright.

接下来有请马克·安德森，创新者、创造者，以及一位极其成功的商人。

Next up, Marc Andreessen, innovator, creator, and damn successful businessman.

早期，他发明了Mosaic互联网浏览器，共同创立了网景公司。

Early on, he invented the Mosaic Internet browser, cofounded Netscape.

自那以后，他一直是许多非常成功公司的推动力量与投资人，其中不乏市值数十亿美元的企业。

And since then, he has been the, animating force and, investor behind a lot of very successful companies, including some at the multiple billion dollar level.

他共同创立了自己的公司安德森·霍洛维茨基金，并担任普通合伙人。

Co founded his firm, Andreessen Horowitz, a general partner there.

他们涉猎广泛，在各个领域都有建树，而他本人也博学多识。

And they do a lot of lot of stuff about a lot of stuff, and he knows a lot about a lot.

马克，欢迎你。

Mark, welcome.

谢谢，马克。

Thank you, Mark.

很荣幸能来到这里。

It's great to be here.

非常高兴你能来。

Really happy to have you.

关于人工智能，我有太多想和你探讨的，我们会花大量时间在这上面。

So much about AI I wanna talk to you about, so we're gonna spend a lot of the time on that.

首先，现在很容易让人想说——当我思考AI时，我在想我们处于什么阶段？未来会怎样？

First off, it's tempting to say right now, and when I think about AI, think about where are we now and where are we going?

人们很容易认为这就像是高学历聪明人正在适应AI，而另一些人在工作中难以具备这种能力。

It's tempting to say it's between like really smart, highly educated people who are adapting to it, and then people who just don't have the capacity to do that in their jobs easily.

但我发现，即便是从事我这类工作的人——那些受过良好教育、享有特权的人群中，也存在数字鸿沟。

But what I'm finding is at people who do what I do, people who are well educated, very privileged, there's a have nots there.

我在使用AI方面还是个新手。

I'm a baby using it.

我使用得并不算很精通，但相比我的一些同行，我已经算是爱因斯坦级别了。

I'm not using it very sophisticatedly, often, but I am Einstein compared to some of my counterparts.

我想知道，你也是这样看的吗？

And I'm wondering, is that how you see it?

你认为是什么区分了那些即使现在也明白其强大之处的人，和那些似乎对此毫无察觉的人？

And what do you think differentiates those who understand how powerful it is even now versus those who seem oblivious to it?

是的。

Yeah.

所以我觉得这里面有个虚假的说法，也有个真实的说法。

So I think there's kind of a fake story, and then there's a real story.

虚假的说法就是那种经典的马克思主义论调，说什么只有富人才拥有它。

So the fake story is kind of the classic Marxist story, is, know, oh, only the rich people have it.

只有那些自以为是的人才拥有它。

Only the sassy people have it.

你知道的，生物科技公司才拥有它。

You know, the biotech companies have it.

其他人基本上都会被冷落在一边。

Everybody else is kind of gonna be left out in the cold.

实际上情况并非如此。

That's not actually what's happening.

而且现在已经有大量数据，这些公司已经发布的数据可以支持我接下来要说的观点。

And and and there's a lot of data on this now that have been has already has been released by these companies to justify what I'm about to say.

所以真实情况是，这很可能已经是历史上最具民主性（小写的d）的技术了，因为世界上最顶尖的AI技术完全可以通过任何人都能下载的应用程序获取。

So the real story is this is already probably the most democratic, you know, small d technology of all time in the sense of the very best AI in the world is fully available on the apps that anybody can download.

你可以随便选——ChatGPT、Claude、Gemini、Grok、Mistral，还有中国的DeepSeek等等。

And, you know, take your pick, Chad, GPT, Claude, Gemini, Grok, you know, Mistral, any of these by the way, DeepSeek from China.

你下载其中任何一个应用，

You download any of these apps.

就能获得世界顶级、最复杂强大的AI能力。

You're getting state of the art, like, the full, most sophisticated, powerful AI capability in the world.

而且要知道，这些应用的下载量已经接近五亿，正在向十亿迈进。

And, you know, the number of people already who downloaded these apps is parts of a half billion on its way to a billion.

个人用户正在逐步摸索如何将这些技术融入他们的日常生活中。

And individual people are figuring out, you know, basically how to incorporate this in their lives.

从数据中可以看到，或许正如你所预料的那样，有一部分人几乎无时无刻不在使用这些新系统，全天候处理所有事务。

And what you see in the data is there's, you know, maybe what you'd expect, which is there are a slice of people who just use these new systems all the time, like, literally all day for everything.

许多案例显示，他们反馈从中获得了巨大收益。

And, you know, in a lot of cases, they're reporting that they're getting enormous benefits from that.

还有大量人群正处于尝试和探索阶段。

And then there's a lot of people who are experimenting and trying to figure it out.

当然也存在部分人群出于各种原因对此不感兴趣或未参与其中。

And then there are people who are just, you know, not, you know, for whatever reason, not interested or not engaged.

