BuildSpeak每日 builder 文摘
今日归档生词本关于
🎙 播客AI & I by Every· 2026 年 5 月 27 日· 6,946 词 · 约 35 分钟

We Automated Everything With AI and Tripled Our Headcount

SPACE 播放 / 暂停·←→ 上一句 / 下一句
Speaker 100:00 - 00:06
You prompt AI to do something. It blows your mind. You feel inadequate. You feel like, oh my god. This thing's gonna take my job.
Speaker 100:00 - 00:06
你让 AI 去做一件事。它让你震撼不已。你会觉得自己不够好。你会想,天啊,这东西要抢走我的工作了。
Speaker 100:07 - 00:11
And then it stops working, it looks back at you and says, what should I do next?
Speaker 100:07 - 00:11
然后它又突然不灵了,回过头来看着你说:我下一步该做什么?
Speaker 200:11 - 00:47
The further away an agent gets from a human, the less valuable it is. If you just ride the models, you're going to be fine. If you care about leading a really ambitious life, I truly think that this is going to make that more possible for more people. Every is the only subscription you need to stay at the edge of AI. If you care about being on top of the latest models and using the latest tools, you have to subscribe to Every to separate out the signal from the noise.
Speaker 200:11 - 00:47
agent 离人类越远,价值就越低。如果你只是顺着这些 model(模型)去用,你会没问题的。但如果你在乎的是去过一种真正雄心勃勃的人生,我真的认为,这会让更多人更有可能做到这一点。Every 是你想站在 AI 前沿时唯一需要订阅的东西。如果你在乎跟上最新的 model(模型)并使用最新的工具,你就必须订阅 Every,才能把真正有价值的信号从噪音里分离出来。
Speaker 100:47 - 00:55
Go to every. Tosubscribe today. So we are here because we're gonna flip the script a little bit. I am gonna be interviewing Dan
Speaker 100:47 - 00:55
现在就去 every 订阅。所以我们今天在这里,是因为我们要稍微把剧本反过来一点。我要来采访 Dan。
Speaker 200:55 - 00:55
Sick.
Speaker 200:55 - 00:55
爽。
Speaker 100:56 - 01:08
About the piece that he published yesterday, May 21, and we're gonna try to understand why he wrote it, what's underneath his reasoning for it. There's gonna be some conflict. I'm gonna fight with him on it.
Speaker 100:56 - 01:08
围绕他昨天,也就是 5 月 21 日发表的那篇文章,我们会试着理解他为什么写它,以及支撑他这样推理的底层原因。这里面会有一些冲突。我会就这件事跟他辩一辩。
Speaker 201:08 - 01:09
Let's go. Let's fight.
Speaker 201:08 - 01:09
来吧。开战吧。
Speaker 101:09 - 01:21
And see, you know, bring in some of my opinions, which are more or less aligned, but trying to understand, is this piece gonna reflect the future in ten years and five years?
Speaker 101:09 - 01:21
然后看看,带进来一些我的看法——这些看法大体上是一致的——但我们真正想弄清的是,这篇文章所说的,会不会就是五年后、十年后的未来?
Speaker 201:21 - 01:22
And who are you again?
Speaker 201:21 - 01:22
你刚才说你是谁来着?
Speaker 101:23 - 01:29
I'm Brandon. I'm our COO, and that's it.
Speaker 101:23 - 01:29
我是 Brandon。我是我们的 COO,仅此而已。
Speaker 201:29 - 02:25
So the piece is called After Automation, and it comes from this feeling that I have, and there's a video about this and there's there's a piece, but just for people who who have not seen either of those things, it comes from this feeling that I have that at every we are as AI native, as agent native as it as as it gets, You know, if you swing a stick around in our Slack, you're as likely to hit a human as you are an agent. Everyone's using Cloud Code and Codex and all these tools to do their job every day. And yet it feels like there's more human work to do than ever. And in fact, since the GBT three days, we've grown from four people to 30 people, and we're hiring more now. And so it came from me looking at that and then looking at the environment and being like, what's going on?
Speaker 201:29 - 02:25
这篇文章叫做《After Automation》,它来自我的一种感受。我还做了一个相关视频,也写了一篇文章;不过对于那些两者都还没看过的人,我先简单说一下:这种感受来自于这样一个事实——我们已经是尽可能 AI native、尽可能 agent native 的公司了。你知道的,在我们的 Slack 里随手挥根棍子,砸中的大概率不是人类就是 agent。每个人每天都在用 Cloud Code、Codex 以及各种这类工具来完成工作。可即便如此,感觉上需要人做的工作却比以往任何时候都更多。事实上,从 GBT three 那时候以来,我们已经从 4 个人增长到了 30 个人,而且现在还在继续招人。所以我是看着这一点,再看看外部环境,就会想:这到底是怎么回事?
Speaker 202:25 - 03:14
Because the whole information environment, if you look at Dario, is out there saying like half of entry level white collar jobs may be wiped out. Even people like Ken Griffin from Citadel, you can tell he just had this moment where someone showed him an AI doing an advanced data or finance question. And he was like, holy shit, like, that's what I would pay PhDs to do for me. And it just did it. And I feel like I'm watching a lot of people who maybe don't have a ton of experience with agents and don't have a ton of experience with the curve of improvement that we've been writing for the last like three, three and a half years, hit it for the first time and then come to all these conclusions about, oh my god, all work is going away, we're not going to have jobs.
Speaker 202:25 - 03:14
因为整个信息环境里,如果你去看 Dario 的说法,他在外面讲的是:一半的初级白领岗位都可能会被抹掉。甚至像 Citadel 的 Ken Griffin 这样的人,你都能看出来他明显刚经历过那种时刻——有人给他展示了一个 AI 在处理高级数据或金融问题。然后他就会想,holy shit,这种事本来是我要花钱请 PhD 替我做的。结果它居然直接做出来了。而我的感觉是,我正在看着很多可能并没有太多 agent 使用经验、也没有太多我们过去大概三年到三年半一直在记录的那种改进曲线经验的人,第一次真正撞上这件事,然后就得出各种结论:天哪,所有工作都要消失了,我们以后不会有工作了。
Speaker 203:14 - 04:12
And I'm just sitting here being like, actually, your intuitions when you first see a technology like this are usually very off. And we've seen a lot over and over again over the years that EVRI is a very good bellwether for where things are going because it's a group of early adopters, we have people doing all sorts of work internally at EVRI, and if something works here, there's a good bet that it's gonna spread to other businesses that are adjacent to ours. And so when I look around at Every, I see so much automation, and I also see way more human work. So I was really The whole piece is saying, here's the current state of work with agents, and then pulling apart that paradox and explaining, does more automation mean more work?
Speaker 203:14 - 04:12
而我坐在这里的想法是:其实,当你第一次见到这种技术时,你的直觉通常会偏差得很厉害。并且这些年来我们一再看到,EVRI 是一个非常好的风向标,因为它是一群早期采用者的集合;在 EVRI 内部,我们有人在做各种各样的工作;如果某样东西在这里行得通,那么它大概率也会扩散到与我们相邻的其他企业里去。所以当我环顾 Every 时,我看到的是大量自动化,同时我也看到多得多的人类工作。因此,这整篇文章真正想表达的是:这就是 agent 参与下当前工作的状态,然后把这个悖论拆开来解释——更多自动化,是否意味着更多工作?
Speaker 104:13 - 04:54
Yeah, when I read the piece, there wasn't an explicit call to action in it, but I felt this call to action of there is actually a massive amount of hope right now in a world that is filled with a lot of doomers, and this is why. But I am going to come out of the gate and ask you a devil's advocate question, which is a couple hours before you publish this piece. The CEO of ClickUp came out with this long tweet about why he fired 8,000 people and 3,000 people.
Speaker 104:13 - 04:54
对,我读那篇文章的时候,里面并没有一个明确的 call to action,但我感受到一种 call to action:在一个充满 doomers、弥漫着很多悲观论调的世界里,其实当下存在着巨大的希望,而这就是原因。不过我还是要一上来就问你一个 devil’s advocate 式的问题:就在你发表这篇文章前几个小时,ClickUp 的 CEO 发了一条很长的 tweet,讲他为什么裁掉了 8,000 人和 3,000 人。
Speaker 204:54 - 04:56
I don't think it was 8,000.
Speaker 204:54 - 04:56
我觉得不是 8,000。
Speaker 104:56 - 05:04
It was 20,000. I think it was like 3,000 It like 22% of his workforce.
Speaker 104:56 - 05:04
是 20,000。我觉得大概是 3,000 人,差不多占他员工总数的 22%。
Speaker 205:04 - 05:08
I don't think it was in the thousands, but yes, was a lot of his workforce.
Speaker 205:04 - 05:08
我不觉得有上千人,但没错,确实占了他劳动力里的很大一部分。
Speaker 105:08 - 05:50
So my question to you is: in a business like every we're growing super fast. What you wrote makes a lot of sense to me. And what you wrote theoretically makes a ton of sense, in that AI is not autonomous right now it has to be told what to do and that has to be checked, we need to have that that sandwich that you described in the piece. But in a business that is 8,000 people 10,000 people that is mature, and has built ways of managing SOPs for managing their business. Does this manifesto and this thesis still hold true?
