– 英语/中文 双语歌词

歌词与翻译

[中文]

为了帮助大家迈出使用 AI 的第一步，我整理了四个指导原则，通常我会建议大家从这四条开始。

第一个原则是“永远邀请 AI 参与”。在合法且合乎道德的前提下，尽可能在所有事情上使用它。

原因在于，我们其实并不清楚 AI 究竟能做到什么。甚至连你所使用的 AI 系统的开发人员，

也不一定清楚它能为你、你的行业、工作和生活带来什么。你必须亲自去用，才能摸索出来。

如果你有了新想法，可以去问问它；

在征得在场人员同意的情况下，让它帮你撰写备忘录；让它记录会议过程，并对你的表现提出反馈；

让它帮你总结笔记、编写代码、分析数据。

只有亲自去用，你才会知道它擅长什么、不擅长什么。

在这个过程中，你可能会对它出色的能力感到惊喜，

也可能会对某些结果感到失望。这都是探索过程的一部分。所以，你必须通过实践来了解它的运作方式。

我的第二个原则是“保持人类在主导环节”（Human in the loop）。

这个概念源自控制系统领域，指的是最终的决策过程应该始终有人的参与。

我认为这一点至关重要。当我们开始让 AI 执行任务时，必须明确人类在决策链中所处的位置。

这也是一个非常个人化的原则，因为 AI 在许多任务上的表现确实相当出色。

甚至你工作中的某些任务，AI 现在可能做得比你还要好。

因此，你需要思考的是：我究竟想亲自完成哪些任务？

很多时候，我们真正想做的是专注于那些体现人类独特价值、也是自己最喜欢的任务，

并思考如何把不想做的事情交给 AI 来协助完成。

我们应该如何从 AI 那里获取支持，而不是把它视为潜在的替代者或竞争对手？

这背后涉及一系列深刻的问题，需要我们去反思：我们究竟想用自己的生命去做什么？

我们真正想要专注于什么？我们的才能在哪里？

我认为，善用这些才能，是理解 AI 如何赋予我们所有人力量的一种非常有效的方式。

我的第三个原则是“像对待人一样对待 AI，并告诉它它扮演的是什么角色”。

像对待员工或实习生那样与它沟通，你就能解决大部分问题。

除此之外，还要明确告诉它要扮演什么角色。因为 AI 在有上下文背景时表现更好。

你可以通过给予身份设定来提供背景，比如告诉它：“你是一位资深营销专家”或“你是一位课程大纲设计师”。

这样能帮助 AI 理解你提问的意图和方式。

不过，像对待人一样对待 AI 也有风险，那就是你可能会犯下 AI 研究人员的大忌——把计算机误当成人类。

我们有充分的理由不要去假设计算机是人类：

它不是人类，思维方式也与人类不同。如果你把它当成人类，就更容易被它说服，

可能会忽视它的偏见，甚至高估它的能力。

AI 的运作方式有点像占卜师，它非常擅长

揣摩你想听到什么，然后把你想听的话说给你听。

因此，如果你开始把它当作人类来看待，就很容易忽视这一点。

你可以像对待人类一样去使用它，但内心深处必须时刻保持清醒：

“我是在和一台机器对话。它背后没有思想，也没有任何情感或个性。”

第四个也是最后一个原则是“假设这是你用过的最差的 AI”。

我们正处于技术进步的洪流之中，而不是终点。AI 的发展速度极其迅猛，

而且很难被直接衡量。虽然有许多测试分数和测量评估，但倾向于从主观体验来看待它的进步。

想想 2021 年推出的早期 AI 模型 GPT-3，

它的写作水平大约相当于六年级小学生，这在当时已经令人惊叹了。我当时想：“哇，真不敢相信它能写得这么好。”

到了 2022 年底，随 ChatGPT 一起发布的 GPT-3.5，

写作水平已经达到了大二学生的标准；

而 2023 年发布的 GPT-4，其写作水平已经接近博士生一年级的水平。

这是非常惊人的进步速度。我们知道 GPT-5 以及其他技术也在路线图中，

当你看到这段视频时，可能已经可以使用这些新技术了。

这种发展速度相当快。我们无法预知它会发展到什么高度，

也不知道这些系统最终会变得多么强大。我们无法控制这些系统进化得有多快，

但我们能够控制自己如何决定去使用它们，以及如何应用它们。

作为管理者和领导者，你可以做出选择，通过部署这些系统来促成人类的全面发展与繁荣，

而不是仅仅用来取代员工或对他们进行更严格的监控。

作为个体，我们可以决定如何成为那个善用这些系统的主导者，

而不是让系统来指挥我们怎么做。通过这种方式，我们可以掌握许多选择的主导权。

想要观看完整课程，请访问 bigthink.com/membership 注册成为会员。

[英语]

To get you started with AI, I have four guiding  principles that I like to start people with. The  

first principle is always invite AI to the  table. Use it for everything you legally and  

ethically can. And that's because we you don't  know what AI can do. And likely the creators  

of the AI system you're using don't know what  it can do for you and your industry, your job,  

your life. You have to use it to figure it  out. Ask it questions if about idea that  

you have. Help it write a memo for you with  the permission of people in the room. Have it  

record what's happening and give you feedback  on your performance. Summarize notes for you.  

Write code. Analyze data. You won't know what  it's good or bad at until you use it. And you  

might find some really surprising insights of  what it can do well. And you might find some  

disappointments. That's part of the process.  So you have to use it to learn how it works.