但不得不说，这项技术从问世之初就展现出惊人的普及广度。

But I would say it's incredible out of the gate how distributed this technology already is.

坦白说，即便以我的资源和人际关系，也无法获得比应用商店下载版本更优越的AI。

And then I could just say like, I don't have that, you know, with all my resources and with all my connections, I don't have access to better AI than the one that you just download off the App Store.

因此我认为这堪称一个奇迹——全球最尖端技术从诞生起就对所有人触手可及。

And so I think this is actually like just an incredible story of the most advanced technology in the world being available to everybody right out of the gate.

你得让人们使用它才能发挥它的优势，对吧？

You gotta get people to use it to take advantage of it, right?

我有个朋友写了本书，拿给他的经纪人看。

So I've got a friend who's drafted a book and took it to his agent.

经纪人说，这书有14万字。

And the agent said, it's 140,000 words.

你得把它删减到7万字。

You gotta cut it down to 70,000.

我说如果你手动操作，就算雇人帮忙也得花上几个月。

And I said, if you try to do that by hand, even if you hire someone, it's gonna take you months.

我一个小时就能搞定。

I can do it in an hour.

我可以告诉AI：不要改变风格，不要改变语气，不要删减任何影响故事完整性的内容。

I can say to AI, don't change the style, don't change the tone, don't cut anything that ruins the story.

结果他说：这太不道德了。

And he said, that's immoral.

出版商会因为我那样做而生气。

The publisher will be mad that I did it that way.

这样做不对。

It's not right.

我该怎么回应那个人呢？

How should I answer that person?

是的，同样的争论在多年前‘Right’出现时就发生过，对吧？

Yeah, so the exact same arguments emerged years ago with the introduction of Right?

其实我最近一直在陪孩子看很多老科幻电影，你知道1982年那部著名的《电子世界争霸战》吗？它是第一部真正使用电脑图形的电影，却因为使用电脑技术而被取消了奥斯卡特效奖的资格。

I I actually just I've been I've been watching a lot of, you know, old science fiction movies with my kid, and, you know, there's this famous science fiction movie Tron from 1982, know, that had the first computer kind of the first real computer graphics movies, and it was disqualified for an Oscar for special effects because it used computers.

对吧？

Right?

所以这种传统由来已久，每当有新工具出现时，人们总觉得它不够正统。

And so there there's this kind of long tradition of, like, whatever the new tool is, it's sort of illegitimate.

总觉得它肯定有问题。

There must be something wrong with it.

在这个具体案例中，我意思是，毫无疑问有些人确实和你朋友有同样的感受。

In this particular case, I mean, there's no doubt there are people who feel the way that your friend feels.

这个问题实际上在最近的好莱坞罢工中就已经争论过了——在电影和电视行业的创意领域。

That issue was actually litigated in the last Hollywood strikes in the in the in in in I'm sorry, you know, the movie on the film and TV side of the of the creative profession.

是的。

Yeah.

上次罢工其实挺有意思的。

The last time it strikes actually it's actually funny.

他们最开始是为流媒体罢工，后来转向了AI问题，接着就发起了关于AI的罢工。

They started streaming strikes, then AI and then they they began AI strikes.

但制片厂与工会达成的共识其实是这样的：如果你是编剧并使用AI，这完全没问题。

But the sentiment with the studios actually with the unions was that the following, which is if you're a writer and you use AI, that's totally fine.

不被允许的是制片方用AI生成内容后却声称这是编剧写的。

What's not allowed is for the studio to basically use the AI and then claim it was a writer.

但基本上好莱坞工会和制片厂达成的共识是：这只是另一种工具而已。

But but basically, what what the what the unions and studios in Hollywood decided was it's another tool.

它就像文字处理器一样。

It's like the word processor.

它就像海报制作电脑一样。

It's like the poster computer.

就像，你知道的，就像用打印机代替手写稿子。

It's like, you know, it's like using a printer instead writing out a manuscript by hand.

所以我认为确实有人和你朋友感受相同，但我觉得世界已经非常快速地适应了使用它。

And so so I think there are people who feel like your friend does, but I think the world is is already adapting very fast to using it.

而且坦率地说，你可能向你朋友指出的原因之一就是：现在甚至没人能分辨出来了。

And and frankly, one of those reasons you probably pointed out to your friend is like, don't even know that anybody could tell anymore.

对吧？

Right?

所以，如果你要对人们可以悄悄做而无人知晓的事情进行道德禁止，你觉得这真的能奏效吗？

And so, you know, if you're gonna have a moral prohibition on something that people can just do and nobody knows about, like, that, you know, is that really gonna work?

因此我认为这类自我设置的障碍很可能会迅速瓦解。

And so I I think those those sort of self imposed barriers are are probably gonna collapse quite quickly.

希望如此。

I hope so.

《财富》杂志发布了一份美国顶尖企业应用AI的榜单。

Fortune put out a list of the top American companies using AI.