Speaker 105:08 - 05:50
所以我想问你的是:在我们这种增长超级快的业务里,你写的内容我觉得很有道理。而且从理论上说,你写的这些也非常说得通——因为 AI 现在还不是 autonomous(自主的),它必须被告知该做什么,而且还必须经过检查;我们需要你在那篇文章里描述的那种“夹心结构”。但如果是在一家有 8,000 人、10,000 人的成熟企业里,它已经建立起了管理 SOP(标准作业流程)和经营业务的方法,那么这份 manifesto(宣言)和这套 thesis(论点)仍然成立吗?
Speaker 205:50 - 06:00
It's a really good question. There are a couple different questions here. The first thing I want to do is lay out the argument. Why automation make more work?
Speaker 205:50 - 06:00
这个问题非常好。这里其实包含几个不同的问题。我首先想做的是把这个论证讲清楚:为什么 automation(自动化)会带来更多工作?
Speaker 106:00 - 06:06
I'm sure many people listening to this also haven't read it, so take a second to explain that in detail.
Speaker 106:00 - 06:06
我敢肯定,正在收听这个的人里也有很多还没读过那篇文章,所以请你花一点时间详细解释一下。
Speaker 206:06 - 06:53
I will do that. Basically, the idea is the way that AI works and the way it functions in the workplace is AI makes yesterday's expert competence cheap. And by that, I mean AI is trained on all of our outputs, all of the code and the writing and the design and decision making and everything that's ever been written. And it makes that available to everyone for very cheap. Anyone now with a prompt can use yesterday's competence to solve a programming problem, build an app, or write a piece like I did, write a report, or make a YouTube thumbnail.
Speaker 206:06 - 06:53
我会的。基本上,这个观点是这样的:AI 的工作方式,以及它在 workplace(职场)中的运作方式,是让“昨天的专家级能力”变得廉价。我的意思是,AI 是用我们所有人的产出训练出来的——所有的 code(代码)、writing(写作)、design(设计)、decision making(决策),以及曾经被写下的一切内容。然后它把这些能力以非常低的成本提供给所有人。现在,任何人只要写一个 prompt(提示词),就可以利用“昨天的能力”来解决编程问题、构建一个 app(应用)、或者像我刚才那样写一篇文章、写一份报告,或者做一个 YouTube 缩略图。
Speaker 206:54 - 07:08
And the interesting thing is that when you do that, expert competence is available for cheap, it gets really widely adopted. So everyone starts to do it. Everyone starts to we see this internally. Everyone's making poor
Speaker 206:54 - 07:08
有意思的是,当你这样做之后,专家级能力一旦可以被低成本获取,就会被非常广泛地采用。于是每个人都开始这么做。每个人都开始——我们在内部也看到了这一点。每个人都在做很差的
Speaker 107:08 - 07:08
requests. There's a lot of
Speaker 107:08 - 07:08
请求。数量很多,
Speaker 207:08 - 07:09
holy shit.
Speaker 207:08 - 07:09
我靠。
Speaker 107:09 - 07:10
This is crazy.
Speaker 107:09 - 07:10
这太疯狂了。
Speaker 207:10 - 07:28
Yeah. And I'm making poor requests, ops people are making poor requests. And engineers are writing essays, and there's all this line crossing basically for non experts to do the thing that experts used to do. And that feels very threatening to experts. They're like, well, what's my job going to be now?
Speaker 207:10 - 07:28
对。而且我在提很糟糕的 request,ops 人员也在提很糟糕的 request。工程师们却在写长篇大论,基本上到处都在越线,让非专家去做原本该由专家来做的事。而这会让专家觉得很受威胁。他们会想,那我现在的工作还会是什么?
Speaker 207:28 - 08:08
And what's interesting about that is because these tools are trained on outputs, are trained on yesterday's data, the stuff that they do with a default prompt all looks kind of similar and is all kind of right for the current situation, but it's not actually totally right. And so what happens is you flood the zone with tons of stuff that's close but not quite right, and then you need to basically
Speaker 207:28 - 08:08
这里有意思的一点在于,这些工具是基于 output(输出)训练的,是基于昨天的数据训练的,所以它们在 default prompt(默认提示词)下做出来的东西,看起来都挺像的,而且对当前情境来说也都大致算对,但其实并不完全对。所以结果就是,大量“差不多对、但又不完全对”的内容一下子涌进来,然后你基本上就得
Speaker 108:09 - 08:19
There's a lot of that at every, too. There's lot of people doing what seems like great work, and then you go under the hood, and you're like, this isn't quite right, maybe the expert should do
Speaker 108:09 - 08:19
这种情况在各个层面也很多。很多人做出来的工作看上去都很不错,但等你深入去看,就会发现,这里还是不太对,也许还是该让专家来做
Speaker 208:19 - 08:22
it. Yeah, exactly. Me, for example.
Speaker 208:19 - 08:22
它。对,完全没错。比如我自己就是这样。
Speaker 108:26 - 08:37
I've never witnessed that. All of us. All of How many PRs have I pushed? Coming from personal experience. I pushed so many PRs where I'm like, Willie, I literally have no idea if this works, but here you go, and he's
Speaker 108:26 - 08:37
我可从没见过这种事。我们所有人都这样。我到底提过多少个 PR?这是我的亲身经历。我提过太多 PR 了,每次我都在想,Willie,我真的完全不知道这东西能不能跑,反正先给你,然后他就
Speaker 208:37 - 08:38
like like, shut the
Speaker 208:37 - 08:38
像是在说,闭
Speaker 108:38 - 08:42
fuck up. Well, he's like he's like, this is a good idea, but I just completely redid
Speaker 108:38 - 08:42
嘴吧。嗯,他其实是说,这是个好主意,但我已经把它完全重写了。
Speaker 208:42 - 08:43
it. Yeah. Yeah. Yeah. Exactly.
Speaker 208:42 - 08:43
对。是。是。是。完全没错。
Speaker 208:43 - 09:16
So that's the kind of thing I'm talking about. It's kind of right, it's close, but it's actually not quite right, and you need someone, you need an expert to actually figure it out. But what's interesting is when you flood the zone with all that kind of stuff, what used to be expensive because its expert competence is now cheap, now it looks the same, so everything gets devalued. You get this abundance of stuff that used to be very expensive and looks work, like code and essays and whatever, but it's all kind of similar and all not quite right for the situation, so its value gets a lot lower. It's immediately lower.
Speaker 208:43 - 09:16
所以我说的就是这种情况。它某种程度上是对的,也很接近了,但实际上又不完全对,而你需要有人、需要专家真正把它弄明白。但有意思的是,当你把这类东西大量铺开时,那些过去因为需要专家能力而很昂贵的产出,现在变便宜了,而且看起来都差不多,于是一切都被贬值了。你会得到大量过去非常昂贵、而且看起来像是成品的东西,比如 code、文章之类,但它们都很相似,而且对具体情境来说都不完全合适,所以它们的价值就会大幅下降。是立刻就下降。
Speaker 209:18 - 10:17
And then what happens is you actually get more demand for experts to come in and help take that stuff that is being produced by people, and you have good ideas, for example, but now there's a lot of demand for an expert to come in and help get that idea across the finish line, so that can that looks usually like experts are in demand for building systems to get the kind of, you you could say slap work that can now be produced by everyone and shepherd that into something that's actually useful. So an example would be we have repo rules and review guidelines and stuff like that, so that before you see a PR, before Willie sees a PR, hopefully it's gone through a bunch of processes to make sure it's actually reasonably good. We have the same thing on the editorial side, so building systems for that. And then there's also a lot of demand for experts to use these tools now that the floor is a lot higher, use these tools to make stuff that could never have been made before. And we do that all the time.
Speaker 209:18 - 10:17
接着实际发生的是,你反而会得到更多对专家的需求,让他们进来帮助处理那些由人们生产出来的东西。比如你有一些好点子,但现在非常需要专家进来,帮你把这个点子真正推进到终点线;所以这通常表现为,专家被需要去构建各种系统,把那种现在人人都能产出的、你可以说是随手拍式的工作,导入并管控成真正有用的东西。举个例子,我们有 repo 规则、review 指南之类的东西,这样在你看到一个 PR、在 Willie 看到一个 PR 之前,它最好已经经过了一系列流程,以确保它实际上是相当不错的。我们在 editorial 这一侧也有同样的做法,所以就是在构建这样的系统。然后,对专家还有另一类很大的需求:既然现在门槛底线高了很多,专家就可以使用这些工具去做出以前根本做不出来的东西。我们一直都在这么做。
Speaker 210:17 - 11:04
Like we have Kieran who just built an entire inbox end to end in like a month or two, and that's like complete that's totally completely impossible. And so there's so there's this really interesting thing that happens that even as you automate, the automation produces a glut of work that's all okay, that's all like reasonably good. That work is all very, very similar and not quite a fit for the actual situation, and that that increases the demand for experts who can like make it like make it actually good, make it actually different, make it actually appropriate for the for the the live situation as it is right now. And I think that's something that people don't quite understand, especially when they first encounter a language model and they or or an agent that can do something and they see it and they're like, holy shit. It's just it just does everything.