My second principle is to be the human in  the loop. And this is an idea from control  

systems that you should always have a human  involved in final decision-m I think that's  

important. We have to decide where humans fit in  decision-m when we start letting AI do things.  

But it's also a deeply personal principle because  AI actually does a lot of tasks quite well. And  

there's a good chance that some of the tasks you  do for your job, the AI might already do better  

than you. Part of what you want to think about  is what tasks do I actually want to do? A lot of  

times what you want to do is actually focus on  what makes you human, what human task you like  

the best and think about how do I give the stuff I  don't want to do to the AI to help me with? How do  

I get support from AI rather than thinking about  it as a potential replacement or as a competitor?  

And I think there's a deep set of questions we  need to think about about what do we actually  

want to do with our lives? What do we actually  want to focus on? Where is our talent? And using  

that talent, I think, is a very powerful way  of thinking about how an AI empowers us all.

My third principle is to treat AI like a person  and tell it what kind of person it is. Talk to it  

like a an employee, like an like an intern, and  you'll get a large part of the way there. And  

on top of that, treat tell it what kind of person  it is. So AI works better with context. Give it a  

context by telling it you are an expert marketer.  You are a lesson plan designer. And that gives the  

AI context to understand what kind of questions  you're asking and how. The danger of treating the  

AI like a person is you're committing the cardinal  sin of AI researchers, which is to pretend that a  

computer is a human. And there's a lot of reasons  you don't want to pretend a computer is a human.  

It's not a human. It doesn't think like a human.  You might start to become more persuaded by it.  

You might become blind to its biases. You  might think it's more capable than it is.

AI kind of works a little bit like  a psychic. It's it's really good at  

figuring out what you want to hear from  it and then telling you what you want to  

hear. And so it's very easy to lose track  of that if you start treating like a human  

being. Use it like a human, but also  always keep in the back of your mind,  

I'm talking to a machine. There's no mind  behind this. There's no emotion or personality.

The fourth and final principle is assume this  is the worst AI you're ever going to use.  

We are in the middle of technological progress,  not at the end. The progression of AI has been  

extraordinarily rapid and it's hard to measure  directly. There's a lot of test scores and  

attempts to measure it, but I tend to think of it  kind of subjectively. When you think about GPT3,  

which was a earlier AI and it came out in uh 2021,  it wrote at about the level of a sixth grader,  

which was really amazing. I was like, "Wow, I  can't believe it could write this well." By the  

time GPT 3.5 came out, which was the version of  that came with chat GPT in the end of 2022, that  

wrote about as well as a sophomore in college, and  then GPT4, which came out in 2023, was as about  

as good a writer as a freshman PhD. That's a very  fast progression. And we know that GPT5 and other  

technologies in the timeline. You may already have  access to those by the time you watch this video.  

And this is a pretty fast progression. We don't  know how far it's going to go. We don't know how  

good these systems are going to get. We're not  in control of how fast these systems improve.

We are in control of how we decide to use  them and how we decide to apply them. As  

managers and leaders, you get to make these  choices about how to deploy these systems to  

increase human flourishing rather than necessarily  replacing people or watching them more closely.  

As individuals, we get to decide how to be the  human who uses these systems well rather than  

having the systems tell us what to do. And  we get to make a lot of decisions this way.

To watch the full class, become a  member at bigthink.com/membership.

重点词汇

词汇	含义
principle /ˈprɪnsəpəl/ B2	noun - 原则 (yuánzé)
ethically /ˈeθɪkli/ C1	adverb - 伦理上 (lúnlǐ shàng)
permission /pərˈmɪʃən/ B1	noun - 允许 (yǔnxǔ)
surprising /sərˈpraɪzɪŋ/ A2	adjective - 令人惊讶的 (lìngrén jīngyà de)
insight /ˈɪnsaɪt/ C1	noun - 洞察力 (dòngchálì)
disappointment /ˌdɪsəˈpɔɪntmənt/ B2	noun - 失望 (shīwàng)
potential /pəˈtenʃəl/ B2	adjective - 潜在的 (qiánzài de) noun - 潜力 (qiánlì)
competitor /kəmˈpetɪtər/ B1	noun - 竞争者 (jìngzhēngzhě)
empower /ɪmˈpaʊər/ C1	verb - 赋权 / 授权 (fùquán / shòoquán)
cardinal /ˈkɑːrdɪnəl/ C2	adjective - 首要的 / 基本的 (shǒuyào de / jīběn de)
persuade /pərˈsweɪd/ B2	verb - 说服 (shuōfú)
bias /ˈbaɪəs/ B2	noun - 偏见 (piānjiàn)
psychic /ˈsaɪkɪk/ C1	noun - 通灵者 (tōnglíngzhě)
extraordinarily /ɪkˈstrɔːrdənərəli/ B2	adverb - 非常 / 非凡地 (fēicháng / fēifándì)
subjectively /səbˈdʒektɪvli/ C1	adverb - 主观地 (zhǔguāndì)
sophomore /ˈsɑːfəmɔːr/ C1	noun - 大学二年级学生 (dàxué èr niánjí xuéshēng)
freshman /ˈfreʃmən/ B2	noun - 大学一年级学生 (dàxué yī niánjí xuéshēng)
deploy /dɪˈplɔɪ/ C1	verb - 部署 / 运用 (bùshǔ / yùnyòng)
flourishing /ˈflɜːrɪʃɪŋ/ C1	noun - 繁荣 / 兴旺 (fánróng / xīngwàng)