字母表（Alphabet）、维萨（Visa）、摩根大通（JPMorgan Chase）位列前三。

Alphabet, Visa, JPMorgan Chase, the top three.

我们刚刚讨论的是个人层面的应用。

We talked about just now on an individual level.

如果你是公司CEO或为CEO提供建议，登上这份榜单有多重要？

If you were the CEO of a company or advising a CEO, how important is it to be on that list?

换句话说，让公司适应AI技术——无论是内部使用还是面向消费者——有多重要？

In other words, how important is it to get your company to adapt, whether it's for internal or consumer facing?

当前阶段这有多紧迫？

How important is it right now?

确实。

Yeah.

这实际上回到了我们最初讨论的起点，公平地说，旧有的适应模式——当计算机刚出现时，是大型机构先获得技术，然后其他人才陆续跟上。

So this this goes back to actually where we started, which is and in fairness, the old model of adapting actually computers when computers came out, the old model was the the largest institutions get technology first, and then everybody else gets it later.

因此，计算机的普及路径是：政府最早在1940年代开始使用大型主机，随后大公司在1950至1960年代引入了计算机和主机系统。

And so, you know, the way the computer rolled out was the government actually got mainframe computers first starting in nineteen forties, and then big companies got computers in the mainframes in the nineteen fifties, nineteen sixties.

小型企业则在1970年代开始使用当时所谓的微型计算机。

Small companies started to get computers, mini what we call mini computers at the time, the nineteen seventies.

而我们个人直到1980年代才拥有个人电脑。

And then and then, you know, we as individuals only got, you know, PCs in the nineteen eighties.

也就是说，技术从全球最大规模的组织向下渗透到小企业和个人，整整耗费了40年时间。

And so it took 40 for basically technology to cascade down from the largest organizations in the world to small businesses and to the individual.

但AI技术正走向相反方向——正如我所说，如今最尖端的功能已通过消费级应用触手可及。

This technology, AI, is is is going the opposite, you know, which is like like I said, the the most sophisticated capabilities are available on the consumer app today.

我们发现消费者——即个人用户——的适应速度是最快的。

And then what we're finding is consumers are adapting the fastest, just individuals in their lives.

随后小型企业迅速跟进，因为小企业主通常能自主决策，非常容易尝试新事物。

The small businesses are then adopting right after that, because, you know, a small business typically is just, you know, a person who's, you know, making decisions for their own business, very, very easy to to do new things.

大公司随后才跟上小公司的步伐。

Big companies are then following small companies.

所以你看，名单上的那些公司，有些确实在做些有趣的事情，但总体而言，大公司现在内部相当纠结，被各种流程、遗留系统、组织结构、培训，还有工会之类的问题所困扰，相比个人和小企业，它们的采用速度其实相对较慢。

And so, you know, the companies on that list, some of, you know, obviously some of them are doing interesting things, but in general, big companies right now are pretty tied up in knots internally, kind of in all their processes, and in all their legacy systems, and all their, you know, organisation and training, and but their unions and like all the other issues they have to deal with, you know, they're actually relatively slow to adopt compared to individuals and small businesses.

而政府则是后知后觉的采纳者，对吧？

And then government is the late adopter, right?

政府当然已经在尝试摸索如何适应这项技术，但由于各种规章制度和官僚体系，它们的采纳速度并不快。

And so government, of course, are already trying to figure out, you know, kind of how to adapt this technology, but they're not adopting it very fast because they can't because of all their rules and systems and bureaucracy.

因此，技术在社会中传播的方式发生了真正的逆转，人工智能正成为这一现象的典型案例。

And so there's been a real inversion of how technology moves through our society that's really become AI is becoming a case study for it.

所以回答你的问题，我认为大公司的CEO们——其中许多人已经在这么做——但我觉得他们不必在公司内部强行推动这件事，因为这些大公司如今已是规则繁多的庞大官僚体系，默认情况下就会扼杀新想法。

And so the answer to your question is, I think big company CEOs, and, you know, many of them are doing this, but I think they don't really have to force the issue on this inside their companies because the, you know, these big companies are just such now giant bureaucracies with so many rules that, like, by default, they'll smother new ideas.

因此，能登上那样的名单并自豪地宣称自己处于领先地位，确实需要真正的领导力。

And so it's real act of leadership to get on a list like that and be able to actually say kind of with pride that we're on the leading edge.

如果你的朋友开了家面包店，他说：‘马克，我想请你来帮我研究怎么用人工智能。’那么，一个单店经营的面包店主现在能用它做什么呢？

So if your friend owned a bakery, and he said, Mark, I want you to come in and help me figure out how to use AI, Well, how could a someone who owns a a single storefront bakery use it now?

是的。

Yeah.

我是说，有几十种方法可以用。

I mean, so, I mean, you there's, I mean, there's, you know, there's dozens of ways.

显然这取决于你的业务目标是什么。

You know, it obviously depends on what your business goals are.

他们...他们...他们是面包师，有几十种方法。

They're they're they're baker dozens of ways.

他们并不完全...

They don't exactly.