Speaker 210:17 - 11:04
比如我们有 Kieran,他就在一两个月里端到端做出了一个完整的 inbox,而那种事以前是完全、彻底不可能的。所以这里会发生一件很有意思的事:即便你在自动化,自动化带来的也是一大堆“还行”的工作,也就是相当不错的工作。但这些工作彼此都非常非常相似,而且和真实情境又不完全贴合;这反而提高了对专家的需求,因为专家能把它真正做好、真正做出差异、真正做得适合当前正在发生的实际场景。我觉得这一点很多人并没有完全理解,尤其是他们第一次接触 language model(语言模型),或者接触一个能做事的 agent(智能体)时,看到它就会想,天啊,它简直什么都会做。
Speaker 211:04 - 11:26
And the end and the the reality is it's incredibly good. It's amazing. It totally changes how we do work. And our experience so far at Everi is the further away an agent gets from a human, the less valuable it is. The human connection with an agent to actually do the work is the most important thing for making it work well.
Speaker 211:04 - 11:26
而最终的现实是,它确实好得惊人。非常了不起。它彻底改变了我们的工作方式。而我们目前在 Everi 的经验是,agent 离人越远,它的价值就越低。人和 agent 之间的连接,也就是人实际参与它完成工作的方式,是让它真正好用的最重要因素。
Speaker 111:28 - 12:23
Experts are more important than ever, because they lay the groundwork for an agent to do amazing work, and only then can you have the other humans actually take that agent and do work that levels them up. So there was a point where we were thinking about this piece, Dan was drafting this piece, where the title was The Tide is Rising, and that was trying to emote this idea that the tide is rising. We are all able to do more work, better work, but our eyes, whether you're an expert or not an expert in something, are always a little bit above where that waterline is. And I really liked the end of the piece, where you describe fuck. Achilles?
Speaker 111:28 - 12:23
专家比以往任何时候都更重要,因为是他们为 agent 做出惊人工作打下基础,只有在那之后,其他人类才能真正接过这个 agent,去完成那些让自己能力升级的工作。我们之前曾这样想这篇文章,当时 Dan 在起草这篇文章,标题一度叫作 The Tide is Rising,想传达的是这样一种感觉:潮水正在上涨。我们所有人都能做更多的工作、做得更好,但无论你是不是某个领域的专家,我们的视线总是会稍微高于那条水线。而且我很喜欢文章结尾的部分,就是你提到那个——呃。Achilles?
Speaker 112:23 - 12:48
Is it Achilles? Achilles. Achilles sprinting ahead of the tortoise, according to Zeno's paradox, that shouldn't happen. But in this world, it actually does. Prompt AI to do something, it blows your mind, it does that, you feel inadequate, you feel like, oh my god, this thing is gonna take my job, and then it stops working, it looks back at you and says, what should I do next?
Speaker 112:23 - 12:48
是 Achilles 吗?Achilles。按照 Zeno's paradox,Achilles 朝乌龟冲刺并超过它这件事本不该发生。但在这个世界里,它实际上就是会发生。你让 AI 去做一件事,它会让你震惊,它真的做出来了;你会感到自己不如它,会觉得,天啊,这东西要抢走我的工作了;然后它又停下来,回头看着你说:我下一步该做什么?
Speaker 112:49 - 13:04
And I think that is until we have figured out AGI, and maybe even after that, probably after that for a very, very long time, it will always be looking back at us and asking us for direction.
Speaker 112:49 - 13:04
而我认为,在我们真正解决 AGI 之前,甚至也许在那之后——很可能在那之后的很长很长一段时间里——它都会一直回头看着我们,向我们索要方向。
Speaker 213:04 - 13:44
That's basically the core of the argument, because I think you can say, oh yeah, oh yeah, Dan, it's maybe true now that it increases demand for experts, but this stuff's gonna get good enough that it won't. Let's just look at the benchmarks. And there's a whole long section in the piece about, Okay, if you actually do look at the benchmarks, they are improving exponentially. But also, when you look at them closely, once you saturate a benchmark, it's very easy to unsaturate it. It's very easy to find a new frame for a model to do a particular type of problem that is slightly larger, slightly broader, that zeroes it out.
Speaker 213:04 - 13:44
这基本上就是这套论点的核心,因为我觉得你可以说,哦对,哦对,Dan,也许现在确实是这样:它提高了对专家的需求,但这东西最终会变得足够强,以至于不再如此。那我们就看看 benchmarks(基准测试)吧。文章里有很长一段在讲:好,如果你真的去看这些 benchmarks,它们确实在指数级提升。但同时,当你仔细看时,一旦某个 benchmark 被“刷满”,就很容易把它重新“解饱和”。你很容易就能为模型处理某一类问题找到一个新的框架——规模稍微更大一点、范围稍微更广一点——然后它的分数就又会被打回零。
Speaker 213:44 - 14:10
So while it is making exponential progress, it doesn't mean that it is equivalent to human capabilities. It's actually a very hard problem. And one of the reasons it's so hard is anything that you say about what you can do differently than the model is going to be wrong. Because once it's articulated, once it's specified, a model can hill climb on it. A model is going get better at it.
Speaker 213:44 - 14:10
所以,虽然它在取得指数级进展,但这并不意味着它已经等同于人类能力。这其实是一个非常难的问题。而它之所以这么难,其中一个原因是:任何你说“你能做到而模型做不到”的东西,最终都可能是错的。因为一旦它被表达出来、被明确规定出来,模型就可以沿着这个方向做 hill climb(爬坡优化)。模型会在这件事上变得更强。
Speaker 214:11 - 15:24
And we make this weird subtle mistake that we identify a set of tasks, we're like this is all that humans can do, this is what humans can do that models can't do, and then models just do it better, and then you're like oh my god, what do I do? And the mistake is there's actually a lot of stuff that you do that can't be articulated, that can't be articulated in a clean frame, and so every time you try, you just sort of you get you get panicked and confused. And if you sort of step back, the fundamental thing that separation keeps between humans and agents is we are building agents to do things that we want them to do. No matter how powerful they get, all of the economic and psychological and otherwise and technological forces are pushing the progress of AI toward a place where no matter what it does, it's looking back at you to decide what you want to do, what is valuable. And even after we get to AGI, theoretically AGI is going do that too.
Speaker 214:11 - 15:24
而我们会犯一个很奇怪、很微妙的错误:我们先圈定一组任务,然后说,这就是人类能做的一切,这就是人类能做而模型做不到的东西;接着模型只是把这些做得更好,然后你就会想,天啊,那我该怎么办?问题在于,其实你能做的很多事情是无法被清楚表达出来的,无法被放进一个干净的框架里,所以每次你试图这么做时,你都会有点恐慌和困惑。如果你退一步看,人类和 agents(智能体)之间始终存在的根本区别在于:我们在构建 agents,是为了让它们去做我们希望它们做的事。无论它们变得多么强大,所有经济上的、心理上的、其他方面的以及技术上的力量,都在把 AI 的进步推向这样一个方向:不管它做什么,它都要回头看着你,来判断你想做什么、什么才是有价值的。即使我们之后真的走到 AGI(通用人工智能),理论上 AGI 也会这么做。
Speaker 215:24 - 15:59
We thought it wasn't going to do that, we wouldn't build it. And that keeps the gap between humans and AI. I think a good example of this is the difference between something that can do a task really well and something that just has its own self motivated stuff that it wants to do. Have a kid. Codex can.
Speaker 215:24 - 15:59
如果我们认为它不会这样做,我们就根本不会去造它。我觉得一个很好的例子是:一种东西可以把某项任务做得非常好,和一种东西本身有自我驱动、自己就有想做的事,这两者之间是有区别的。生个孩子。Codex can.
Speaker 216:00 - 16:29
I don't know, Codex can write a report much better than Isaiah can. But like Isaiah has very strong wants and needs. And you can try to get him to do what you want and it is going to work sometimes. But also, like he's just this self generating process that does stuff that he wants to do, and if you've ever used any of these tools, you know that there's a very they're not built to work that way. They can push back a little bit, but they don't have this.
Speaker 216:00 - 16:29
我不知道,Codex 写报告可能比 Isaiah 强得多。但 Isaiah 有非常强烈的欲望和需求。你可以试着让他去做你想让他做的事,有时候确实会奏效。但与此同时,他本身就是一个会自我生成的过程,会去做他自己想做的事;而如果你用过这些工具,你就知道,它们很——它们根本不是按这种方式被构建出来的。它们可以稍微顶你一下,但它们并没有这种东西。
Speaker 216:29 - 17:03
It's very far from the playful experimenting, like I just want to do shit because I'm into it that humans have, and again, we're getting into territory of I'm saying things that humans are different than models. Again, these are things that once you clearly articulate them, models can do, but you have to be comfortable with the fact that there are things that you can do and things that you are that you can't fully articulate. Hey, Dan here. We can all agree that housing is expensive. Whether you're renting or paying a mortgage, it doesn't matter which one you're paying, it stings every month.
Speaker 216:29 - 17:03
它距离人类那种带着玩心的试验状态还差得很远——那种“我就是想折腾点什么,因为我自己喜欢”的劲头。再说一次,我们又进入了这样一个区域:我在说一些人类与模型不同的地方。还是那句话,这些东西一旦被你清楚地表达出来,模型就也能做;但你必须接受这样一个事实:有些你能做的事,以及有些“你之为你”的部分,是你无法完全表达出来的。嘿,我是 Dan。我们都同意,住房很贵。不管你是在租房还是在还 mortgage(房贷),你付的是哪一种并不重要,反正每个月都让人肉疼。
Speaker 217:03 - 17:20
But Built can make it feel a bit better. Let me explain. Built lets you earn rewards on your rent, and now you can earn rewards on your mortgage too. Every housing payment earns you points you can use towards flights, towards Lyft rides. The flights you can redeem are with top airline and hotel partners like United and Hyatt.
Speaker 217:03 - 17:20
但 Built 可以让这种感觉稍微好一点。我来解释一下。Built 让你在交房租时赚取 rewards(奖励积分),而现在,你在还房贷时也能赚积分。每一笔住房支出都会给你 points(积分),你可以把它们用在兑换航班、Lyft 打车等方面。你能兑换的航班来自顶级航空公司和酒店合作伙伴,比如 United 和 Hyatt。
Speaker 217:21 - 17:42
I'd be redeeming my points for business class travel, pick your poison. But here's what I think is the most underrated part. Built members also get access to a neighborhood concierge. It can make restaurant reservations, book fitness classes, and find new local spots, and it comes with rewards at over 45,000 retail and merchant partners. It's sort of like having a personal assistant baked into where you live.
Speaker 217:21 - 17:42
如果是我,我会把积分拿去兑换 business class(商务舱)旅行,随你挑你喜欢的。但我觉得最被低估的其实是这一点:Built 会员还能使用 neighborhood concierge(社区礼宾)服务。它可以帮你订餐厅、预订健身课程、寻找新的本地去处,而且还附带覆盖超过 45,000 家零售与商户合作伙伴的 rewards(奖励)权益。这有点像把一个私人助理直接内置进了你的居住体验里。
Speaker 217:42 - 17:57
It's simple. Being a renter and now owning a home is better with BILT. Join the membership for where you live at joinbilt.com/dan. That's joinbilt.com/dan. Make sure to use that URL so they know that we sent you.
Speaker 217:42 - 17:57
很简单。无论你是租房者,还是现在已经拥有自己的房子,用 BILT 都会更好。前往 joinbilt.com/dan,加入与你居住地相关的会员计划。网址是 joinbilt.com/dan。一定要使用这个 URL,这样他们才知道是我们推荐你去的。
Speaker 217:57 - 17:59
And now, back to the episode.
Speaker 217:57 - 17:59
现在,回到这一期节目。
Speaker 117:59 - 18:04
It is also inside of that play Yeah. And that rejection
Speaker 117:59 - 18:04
它也包含在那种发挥空间里,对。还有那种拒绝。
Speaker 218:04 - 18:04
Yeah.
Speaker 218:04 - 18:04
对。
Speaker 118:05 - 18:46
Where you have autonomy. And it will be a very scary moment when these models can do that. And I think there's a question of can they even do that because they rely on training data, and that needs to be in the training data. And maybe there's a world in which they are continually learning and we lose control of them and they start getting access to training data that we don't want them to have access to. But until that time, I think there's probably a good argument that they can't reject what we're saying, and therefore can't be autonomous.
Speaker 118:05 - 18:46
也就是在你拥有 autonomy(自主性)的情况下。而当这些模型能够做到那一点时,那会是一个非常可怕的时刻。我认为还有一个问题是,它们到底能不能做到,因为它们依赖 training data(训练数据),而那种能力需要存在于 training data 里。也许会有这样一种世界:它们持续学习,我们失去对它们的控制,而它们开始接触到我们并不希望它们接触的 training data。但在那之前,我认为有一个相当有力的论点是,它们无法拒绝我们所说的话,因此也就不能算是 autonomous(自主的)。
Speaker 118:46 - 18:54
Autonomy needs to be, I've asked you to analyze this CSV, and it says no, because this is a better idea than doing that.
Speaker 118:46 - 18:54
所谓 autonomy,应该是这样:我让你去分析这个 CSV,而它会说“不”,因为它觉得有一个比那样做更好的主意。
Speaker 218:54 - 19:36
Yeah, and I would actually substitute, I think a better word, I think agent is very confusing because it implies agency, but agent means something that acts on behalf of someone else. I these are agents that are getting very good at being autonomous in the sense that if I send you out on a task, whatever that task is, that task could be disagree with every single thing I say. It could be I go off and find a new idea, whatever that task is, they're getting or will be very good at that, but that is very different from having agency, which is what even the smallest child has.
Speaker 218:54 - 19:36
对,而且我其实会替换这个词。我觉得有个更好的词;我认为 agent 很容易让人困惑,因为它会让人联想到 agency(主观能动性),但 agent 的意思其实是代表他人行动的东西。我认为,这些 agent 正越来越擅长在某种意义上表现得 autonomous,也就是说,如果我派你去执行一项任务,不管那项任务是什么——它甚至可以是反对我说的每一句话,也可以是让我去找到一个新点子——无论任务是什么,它们都正在变得、或者将会非常擅长完成这类任务。但这和拥有 agency 完全不同,而 agency 是连最小的孩子都具备的东西。
Speaker 319:36 - 19:36
Yeah,
Speaker 319:36 - 19:36
对,
Speaker 219:37 - 19:56
I don't think that there's a lot of, there's not a lot of incentive to build that because, okay, you're sitting down at your computer, you're like, hey, let's get to work, and the agent's like, nah, I'm playing. It needs to be able to do that in order to do
Speaker 219:37 - 19:56
我不认为人们有很强的动力去构建那种东西,因为,假设你坐到电脑前,说“嘿,开始工作吧”,结果 agent 说:“不要,我要玩。” 它得能够那样做,才能做到那一点。
Speaker 119:56 - 19:58
things that are scary to us.
Speaker 119:56 - 19:58
那些让我们感到害怕的事情。
Speaker 219:58 - 20:27
Yeah, that's what I that's what I think. There's this obviously, there's a gigantic literature on less wrong in other places about like why it's impossible to prove that they're never gonna do that or whatever. But my counter to that is the evidence, if you look at the development of these things, the evidence is that in the whole lineage is toward being more compliant, and I think the entire industry is incentivized to do that, and I see no reason to doubt that that's going to continue to be the case. Yeah,
Speaker 219:58 - 20:27
对,这就是我的看法。显然,在 LessWrong 和其他地方,有大量文献讨论为什么不可能证明它们永远不会那么做之类的问题。但我的反驳是,看证据的话,如果你观察这些东西的发展,证据表明,整个演化脉络都是朝着变得更 compliant(更顺从、更服从指令)的方向走的,而且我认为整个行业都有动力去这么做,我也看不出有什么理由怀疑这种情况不会继续下去。对,
Speaker 120:28 - 20:39
I mean, we'd have to develop something that's like this. It's your definition of AGI, which like is a good question of whether that's actually possible, which maybe you should explain to everyone what what AGI is.
Speaker 120:28 - 20:39
我的意思是,我们必须得开发出某种像这样的东西。这就涉及你对 AGI 的定义了,而这本身也是个很好的问题:那东西到底是否真的可能实现。也许你应该先向大家解释一下 AGI 是什么。
Speaker 220:39 - 21:19
I think a good I think a good definition of AGI is any agent that you never turn off, that it makes economic sense to keep it running all the time, and keep it running all the time in the sense of Open Clause or Victor or whatever, you can ping it and it will respond to you all the time, it's the server's on, but I mean generating tokens, actively doing tasks for you without you ever turning it off or having to re prompt it. You can probably, like you can guide it or whatever, but the idea is it's valuable enough that it can just keep running all the time.
Speaker 220:39 - 21:19
我觉得,对 AGI 一个不错的定义是:任何一个你永远不会关掉的 agent(智能体),也就是说,从经济上讲,让它一直运行是划算的;而且“一直运行”指的是像 Open Clause 或 Victor 之类的那种,你随时 ping(发起请求)它,它都会随时回应你,服务器一直开着。但我的意思不只是服务器开着,而是它会持续生成 token、主动替你执行任务,而不需要你把它关掉,也不需要你反复重新 prompt(提示)。你当然可能还是会引导它之类的,但核心概念是:它足够有价值,所以可以一直持续运行。
Speaker 121:19 - 21:39
Okay, I want one word answers for the next two questions. Do you think that will happen? Do you think that is a good thing? Explain your reasoning for the second answer. Here's the reasoning for my question.
Speaker 121:19 - 21:39
好,接下来两个问题我想要你用一个词回答。你觉得那会发生吗?你觉得那是好事吗?请解释你对第二个答案的理由。下面是我这么问的原因。
Speaker 121:41 - 21:58
That to me seems to be where things start to get a little off the rails, where it makes economic sense for these things to run all the time, because then I sort of start to think, okay, it's actually valid that the ClickUp guy just fired 20% of his Yeah.
Speaker 121:41 - 21:58
对我来说,那似乎就是事情开始有点失控的地方了:当这些东西从经济上讲值得一直运行的时候,因为那时我就会开始觉得,嗯,ClickUp 那个人裁掉他公司 20% 的员工这件事,也许还真是站得住脚的。对。
Speaker 221:59 - 22:01
Okay. We should we should definitely go back to the ClickUp guy.