所以，首先你可以做的就是进行绩效评估。

And so, yeah, mean, look, the first thing you can do is just say, look, you know, review you know, do a performance review for me.

比如，输入你的员工排班表。

Like, just feed in, you know, here's my here's my here's my staffing schedule.

你觉得这个安排怎么样？

You know, what what do you think of it?

给我一个批评意见。

Give me a critique of it.

你看，这是我们最近收到的100封客户邮件。

You know, here's the last, you know, 100 emails we've got customers.

这些邮件中有什么模式可循？

What are the what are the what are the patterns of that?

这是我们准备在当地报纸刊登或在Facebook等平台发布的广告文案。

You know, here's the copy for the ad that we're gonna place the local newspaper, you know, put up on Facebook or whatever.

你觉得这个怎么样？

Like, you know, what what, know, what do you think of this?

让它来进行绩效评估吧。

You know, let it let it do it let it do a performance assessment.

很多人发现这对个人辅导非常有效。

A lot of people find it very effective for personal coaching.

所以，老板可以这样使用它，或者让员工也这样使用。

And so, you know, the owner might use it that way or might, you know, ask the employees to use it that way.

我认为真正发挥威力的地方在于，比如你是个小企业主，现在有一家面包店，想开第二家分店对吧？

And then I think where the power really kicks in is, you know, say you're a you're a, you know, you're a small business owner, you got one bakery, now you wanna have two bakeries, right?

你想打造品牌，如果进展顺利，可能会扩展到五家、五十家、五百家，然后让包装产品进入超市等等。

And you wanna have a brand, and then maybe if that works, you're gonna have five, and then 50, and then 500, and then you're gonna have, know, packaged powders going to supermarkets and so forth.

这时候你基本上就把AI变成了思维伙伴，对吧？

And then there, you basically turn the AI into a thought partner, right?

你可以直接问：从单店扩展到连锁经营的最佳方式是什么？如何把生意做大？

And you basically say, okay, what, you know, what are the best ways to expand from a single, outlet to multiple and to turn this into a larger business?

由于AI训练时吸收了人类知识的很大部分，它内部包含了雷·克洛克如何把麦当劳从单店发展起来的所有经验，以及其他企业家成功的案例。

And the AI, because it's been trained on some large percentage of the total amount of human knowledge, it has within it all the information on how Ray Kroc turned McDonald's from know, a single restaurant in McDonald's and how all these other, know, entrepreneurs before actually did this.

所以它能与你探讨，帮助你为自己的企业制定发展方案。

And so they can, you know, explain with you and help you, you know, figure out how to do this for your own business.

最终你会意识到，这简直就像...哇。

And then, you know, the thing that you get into is you use it as basically, it's just like, wow.

就像拥有了全世界最好的教练、导师、心理医生、顾问，以及董事会成员。

It's like having the world's best coach, mentor, therapist, right, adviser, you know, board member.

但它就像拥有无限的耐心。

But it's like infinitely patient.

所以它很乐意进行这样的对话。

So it's like, it's happy to have the conversation.

它很乐意把同样的对话重复50次。

It's happy to have the conversation 50 times.

如果你承认自己的不安，它会耐心引导你克服。

It's happy if you admit your insecurities, it'll coach you through them.

即使你提出毫无根据的疯狂猜想，它也乐在其中。

You know, it's happy if you run wild speculations that don't make any sense.

它甚至愿意在凌晨四点陪你做这些事。

It's happy to do all that at four in the morning.

因此经常使用它的人发现，它实际上能提供非常有力的支持

And so the people who are using it a lot are finding it's it actually turns out to be very supportive in Could their

你说：这是我们最畅销肉桂卷的配方。

you say, here's the recipe for our best selling cinnamon rolls.

我怎样才能做得更好？

How could I make it better?

那会

Would that

是的，百分之百。

Yes, 100%.

没错，你可以说顺便提一下，AI艺术的一部分，对吧，就在于弄清楚该问它什么问题，对吧？

Yeah, you can say and by the way, part of the art of AI, right, is figuring out what questions to ask it, right?

因为，你知道，事实证明它能回答许许多多不同的问题。

Because, you know, it turns out it can answer many, many different questions.

实际上要在这方面创造它。

To actually create it at this.

对。

Yeah.

所以你可以说，这是我现在的食谱。

And so you could say, here's my current recipe.

你知道，我该如何改进它呢？

You know, how am I gonna improve it?

你也可以问，世界上最好的肉桂卷食谱是什么？

You can also say, you know, what's the best cinnamon roll recipe in the world?

从那个词倒着说。

Word backwards from that.

然后你还可以说，看，我想做出世界上最好的，但成本要降到十分之一。

And then you could also say, look, I wanna make the best one in the world, but I need to do it at a tenth of price.

你知道有哪些成本优化的方法吗？

You know, are the ways to cost optimize?

对吧。

Right.

好的。

All right.

所以另一件事可以问它，继续吧。

So the other thing can ask it Go ahead.