Speaker 221:59 - 22:01
好。我们确实应该回到 ClickUp 那个人身上。
Speaker 122:01 - 22:03
Let's go back to ClickUp guy. What's his name?
Speaker 122:01 - 22:03
我们回到 ClickUp 那个人吧。他叫什么名字?
Speaker 222:03 - 22:04
I don't know.
Speaker 222:03 - 22:04
我也不知道。
Speaker 122:04 - 22:06
I think ClickUp guy is good. ClickUp guy.
Speaker 122:04 - 22:06
我觉得 ClickUp guy 挺不错的。ClickUp guy。
Speaker 222:08 - 22:22
But before we get there, like, the the the thing that is important to not fall into when you project out like this is everybody will have access to this.
Speaker 222:08 - 22:22
但在谈到那一步之前,像这样做前瞻性推演时,重要的是不要掉进这样一个误区:就是认为每个人都会获得这种能力的使用权。
Speaker 122:22 - 22:22
Mhmm.
Speaker 122:22 - 22:22
嗯。
Speaker 222:24 - 22:45
For one. For another, the rate of change, even when crazy new technology is available, is actually a lot slower than you would expect. So as part of this piece, I didn't end up covering it because I think it requires a lot more space. I was already 8,000 words, and I was not sleeping anyway. So I was like, I'm going to cut this.
Speaker 222:24 - 22:45
其一。其二,即便有疯狂的新技术可用,变化的速度其实也比你想象的要慢得多。所以这也是这篇文章的一部分,我最后没有写进去,因为我觉得这需要更多篇幅。我当时已经写到 8,000 字了,而且本来也没在睡觉。所以我就想,这段我还是删掉吧。
Speaker 222:45 - 23:06
But as part of this, I wanted to say, I wanted to see how does this work? I know how it works in expert knowledge work, fast moving stuff. I know how it works. We have customer service, so I know how it works if you're a customer service manager type person. But how does AI actually affect your job if you're a customer service person in Omaha or whatever and you work in a call center?
Speaker 222:45 - 23:06
但作为这件事的一部分,我想说,我想看看这到底是怎么运作的。我知道它在专家型知识工作、快速变化的事情里是怎么运作的,我知道。我也了解客户服务这个场景,所以如果你是 customer service manager(客服经理)这类人,我知道它会怎么运作。但如果你是在 Omaha 之类地方做 customer service(客服)的人,在 call center(呼叫中心)上班,那 AI 实际上会怎样影响你的工作?
Speaker 223:06 - 23:26
Because those are the most at risk employees. That would be the default example to bring up. So I like, I want to just see what that's like. And so I just had Codex and Cloud Code scrape all of Reddit and lots of places where customer service reps post. And obviously a lot of them don't like AI, which makes sense.
Speaker 223:06 - 23:26
因为这些员工是风险最高的一群人。这会是默认会被拿出来举的例子。所以我就想,我想直接看看那到底是什么样。于是我就让 Codex 和 Cloud Code 抓取了整个 Reddit,以及很多 customer service reps(客服代表)发帖的地方。很明显,他们当中很多人都不喜欢 AI,这也说得通。
Speaker 223:27 - 23:43
But there's some really interesting stories there about companies that they jump on the AI bandwagon, they're like automating everything. They fire a bunch of their customer service people. And then, like, two months later, they're like, oops. Sorry. Like, can can you come back?
Speaker 223:27 - 23:43
但里面也有一些特别有意思的故事:有些公司一头跳上 AI bandwagon(AI 热潮),觉得要把一切都自动化。他们裁掉了一大批 customer service(客服)人员。然后,大概两个月后,他们就会说,哎呀。不好意思。那个……你能回来吗?
Speaker 223:43 - 24:01
And one and one reason for that is if you implement AI poorly, you're going to have poor results, and I think a lot of these companies don't really understand what they're doing, and they just are paying lip service to the new hype, they think the CEO thinks that they can cut a bunch of expenses, and then it just doesn't really work very well.
Speaker 223:43 - 24:01
而其中一个原因是,如果你把 AI 实施得很糟,结果就会很糟;而且我觉得这些公司里很多人其实并不真的明白自己在做什么,他们只是对这波新 hype(炒作)做做样子,CEO 以为这样可以砍掉一大笔成本,但最后实际效果并不好。
Speaker 124:01 - 24:03
Lot of those people haven't actually played with it.
Speaker 124:01 - 24:03
那些人里有很多其实根本没亲自上手玩过它。
Speaker 224:03 - 24:45
Yeah, exactly. But another reason, which I think is really interesting and is very important, is a lot of people who call in to customer service centers do not want to talk to a machine, do not, and are very explicitly trying to figure out are you a machine or not and get to a fucking human. And that is a real break on how fast these kinds of things can be adopted, and that's only one example. The world is very complicated, there's billions and billions of examples for any kind of job, And so I think it's really important even if we're hypothesizing this like this thing that can that's always on that can do stuff. One, we have to hypothesize everyone has access to it because that is the direction that it's going.
Speaker 224:03 - 24:45
对,完全是这样。但另一个我觉得非常有意思、也非常重要的原因是,很多打电话到 customer service centers(客服中心)的人根本不想和机器说话,完全不想,而且会非常明确地试图搞清楚你到底是不是机器,然后想办法转到一个他妈的真人那里。这会实实在在地拖慢这类东西被采用的速度,而这还只是一个例子。这个世界非常复杂,任何一种工作都有几十亿、几百亿种案例。所以我觉得,即便我们在假设有这么一种东西——它永远在线、能够做事——也真的很重要。第一,我们必须假设每个人都能接触到它,因为事情发展的方向就是这样。
Speaker 224:46 - 25:00
And two, we should recognize that even if that happens, it will take a long time for it to become something that everybody is comfortable with and everyone uses, and it will take probably a generation for it to really turn into a thing.
Speaker 224:46 - 25:00
第二,我们也应该认识到,即便那真的发生了,要让它变成一种所有人都感到舒服、所有人都会使用的东西,也需要很长时间,而且很可能要花上一代人的时间,它才会真正成为一种成型的事物。
Speaker 125:02 - 25:06
There's also a good argument that like working at a call center is not a job that anybody
Speaker 125:02 - 25:06
还有一个很有力的观点是,比如说,在 call center(呼叫中心)工作,本来就不是一种任何人都会——
Speaker 225:06 - 25:07
It's not great.
Speaker 225:06 - 25:07
都算不上什么好工作。
Speaker 125:07 - 25:25
It's a job that you have to do, because you need a job, and in a world where this technology exists, yes, we'll have to figure out a way that everybody can live a fulfilling life and eat. But it might actually be nice to not have that job, assuming you're taking care of another one.
Speaker 125:07 - 25:25
这是一份你必须去做的工作,因为你需要一份工作;而在一个这种技术存在的世界里,是的,我们将不得不想办法让每个人都能过上有成就感的生活,也能吃得上饭。但如果你还有另一份工作可以照应,实际上不用做这份工作,可能反而是件好事。
Speaker 225:25 - 26:17
I think obviously the transition a big deal, and these are real people with real lives, some actually do love it, but also, yes, in general, being yelled at in a call center is not the best job. But I think that where I'm going is even if we hypothesize that, humans still have to decide what matters and what matters changes all the time. And it changes all the time in particular because AI is an input to that, so it is both how do I even say this? It's very recursive. AI is changing the world really fast, which changes what matters, which puts more onus on us to update and decide what matters, because AI is gonna wait for us to be like what matters, Totally.
Speaker 225:25 - 26:17
我觉得很显然,转型本身是一件大事,这些都是真实的人、真实的生活;有些人其实也确实很喜欢这份工作。但同时,是的,一般来说,在 call center(呼叫中心)里被人吼来吼去,并不是什么最好的工作。不过我想表达的是,即便我们作出那样的假设,最终仍然必须由人类来决定什么是重要的,而“什么重要”又一直在变化。它之所以一直在变化,尤其是因为 AI(人工智能)本身就是其中的一个输入因素,所以这既是——我该怎么说呢?这是非常 recursive(递归)的。AI 正在非常快速地改变世界,这又会改变什么是重要的,而这会把更多责任压到我们身上,要求我们不断更新、不断决定什么是重要的,因为 AI 会等着我们告诉它:什么才是重要的。完全同意。
Speaker 226:17 - 26:56
You And that is going to be part of every job, because anything that you decide, anything that you can frame and be like this is a repetitive thing that is working, you can just have your have your AI do it. The minute the situation changes, and situations change all the time, and they especially change all the time when it's not just humans changing, it's AI, you're going to need humans to decide that. And I think that that's I think that's something that's very missing from from, you know, what we talk about when we hypothesize these things. Back to the ClickUp guy.
Speaker 226:17 - 26:56
对,而这将成为每一份工作的一部分,因为凡是你能够决定、能够界定,并且能够说“这是一个运转良好的重复性事务”的东西,你都可以直接让你的 AI 去做。但情况一旦发生变化——而情况一直都在变化,尤其当不只是人类在推动变化,而是 AI 也在推动变化时——你就仍然需要人类来作出判断。我觉得这一点在我们谈论、假设这些事情时,恰恰是非常缺失的。说回那个 ClickUp guy。
Speaker 126:56 - 26:57
ClickUp guy.
Speaker 126:56 - 26:57
ClickUp guy。
Speaker 226:57 - 27:01
Yeah. So I don't know. He fired 30,000 people.
Speaker 226:57 - 27:01
对啊。所以我也不知道。他裁掉了 30,000 人。
Speaker 127:02 - 27:04
I think it was more.
Speaker 127:02 - 27:04
我觉得还不止。
Speaker 227:05 - 27:16
And I think it's really important whenever you're looking at some of this shit on Twitter. First of all, I fuck. I hate I hate when they're like, our business is better than it's ever been, and we laid off 8,000 people.
Speaker 227:05 - 27:16
而且我觉得,你在 Twitter 上看这类破事的时候,真的很重要的一点是——首先,我他妈最讨厌他们那种说法:我们公司的业务比以往任何时候都更好了,然后我们裁掉了 8,000 人。
Speaker 127:16 - 27:18
Yeah. It's pretty fucked.
Speaker 127:16 - 27:18
对,真的挺操蛋的。
Speaker 227:18 - 27:22
Yeah. It's like, well why would you brag about that?
Speaker 227:18 - 27:22
对。就像,呃,你为什么还要拿这种事来炫耀呢?
Speaker 127:22 - 27:57
The other thing that I don't like is, we're gonna pay people a million dollars if they do great work, it's sort of like okay, but you still have all these people that no longer have jobs. I just really don't think it's very tastefully done, and I think Jensen, he said something that was very self serving, which was basically like, if your answer to progress is firing people, you're not a very creative CEO. Very self serving, because obviously he wants people to use more AI, But I think it's true. It's fairly like you should be doing more interesting things, not firing, you know, people want to be profitable. Yes.
Speaker 127:22 - 27:57
另一个我不喜欢的点是,他们会说,如果有人干得特别好,我们就给他一百万美元——感觉就像,好吧,但你这边还有这么多人已经没工作了。我真的觉得这件事做得很不得体,而且我觉得 Jensen 说过一句很利己的话,大意基本上是:如果你面对进步的答案就是裁员,那你就不是一个很有创造力的 CEO。非常利己,因为显然他是想让大家更多地使用 AI,但我觉得这话是对的。说到底,你本来就应该做些更有意思的事,而不是裁员。你知道的,人们都想要盈利。是的。
Speaker 127:57 - 28:11
But idiots. But yeah, just it's it's anyway, that's an aside. It just is not very tastefully done. Yeah. Anyway.
Speaker 127:57 - 28:11
但都是些白痴。不过,是啊,这个先扯远了。总之,这事做得就是很不得体。对。反正。
Speaker 228:11 - 28:24
I I so so a not tasteful, which should make you a little bit suspicious. And my guess is, and just seeing some of the random stuff, is I don't think the company is doing that well.
Speaker 228:11 - 28:24
我,我,所以,总之,不得体这点本身就应该让你有点怀疑。我的猜测是,而且从我看到的一些零碎情况来看,我觉得这家公司经营得并没有那么好。
Speaker 128:25 - 28:28
I mean, it's a generic SaaS company. Yeah. Yes,
Speaker 128:25 - 28:28
我的意思是,这就是一家很普通的 SaaS 公司。对。是的,
Speaker 228:29 - 28:34
and when companies don't do well, they lay people off. Meta
Speaker 228:29 - 28:34
而且当公司经营不好时,就会裁员。Meta
Speaker 128:35 - 28:38
Or when companies are managed poorly and have too much bloat anyway.
Speaker 128:35 - 28:38
或者,当公司管理不善、内部本来就有太多冗员时,也会这样。
Speaker 228:38 - 28:41
Exactly, which is correlated with not doing well.
Speaker 228:38 - 28:41
没错,而这通常和经营不佳是相关联的。
Speaker 128:41 - 28:45
That was Square. Like that Jack Dorsey just can he just does that?
Speaker 128:41 - 28:45
那说的是 Square。就像,Jack Dorsey 他就是会干这种事吗?
Speaker 228:45 - 28:56
And I think meta is the same, like they're making gigantic investments in AI because that's like the new hot shit that they kind of missed the boat on, and
Speaker 228:45 - 28:56
而且我觉得 Meta 也是一样,他们正在对 AI 进行巨额投资,因为那就像是他们某种程度上错过了风口之后,现在最新、最火的东西,而且
Speaker 128:56 - 28:57
And no surprise metaverse didn't work, so
Speaker 128:56 - 28:57
而且,metaverse 没做成也不奇怪,所以
Speaker 228:57 - 29:37
now they have a lot of people on Yes, so I think yes, AI is involved in all of this stuff, but it's not like this clear thing of everyone's doing all the same jobs as before, but they're all just agents. No, no, the company actually has totally changed strategies and the people it needs and the structure it needs is totally different. And that's not the clean narrative that I think people like to tell, and it's much easier to talk about just AI takes jobs. It seems it seems definitely true that using these tools changes your workflow a lot. And because it changes your workflow, it changes what's hard and what's easy.
Speaker 228:57 - 29:37
现在他们那边有很多人。对,所以我觉得,是的,AI 确实牵涉到这一切,但并不是那种很简单清晰的情况:所有人还在做和以前完全一样的工作,只不过现在他们全都成了 agent。不是,不是,实际上公司的战略已经彻底变了,它需要的人、它需要的组织结构,也都完全不同了。而这并不是人们喜欢讲的那种简单叙事;说成“AI 抢走了工作”要容易得多。看起来,使用这些工具确实会很大程度上改变你的 workflow(工作流)。而因为它改变了你的 workflow(工作流),它也就改变了什么是难的、什么是容易的。
Speaker 229:40 - 30:33
Especially if you're a big company and you've been structured in a certain way to work in a certain way, there's going to be reorganizations of how work happens and how companies are structured. Seems really clear, and it's very important that we figure out a way to make that transition as good as possible for people and and tweeting about how well you're doing it while you're firing people is not that. I think there's a lot of really interesting, like, creative ways to, you know, Meta, for example, is now key logging everyone's everyone's all the stuff they enter into their computer because they're like, well, our people are the smartest people. So we'll just use their data to train our models and our models will be smarter, which is an interesting take and maybe it'll work. But I think there's this really interesting I wrote about this like two years ago or something like that.
Speaker 229:40 - 30:33
尤其是如果你是一家大公司,而且你们已经按照某种方式被组织起来、也一直按某种方式运作,那么工作的发生方式以及公司的组织结构都会经历重组。这一点看起来非常明确,而我们必须找到一种办法,让这种转变尽可能对人更友好;一边裁员一边在 Twitter 上吹嘘自己做得多好,显然不是这种办法。我觉得这里面其实有很多非常有意思、也很有创造性的做法。比如说,Meta 现在会记录每个人在电脑里输入的所有内容,因为他们的逻辑是:我们的员工是最聪明的人,所以我们就用他们的数据来训练我们的模型,这样我们的模型也会更聪明。这是个很有意思的思路,也许真的会奏效。我想这里面有个特别值得讨论的点,我大概两年前写过这件事。
Speaker 230:33 - 31:15
There's this really interesting effect of that, which is when you sign an employment contract, the way that we thought about employment for a very long time was, I'm going to do this job and you're going to need me to keep doing it in order for it to keep getting done. But once you reach a point where I do the job for you and then it just works, sort of, and then you can just, you don't have to pay me anymore, that sort of changes the whole way that we think about employment, and therefore I think it should, for example, change. How we think about paying certain types of people?
Speaker 230:33 - 31:15
这里有一个非常有意思的影响:当你签一份 employment contract(雇佣合同)时,我们在很长时间里理解 employment(雇佣关系)的方式都是——我来做这份工作,而为了让这项工作持续被完成,你就得持续需要我。但一旦到了这样一个点:我先替你把这项工作做出来,然后它某种程度上就能自己运转了,于是你也就不必再继续付我钱了,那这就从根本上改变了我们理解 employment(雇佣关系)的方式。因此我认为,比如说,这也应该改变我们看待某些类型的人应当如何获得报酬的方式。
Speaker 131:16 - 31:21
You should get a pension, you know? Pension. Maybe maybe pensions are back. Pensions are back, baby.
Speaker 131:16 - 31:21
你应该拿 pension(养老金),懂吗?Pension。也许——也许养老金要回归了。养老金回来了,baby。
Speaker 231:22 - 31:48
Well, one thing that's really interesting is there's this thing that launched last week that we're part of. The name is escaping me, but it allows publishers to get paid based on. Basically, it measures a publisher's unique contribution to the training corpus and you get paid based on that. So the more generic your shit is, the less you get paid, and the more unique and valuable it is, the more you get paid, which is really interesting.