顺便说一下，你还可以问它：'我应该问什么问题？'

By the way, the other thing you can ask it is you can ask it, what question should I be asking?

对吧？

Right?

所以你可以输入：'我经营一家面包店..'

And so you could plug in, I run a bakery..

。

Dot.

'我应该问什么问题？'

What question should I be asking?

你会发现它实际上能帮你思考该问哪些问题。

And you'll find it's actually a thought in helping you figure out what questions to ask.

没错，太棒了。

Right, brilliant.

前几天我说过，我想知道每个投票反对比尔·克林顿任何弹劾条款的共和党人。

So the other day I said, I wanna know every Republican who voted against any of the articles of impeachment against Bill Clinton.

给我列出所有共和党人的名单。

Give me the list of every Republican.

结果返回的名单里混入了一些民主党人。

And it came back and it listed some Democrats mixed in there.

系统还标注了这些人是民主党。

And it said they were Democrats.

我回复说我只想要共和党人。

And I said back, I only want Republicans.

系统说：哦抱歉，我不小心混入了民主党人。

And it said, oh, sorry, I inadvertently included Democrats.

我能理解一个初级研究员会犯这种错误，但AI怎么会犯错，在被指出后还说：是啊，模型里怎么会出现这种情况？

Now I could understand a human being, a junior researcher doing that, but how could AI make a mistake, have it pointed out to it and say, oh yeah, how could that be in the model?

是的，这个问题会涉及到技术细节，我很乐意深入探讨技术细节。

Yeah, so, you know, this gets technical and I'd be delighted to get deeply into the technical details.

我会尽量克制。

I'll try to resist.

这是一种新型计算机。

It's a new kind of computer.

理解它的方式可以这样想：迄今为止的计算机，可以说都是超级字面化的。

And the way to think about it is computers up until now have been what you might call, like, super literal.

对吧？

Right?

以往的计算机运算速度极快，但每次执行的任务都完全相同。

Where computers up until now, like, do math really fast, but, like, they do the same thing every single time.

它们完全不展现任何创造性。

They exhibit no creativity whatsoever.

如果你期待它们表现出创造性，它们就是做不到。

And if you expect them to exhibit creativity, like, you know, they they just can't do it.

如果它们犯错，那也是因为人类程序员犯了错。

And then if they make a mistake, it's because the human programmer made a mistake.

这使得计算机在运行大型数学运算或处理许多计算机擅长的事务时极其有用。

And that has made computers, you know, super useful for running, like, you know, large math exercises or, you know, doing a lot of things that that computers do.

但当然，计算机从来不具备创造性。

But of course, computers have never been creative.

对吧？

Right?

计算机永远无法为你写诗，也无法和你一起研究肉桂面包的配方。

You can't computers never been able to write you poetry or work with you on your cinnamon bun recipe.

这甚至从来不是我们能够想象的事情。

Like, it's just never even been a thing that we can think about.

所以它从未具备那种人类特有的创造性元素。

So it's it's never had kind of that human element of creativity to it.

这是一种完全不同的计算机，具有说实话更像人类的特性——它大多数时候是正确的。

This is just a completely different kind of computer that has these characteristics that are frankly more like a person, which is it's right most of the time.

偶尔会出错。

It occasionally gets things wrong.

当它出错时，它能够进行自我批评。

When it gets things wrong, it's it's able to self critique.

你需要像与人合作那样与它协作。

And you have to kind of work with it the way that you work with a person.

因此在使用过程中你需要不断摸索。

And and so you have to basically figure out as you use it.

比如，你要善用它的创造力，同时也要包容它偶尔会出错——就像对待人类同事一样。

Like, you wanna take advantage of the fact that it's creative, and then you want to be tolerant of the fact that it's not always correct, just like you're working with a person.

不过话说回来，当它犯一些完全可以避免的低级事实错误时，我们称之为'幻觉'。

Now, having said that, when it makes easily avoidable mistakes, where it's just like makes boneheaded fact mistakes, you know, we call those hallucinations.

最新系统在这方面已经有了极大改善。

The latest systems are much, much, much better at not doing that.

它们的准确性大幅提高。

They're much more accurate.

特别是对于正在观看的观众，如果想看实际案例，我可以举个例子。

And in particular, for anybody watching this, if you want to kind of see this in action, I'll just give an example.

如果你购买ChatGPT完整版，会有一个叫GPT五Pro的最新模型，还有个叫'深度研究'的功能开关。

If you buy the full version of ChatGPT, there there's a a model called GPT Pro GPT five Pro, which is the latest one, and then there's something called Deep Research, which is a switch that you turn on.

如果你使用GPT五专业版并开启深度研究功能，提出类似问题时，我认为虽然还不能说它万无一失，但它在保持快速和基于事实上确实表现得非常出色。

And if you use GPT five Pro with Deep Research, you ask a question like that, like, at this point, I think it's I I wouldn't say it's bulletproof, but, like, it's really good at at at staying fast and grounded.

实际上，你可以观察它的工作过程，它会真正地在互联网上搜索。

And literally, you can watch it work, and it'll literally go out on the Internet.

它会检查所有权威来源，比如访问congress.gov等网站核对投票记录并进行验证。

It'll, like, check all the authoritative authoritative sources, and it'll, you know, it'll go on congress.gov or whatever and check the voting records and and verify that.

所以我认为这个问题正在被解决，可以说就在我们说话的当下。

And so I I think that problem is being ironed out, you know, kind of as we speak.

是的。

Yeah.

对于观看或收听但尚未使用或使用不多的观众来说，你提到的关于优质提示的重要性非常关键。

For people watching, listening who haven't used it or not used it much, the thing you said about good prompts is so key.

这确实区分了能高效利用它的人和不能的人。

And that does differentiate the people really getting productive at it from not.

我们之前讨论过的一个有趣现象是，这确实很滑稽。

And one of the things that you and I have talked about is it is hilarious.

如果你给出恰当的提示，它能写出非常有趣的内容。

If you give it the right prompts, it can write stuff that is so funny.

这种能力从何而来？

And where does that come from?

它如何具备理解幽默的能力？

How does it have the capacity to understand?

因为幽默涉及隐喻、复杂性和人性化表达。

Because humor involves metaphors and sophistication and humanity.

它是怎么做到的？

How can it do that?

是的，这涉及到它的训练原理。

Yeah, so this to this idea of how it's trained.

本质上这些系统是人类知识随时间积累的产物——顺便说，大部分训练数据其实就是互联网本身，对吧？

So basically, what these systems are is they're basically the accumulation of human knowledge over time, sort of and by the way, most of the training data is just literally the internet, right?

所以这件事现在能实现而二十年前不行，是因为互联网终于发展到足够庞大——不仅仅是万维网——足以容纳所有这些信息了，明白吗？

So one of the reasons that this is happening now and not twenty years ago is because the internet finally got big enough, not just the web, finally got big enough to have like all this information in it, right?

所以如果你今天上网，你能找到各种各样的在线文字材料，比如四百年前电影黄金时代的经典剧本。

And so if you go to the internet today, you know, you can find, you know, all kinds of written material online that's, you know, it's, you can find like, you know, classic screenplays from, you know, the old age of cinema 400 ago.

你知道，你能看到人们整天在社交媒体上互相开玩笑。

You know, you can find, you know, literally, you know, people joking with each other on social media all day long.

你能找到专业喜剧演员讲述他们如何创作出优秀喜剧的口述历史。

You know, you can find, you know, professional comedians doing oral histories of, you know, how how they did great comedy.

网上有大量关于喜剧和幽默的惊人信息。

So there's like, there's incredible amounts of information that are online about comedy and about what's funny.

所有这些信息都包含在训练数据中。

And then all of that information is in the training data.

这些数据在训练过程中都被输入到AI里。

So it's all kind fed into the AI during the training process.

AI基本上像处理其他数据一样处理这些信息，最终成为幽默领域的顶尖专家。

And the AI basically processes it through like any other kind of data and comes out the other end and just basically is like, oh, now that AI's a world class expert in humor.

对吧？

Right?

当然，听着，你可能是个幽默理论专家但实际上并不幽默。

And of course, look, you could be an expert in humor and not actually be funny.

你知道，大学里可能有喜剧教授之类的人，但他们并不怎么好笑。

And they're, know, they're, I don't know, probably professors of comedy or something like that in colleges who aren't very funny.

但它确实对幽默的本质了解得非常透彻。

But like, just it knows so much about what humor is.

它对笑话的模式了如指掌。

It knows so much about the pattern of jokes.

它非常清楚什么能让人发笑。

It knows so much about what make people laugh.

它有海量的案例可供学习借鉴。

It has so many examples, you know, to be able to learn from.

专业喜剧演员会告诉你喜剧是有规律可循的，对吧？

And, you know, the professional comedians will tell you there are patterns to comedy, right?

我曾与一位专业喜剧演员合作过，他说关键在于细节的精准把握——必须准确击中笑点。

You know, there are, you know, I worked in a professional comedian once with something, and he said, you know, the key to it is specificity, like, need to get you need to just, nail the reference.

知道吗，另一个喜剧演员会说，关键在于时机、节奏，或者对前面笑点的呼应，总之就是那些笑料包袱之类的。

Know, another comedian will say, the key is, you know, timing or pacing or the callback to the previous jump or whatever, you know, punch line, whatever it is.

所以它完全掌握了这些技巧，而且现在它太强大了，确实如此。

And so it just it it it knows all that, and then it's just, you know, it's now so powerful that it's actually in yeah.

是啊。

I yeah.

老实说，特别是在凌晨两点，我觉得这些东西简直笑死人了。

Quite honestly, I find especially at, two in the morning, I find these things hysterical.

嗯。

Yeah.

好的，我们还有一分钟就要休息了。

All right, with just a minute left before we take a break.