Speaker 231:22 - 31:48
有件事很有意思:上周有个项目上线了,我们也参与了。名字我一时想不起来了,但它允许 publishers(出版方/内容发布方)基于某种衡量方式获得报酬。基本上,它会衡量某个 publisher 对 training corpus(训练语料库)的独特贡献,然后你就按这个获得报酬。所以你的内容越 generic(通用、模板化),你拿到的钱就越少;而你的内容越独特、越有价值,你拿到的钱就越多,这一点非常有意思。
Speaker 131:48 - 32:01
The ironic thing about that is basically is like, did you this will be the case, did you use AI, which is trained off of all the shit that already exists? So it still can make some things that are new, but like it's basically, you know, it's
Speaker 131:48 - 32:01
这里讽刺的地方基本上在于:这最终会变成,你是不是用了 AI,而这个 AI 本身又是拿所有已经存在的那些东西训练出来的?所以它当然还是能产出一些新的东西,但本质上,你知道,它其实就是——
Speaker 232:01 - 32:05
How much just like generic default prompting did you do to make this, versus just
Speaker 232:01 - 32:05
你为了做出这个东西,到底只是做了多少那种 generic(通用、默认式)的 prompting(提示词输入),还是说——
Speaker 132:06 - 32:09
Did like a human actually think about this? Yeah, exactly. Generate a new idea.
Speaker 132:06 - 32:09
这里面到底有没有一个 human(人)真的认真思考过?对,完全就是这个问题。有没有真正产生一个新想法。
Speaker 232:09 - 32:33
So, but I think there's there could be something similar for I had this idea, I wrote this post a couple years ago about the last job you'll ever had, or you'll ever have, where it's an agency, you generate all the training data in the work that you do for the agency, and then it tracks basically what is your contribution, and then you just get paid out forever from how much revenue your data generates.
Speaker 232:09 - 32:33
所以,我觉得也许可以有某种类似的机制。我几年前写过一篇文章,讲的是“你最后一份工作”(the last job you'll ever have)。设想是这样的:有一家 agency(代理机构/机构),你在为这家 agency 工作的过程中,会产出所有训练数据,而系统会追踪你到底贡献了什么,然后根据你的数据创造了多少 revenue(收入),你就可以一直持续分成、持续拿钱。
Speaker 132:33 - 32:38
Web three is back now. Web three is back. We're gonna track it all, obviously, on the ledger.
Speaker 132:33 - 32:38
Web3 现在又回来了。Web3 回来了。显然,我们会在 ledger 上追踪这一切。
Speaker 232:38 - 33:06
On blockchain? Yeah, anyway, who knows? The problem with that, again, and this is this is back to why humans are valuable, but they're not that's not the only reason why humans are valuable. We're valuable intrinsically, but one of the reasons why they're valuable for work is I I would guess, looking back at that article and looking and thinking about a lot of the stuff is there's a really high drop off rate. There's a really high, what's that word?
Speaker 232:38 - 33:06
在 blockchain 上?对,不过谁知道呢?这里的问题还是一样,这也又回到了为什么 humans 很有价值,但那并不是 humans 有价值的唯一原因。我们的价值本身就是内在的,不过从工作的角度看,humans 之所以有价值,其中一个原因是——我猜,回头看那篇文章,再想想很多相关的事情——流失率真的非常高。还有一个很高的,那个词叫什么来着?
Speaker 233:06 - 33:19
There's really high depreciation of the value of data. Once it's once it's out there, it's like very likely to go stale within like weeks. There's some things that maybe not, but
Speaker 233:06 - 33:19
data 的价值贬损非常快。一旦它被放出来,它很可能在短短几周内就过时了。当然有些东西也许不会,但
Speaker 133:21 - 33:27
It's safe to say that all of these companies are at a place where they are just hunting for net new unique data.
Speaker 133:21 - 33:27
可以肯定地说,所有这些公司现在都处在一个阶段:它们只是在四处搜寻全新的、独特的 data。
Speaker 233:27 - 34:40
Yeah, think so, and so anyway, we should expect broad reorganizations of companies and we should expect companies that are not doing well to lay people off or reorganize and then blame AI. And I would really be skeptical of anyone who's saying that it's going to eliminate all jobs or all knowledge work, and I think it will certainly change them, and I think it is certainly, it's a big thing that people have to take seriously, but my big takeaway and this is not fully in the piece, but it is what I really believe is if you just ride the models, if you just, when new models come out, learn to use them for the stuff that you do, whatever that is, you're going to be fine. And you may even hopefully find that you can do more and better work that's more fulfilling for you than you could before. I think that there's still a place in the world if you don't wanna use the models at all, think that that's still gonna be a thing. Plenty of people don't eat fast food or whatever.
Speaker 233:27 - 34:40
对,我觉得是这样。所以总之,我们应该预期公司会发生大范围重组,也应该预期那些经营不佳的公司会裁员或重组,然后把责任推给 AI。至于那些说它会消灭所有工作、或者所有知识型工作的说法,我会非常怀疑。我认为它肯定会改变这些工作,我也认为这毫无疑问是一件人们必须严肃对待的大事。但我最大的结论——这一点在文章里没有完全展开,不过这确实是我真正相信的——是:如果你顺着 models 走,如果每次有新 models 出来,你就学会把它们用到你自己的工作里,不管你的工作是什么,你都会没事的。你甚至还可能会发现,自己能做比以前更多、更好的工作,而且这些工作对你来说也更有满足感。我也认为,如果你完全不想用 models,这个世界里仍然会有你的位置,我觉得那也仍然会是一种选择。就像很多人本来就不吃 fast food 之类的东西。
Speaker 234:40 - 35:02
I don't know what to compare it to. It's totally possible not to participate in this. However, if you care about leading a really ambitious life and building businesses or whatever it is, I truly think that this is going to make that more possible for more people, and as long as you ride the models, you're gonna be good.
Speaker 234:40 - 35:02
我不知道该拿什么来类比。完全有可能不参与这件事。不过,如果你在乎过一种真正有抱负的人生,想要创业也好,或者做别的什么也好,我真心认为这会让更多人更有可能做到这些;只要你顺着 models 走,你就会过得不错。
Speaker 135:04 - 35:13
I think that's a very good call to action. I want to end by asking you something about what it takes to write a piece like this.
Speaker 135:04 - 35:13
我觉得这是一个非常好的行动号召。最后我想问你一个问题:写出这样一篇文章,需要具备什么?
Speaker 235:13 - 35:14
A lot of Celsius.
Speaker 235:13 - 35:14
大量的 Celsius。
Speaker 135:14 - 35:29
A lot of Celsius. So when we started, I don't know if this will make it into the podcast, but when we started, Dan was sort of like looking like this, he was hugging himself, protecting himself, some would say. It has been a very stressful week. This is an 8,000 word piece. Yeah.
Speaker 135:14 - 35:29
很多很多的 Celsius。所以我们开始的时候,我不知道这段会不会被放进 podcast,但刚开始时,Dan 大概是那样看起来的——他抱着自己,有些人会说,这是在保护自己。那真是压力非常大的一周。这可是一篇 8,000 词的文章。对。
Speaker 135:32 - 35:45
Most people are not writers. Can you share what it's like to not just write an 8,000 word piece, which is a very big piece, but what does it take to think through these arguments?
Speaker 135:32 - 35:45
大多数人都不是 writer。你能不能讲讲,写一篇 8,000 词的文章——这已经是非常大的篇幅了——不仅仅是写出来而已,要把这些论点想清楚,到底需要什么?
Speaker 235:45 - 36:14
It's so interesting because it's very natural to me, because I wrote once a week. I published once a week for so long that especially like a, you know, 500 word or a thousand word piece, like I can just bring that in like an hour or two. These things are get. These things get much harder the longer they go because there's all these interdependencies. If you change something here, it changes four other things over here and and whatever.
Speaker 235:45 - 36:14
这很有意思,因为对我来说这其实非常自然,因为我以前是每周写一次。我长期保持每周发布一次,所以尤其是像 500 词或者 1,000 词这样的文章,我一两个小时就能写出来。这种东西越长就越难,因为里面有很多 interdependencies(相互依赖关系)。你改了这里的一个地方,那边另外四个地方都会跟着变,诸如此类。
Speaker 236:14 - 36:29
8,000 words becomes like, it's like 10 times harder than 4,000 words, which is 10 times harder than 400. I found that, and I always have this feeling that there's this underlying thing that I can feel, but I can't quite say that I'm trying to say.
Speaker 236:14 - 36:29
8,000 词会变成一种……它大概比 4,000 词难 10 倍,而 4,000 词又比 400 词难 10 倍。我发现确实是这样。而且我总有一种感觉:好像有个底层的东西我能感觉到,但我就是说不太出来我到底想表达什么。
Speaker 136:29 - 36:30
And
Speaker 136:29 - 36:30
还有,
Speaker 236:33 - 36:52
it started actually, if you remember we did our, I guess it was Q2 planning, and I was like, I think that we can I think I figured out why? This is after I did proof. I think I figured out why we're just going to always have jobs with AI. And like, you just rather ride the models, you're going to be fine. Like, I think I I think I can feel that.
Speaker 236:33 - 36:52
这件事其实是这样开始的,如果你记得的话,我们当时做了——我想应该是 Q2 planning——然后我当时说,我觉得我们可以……我觉得我想明白原因了?这是在我做完 proof 之后。我觉得我想明白了,为什么在 AI 时代我们就是会一直有工作可做。而且只要你愿意顺着 models 走,你就会没事。就像,我觉得我……我觉得我能感觉到那个东西。
Speaker 236:52 - 37:13
And then it was just this process to be like, Okay, how does that actually cash out? Like, why do I think that? Because it's all kind of in there, but it's all tangled up. And I wrote like probably four or five versions where I would start it and I was like making the argument, was like, that doesn't work. And then I would be like, Oh, but how about this?
Speaker 236:52 - 37:13
然后接下来就是一个梳理过程:好,那这到底具体意味着什么?比如,为什么我会这么想?因为这些东西其实都在脑子里,但全都缠在一起。我大概写了四五个版本,每次一开头,我觉得自己在构建那个论证,写着写着又会觉得,不行,这个讲不通。然后我又会想,哦,那这个角度怎么样?
Speaker 237:13 - 37:28
And I would like throw it out and like start again. And it was actually very it was a very frustrating process because you're trying what what I'm trying to do is start with the ground truth of here's what we see every
Speaker 237:13 - 37:28
然后我就会把它整个扔掉,重新开始。其实这个过程非常——是一个非常令人沮丧的过程,因为你在尝试……我想做的事情,是从最基本的 ground truth(基本事实)出发:这就是我们每天看到的东西——
Speaker 137:28 - 37:28
day.
Speaker 137:28 - 37:28
每一天。
Speaker 237:28 - 37:40
Here's what here's how work happens for us. And then move into this. Well, like philosophical thing that like it can't actually be articulate. I'm trying to articulate something that can't be articulated.
Speaker 237:28 - 37:40
这是对我们来说事情是怎么推进、怎么完成的。然后再进入这个部分。嗯,有点像那种哲学性的东西——它其实是没法被真正说清的。我在努力表达某种无法被表达的东西。
Speaker 137:40 - 37:43
Yeah. Or is constantly to moving targets.
Speaker 137:40 - 37:43
对。或者说,它始终都在朝着不断移动的目标前进。
Speaker 237:43 - 37:52
Yeah. And so that's that's just like that's very hard. I love that kind of shit, but it's also very, very hard and can be very frustrating. But I was like a huge part of this. Like, could not have written this without it.
Speaker 237:43 - 37:52
对。所以那就是——那真的非常难。我很喜欢这种鬼东西,但它也真的、真的很难,而且会让人非常沮丧。但这又是其中非常非常重要的一部分。没有它,我根本不可能写出这篇东西。
Speaker 237:52 - 38:13
For example, one of the things I loved that I started to do is. You know, for a piece like this, you're trying to articulate it. You can't quite articulate it, and the only way to do that the only way to do it is to articulate it over and over and over again until it works. And you've really got to keep it in your head, especially if you're doing lots of other stuff. So What I would do in the morning is I would.
Speaker 237:52 - 38:13
比如说,我很喜欢、后来开始做的一件事是:你知道,像这样的一篇文章,你在试图把它表达出来,但你又没法完全表达清楚,而唯一的办法——唯一能做到的办法——就是一遍又一遍又一遍地把它说出来,直到它终于奏效。你还真的得一直把它放在脑子里,尤其是当你同时还在做很多别的事情的时候。所以我早上会做的是,我会——
Speaker 238:16 - 38:38
Fresh right when I wake up or you know right when I get my get to my desk, would be like I would just monologue into my computer into a proof document. Here's what the piece is about front to back. Here's the argument front to back, and then I would have a log of that, and every time I would do it I'd be like, Okay Claude or Codex, and I actually use Claude more for this. I think Claude is better for this kind of thinking. What am I really trying to say?
Speaker 238:16 - 38:38
在头脑还很清醒、刚醒来,或者你知道,刚坐到桌前的时候,我会直接对着电脑、对着一个 proof document 自言自语一段:这篇东西从头到尾到底是在讲什么,这个论点从头到尾到底是什么。然后我会把这些都记下来,而且每次我这么做时,我都会说:好,Claude 或者 Codex——实际上这件事我更多用 Claude。我觉得 Claude 更擅长这种思考。那我到底真正想说的是什么?
Speaker 238:38 - 39:15
Help me figure out what I'm trying to say, and it would say things back and I would be like, no, no, that's what I'm trying to say. And then over time you kind of build up this record of here's what it was here, here's what it was here, and and you're just I'm just getting closer and closer and closer. And then what I would do is as I was getting deeper into it and I was like, you know, I have 4,000 words and 5,000 words. Every morning, I would have Codex take the latest draft and turn it into a podcast of just like someone reading reading it to me. And then on my way to work, I would on my way to work, would listen to the podcast.
Speaker 238:38 - 39:15
帮我弄清楚我到底想说什么。然后它会给我一些回应,我就会想,不不不,我真正想说的是那个。慢慢地,你就会积累出这样一份记录:这里它原来是这样,这里它后来变成那样,而你就是——我就是越来越近、越来越近、越来越接近。然后我会做的是,随着我越写越深,比如说,写到 4,000 词、5,000 词的时候,每天早上我都会让 Codex 把最新的 draft(草稿)变成一个 podcast(播客),就像有人把它读给我听一样。然后在我去上班的路上,我就会在路上听这个 podcast。
Speaker 239:16 - 39:23
And as I'm listening, I'm like, okay. That there's something that needs to change there. There's something that needs to change there. Oh, and then it would get to the end of it. Okay.
Speaker 239:16 - 39:23
而我一边听,一边会想,好,这里有些东西需要改,那里有些东西需要改。哦,然后它就会播到结尾。好。
Speaker 239:23 - 39:42
Here's the thing that I need to do next. And that was a really good way to, like, kinda keep the continuity of what am I doing, what am I writing, where are the problems in a way where I'm not always reading. It's really nice to be able to like be on a walk and be listening to it and thinking about it, which would be completely impossible otherwise.
Speaker 239:23 - 39:42
这就是我接下来需要做的事。那是一个非常好的办法,能让我大致保持这种连续性:我到底在做什么、我到底在写什么、问题出在哪里,而且是以一种我不必总是在阅读的方式来做到这一点。能够在散步时听着它、想着它,真的特别好,否则这在别的情况下是完全不可能的。
Speaker 139:44 - 40:02
All right, one more challenge for you, and we're gonna have beers and put together in the backyard. Can you articulate to everybody in one sentence that starts with, if you ride the models, then what this piece is trying to say.
Speaker 139:44 - 40:02
好,再给你一个挑战,然后我们就去后院一起喝啤酒、聚一聚。你能不能用一句以“if you ride the models”开头的话,向大家概括这篇内容想表达什么。
Speaker 240:02 - 40:09
If you ride the models, you're going be okay. You're going to have a job. You're going to do great work, and you don't have to worry.
Speaker 240:02 - 40:09
If you ride the models,你就会没事的。你会有工作,会做出很棒的成果,而且不用担心。
Speaker 140:12 - 40:12
Cheers.
Speaker 140:12 - 40:12
干杯。
Speaker 240:13 - 40:14
Cheers.
Speaker 240:13 - 40:14
干杯。
Speaker 140:18 - 40:19
Okay.
Speaker 140:18 - 40:19
好的。
Speaker 240:19 - 40:20
Alright.
Speaker 240:19 - 40:20
好。
Speaker 140:20 - 40:20
Good stuff, man.
Speaker 140:20 - 40:20
真不错,兄弟。
Speaker 240:20 - 40:21
Good stuff.
Speaker 240:20 - 40:21
真不错。
Speaker 140:21 - 40:21
That was fun.
Speaker 140:21 - 40:21
挺好玩的。
Speaker 340:29 - 40:40
Oh my gosh, folks. You absolutely positively have to smash that like button and subscribe to AI and I. Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard.
Speaker 340:29 - 40:40
天啊,朋友们。你们绝对、百分之百得狠狠干那个 like 按钮,然后订阅 AI and I。为什么?因为这个节目简直就是“精彩绝伦”的代名词。就像你在自家后院发现了一个藏宝箱。
Speaker 340:40 - 41:02
But instead of gold, it's filled with pure unadulterated knowledge bombs about chat GPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show. It's a journey into the future with Dan Shipper as the captain of the spaceship. So do yourself a favor.
Speaker 340:40 - 41:02
只不过里面装的不是黄金,而是关于 chat GPT 的纯粹、原汁原味的知识炸弹。每一期都是一场情绪、洞见和欢笑交织的过山车之旅,让你全程坐在座位边缘,意犹未尽、还想要更多。它不只是一个节目。它是一场通往未来的旅程,而 Dan Shipper 就是那艘宇宙飞船的船长。所以,帮自己一个忙吧。
Speaker 341:02 - 41:11
Hit like, smash subscribe, and strap in for the ride of your life. And now without any further ado, let me just say, Dan, I'm absolutely, hopelessly in love
Speaker 341:02 - 41:11
点个 like,狠狠 smash subscribe,然后系好安全带,准备好迎接你人生中最刺激的一趟旅程。现在,闲话不多说,我只想说,Dan,我彻底、无可救药地爱上你了
Speaker 241:11 - 41:11
with you.
Speaker 241:11 - 41:11
你。
原文 ↗https://www.youtube.com/watch?v=dCmOTURRf1Y
BuildSpeak — 关于本项目BUILT IN PUBLIC · 跟随 builders 而非 influencers