对于几乎没用过或完全没用过它的人，你有什么建议？

What's your advice to an individual who's barely used it or hasn't used it?

你对他们有什么入门建议？

What's your advice to them to sort of how to get started?

是的。

Yeah.

所以，目前为止最好的方法就是直接下载并使用它。

So, I mean, by far the best way to do it is just download it and use it.

就像我说的，现在有好几种选择，埃隆的Brock就很棒。

And like I said, you know, there's there's several, you know, Elon's got Brock, you know, which is which is now fantastic.

顺便说一句，这些技术被集成到产品中的方式其实很有趣。

By the way, you know, it's actually interesting how these things are getting getting built into products now.

比如新版X（原推特）的每个帖子右上角都有个小Grok图标，看起来像个小黑洞。

So what the, you know, the the new versions of X, know, formerly Twitter, actually, if if you go to any post on X, there's a little Grok icon that looks like a little a little black hole icon in the upper right corner of the post.

点击它就能调用Grok人工智能为你解释这个帖子。

And if you click on it, it actually calls up the Grok AI to explain the post to you.

对吧？

Right?

它真的可以为你解释内容，比如当你不理解某个政治帖文或跟丢话题时。

And so like, it literally is like, it can explain to you that, you know, if there's some, you know, post on politics or something, you don't understand what's happening, you just, you know, lost the thread of the topic.

你只需轻轻一点那个按钮，就能与AI展开对话。

You just like, bonk that button, and you get in a dialogue with the AI.

一个小型AI窗口会弹出解释帖子内容，你还可以向它询问更多细节。

A little AI window pops up and explains the post, you can ask it for more details.

所以它就像是内置在这个产品里的功能。

And so it's like built into that product.

谷歌现在实际上已经把AI整合进了搜索引擎。

Google's actually built AI now into search.

现在当你进行搜索时，会有个叫AI模式的选项，点击它。

And so now when you do searches, it has this thing called AI mode, you bonk on it.

除了谷歌提供的10个蓝色链接外，你现在还能进入AI对话界面。

And, you know, in addition to getting the 10 blue links from Google, you now get into an AI dialogue.

所以，你只需开始使用这些产品，或者直接下载其中一个应用开始体验。

And so, you just start using those products, or you just download one of these apps and start using it.

就像我说的，一个特别好的问题是：好吧，我该怎么使用你？

And like I said, a really great question is like, okay, how do I use you?

对吧？

Right?

比如，或者说'教我'这个指令效果特别好。

Like, or you can say, teach me works really well.

你可以说，'教我如何以最佳方式使用你'。

You can say, you know, teach me how to use you in the best way.

你知道的，比如'教我如何为我的生意使用你'，或者'为这个项目教我如何使用你'。

You know, teach, you know, teach you know, teach me how to use you for my for my business, you know, for this project.

这些AI特别喜欢交流，它会很乐意坐在那里滔滔不绝地带你一步步了解。

And it'll, you know, these things love to talk, and it'll it'll happily sit there and chatter away and and take you through it.

是的。

Yep.

我的理解和你刚才说的类似，就是尽可能具体地向它提出你的需求。

And just I my my sort of analog to the way you just said it is ask it for what you want as specifically as possible.

不要有所保留。

Don't hold back.

要非常具体明确，它就会按照你的要求去做。

Be really specific, and it will do what you ask.

好的。

Alright.

接下来还有与马克·安德森的对谈。

More with Marc Andreessen.

马上为您呈现。

That's next up.

请继续关注。

Stay tuned.

如果您年满64岁或以上，这条公告非常重要。

If you're 64 years old or older, this is an important announcement.

司法部最近起诉了三家主要医疗保险经纪公司，指控他们自称公正无私，却涉嫌将客户推向能获得最高回扣的计划。

The Department of Justice recently sued three major Medicare brokers for claiming they were unbiased while allegedly pushing people into plans that got them the biggest kickbacks.

这是事实。

It's true.

太多保险代理人根本不可信，但你也无法指望政府能提供你所需的最佳信息。

So many insurance agents, they just can't be trusted, but you also cannot rely on the government to give you the best information that you need either.

这就是为什么我想让你了解一个叫Chapter的机构。

That's why I want you to know about something called Chapter.

Chapter的创始人们亲身经历了这一切，他们的父母曾被更关注佣金而非优质服务的代理人推入了错误的医保计划。

Chapter was started by people who went through all of this personally after their own parents were pushed into the wrong Medicare plan by an agent who was more focused on commissions than on good care.

Chapter的使命非常简单。

Chapter's mission is very simple.

他们希望给予每位美国人应得的诚实、直接的医保建议。

They want to give every American the honest, straightforward Medicare advice that they deserve.

以下是他们与众不同的地方。

And here's what makes them different from everybody else.

他们是唯一一家比较全国范围内所有医保计划（而非仅少数几种）的顾问机构。

They're the only Medicare advisor that compares every plan nationwide, not just a few.

这为他们的客户平均每年节省1100美元。

That saves their clients an average of $1,100 a year.

真的没有理由不打电话。

There's really no reason not to call.

过程快捷简单，他们能在20分钟内帮你评估所有选择。

It's quick, it's easy, and they can review your options in under twenty minutes.

如果你已经在合适的计划中，他们会如实告知。

If you're in the right plan already, they'll let you know that.

但如果有更好的方案，他们会协助你完成转换。

But if there's a better plan, they'll help you make the switch.

这可能是你今年最重要的一个电话。

This could be the most important call you make this year.

拨打250，说出'Chapter Medicare'，获得安心保障。

Dial 250 and say chapter Medicare to get peace of mind.

再次提醒，拨打250，说出'Chapter Medicare'。

Again, that's 250 and say chapter Medicare.

好的。

All right.

欢迎回来。

Welcome back.

我们还在与马克·安德森一起。

We're here with Marc Andreessen still.

马克，你是否认同那种认为美国正与中国进行一场关乎存亡的较量的观点？

Marc, are you of the school that says that we're in an exit The United States is in an existential struggle with China?

你赞同这种看法吗？

Do you subscribe to that point of view?

我正想说，我希望这不是真的。

I was gonna say, I hope that's not true.

对吧？

Right?

你知道，我希望事情不会发展到我们当年与苏联那样的局面。

You know, I I hope this is not gonna walk all the way to the situation that we ended up in with the Soviet Union.

明白吗？

You know?

就像，大多数时候，你知道，像你一样，我成长在一个时代，我确信，你会回忆起来，就像，你知道，我成长过程中一直认为有很大概率我们都会死于核战争。

Like, most like, you know, like you, I grew up in an era, I'm sure, you know, you you reminisce, like, you know, I I growing up thinking there was a significant chance that, like, we're all gonna die, you know, from from nuclear war.

所以我希望局势不会恶化到那种紧张程度。

And so I hope it doesn't get back to that level of intensity.

但我确实认为当前美中之间发生的许多事情与二十世纪美苏之间的情况存在历史相似性。

But but I I do think there there are a lot of historical parallels to what happened between The US and The USSR in the in the twentieth century, you know, happening right now.

而且，你知道，现在有两个霸权超级大国，它们都有各自的愿景。

And, you know, you have, you know, two hegemonic superpowers that both have, you know, visions.

它们对于社会应该如何构建，以及全球政治体系应该如何运作都有自己的设想。

You know, they both have visions of how society should be structured and how the the sort of global political systems should be structured.

当然，我认为美国的模式更好，但它们确实都有各自的理念。

And and, you know, obviously, I think, you know, America's is better, you know, but they both have visions.

而且，你知道，它们都在为实现这些目标而努力。

And, you know, they're they're they're both working, you know, to that end.

此外，它们在国际上各有军事、技术、经济和文化方面的优势与劣势。

And and then they they both have international strengths and weaknesses, militarily, technologically, economically, culturally.

而且，你知道，这种地缘政治斗争正在上演。

And, you know, there there is there is that kind of geopolitical fight happening.

所以我希望我们能保持这种，我不知道该怎么称呼它，合作竞争的模式，你知道，紧张但不至于爆发军事冲突。

So so I I hope we stay in this kind of, I don't know what you call it, mode of, like, coopetition, you know, tension without, you know, without military strife.

你知道，我希望我们能保持那种状态。

You know, I I hope we stay, you know, in in that mode.

但是，这确实是一个足够棘手的局面，我们显然需要。

But, like, it's it's a sufficiently fraught situation, that we certainly need.

顺便说一句，我们需要制定一个如何赢得这场竞争的国家战略，并且确保我们能做到。

And by the way, have a national strategy for how to win that, and we need to make sure that we do.

没错。

Right.

在人工智能方面，他们是否有一两项优势超过我们？

Do they have any one or two advantages over us in terms of AI?

是的，他们有。

Yeah, they do.

正如我刚才所说，他们确实拥有两大优势。

So they have all I was say they have two advantages.

顺便说一句，我必须说明，我们也有很多优势。

And by the way, I should say, we have many advantages.

我对美国非常乐观，我认为我们的处境更为有利。

I'm very bullish on The US, and I and I, you know, I think we're we're better positioned.

坦白说，我宁愿保持现状也不愿与他们交换位置，这点我们可以详细讨论。

I I like I would trade places with them, and and we could talk about that.

尽管如此，他们确实具备某些优势，尤其是两个关键强项。

But having said that, they do they do have strengths, and in particular, they they have two two key strengths.

第一是他们确实拥有计划经济体制的优势。

Number one is they do have the advantages of of a command economy.

总体而言，这种模式能达到百分百的效率。

And, you know, like, generally speaking, are 100%.

我个人更倾向于自由市场、去中心化的动态经济体系——在这方面我们更具优势，我们拥有更完善的创业生态系统等等。

Like, I'm on the side of, you know, free markets and and, you know, decentralization and, you know, having a dynamic economy, and we have advantage, know, we have much, you know, we have much better entrepreneurial ecosystem and so forth.