【AI趋势】吴恩达：AI最大的机会不在你以为的地方 · 打破

共 2 集

剧集目录

P1 · 【AI趋势】吴恩达：AI最大的机会不在你以为的地方 · 打破AI恐惧 p01 中英字幕 (P1)

【AI趋势】吴恩达：AI最大的机会不在你以为的地方 · 打破AI恐惧 p01 中英字幕 (P1)

发布于 2026-09-02 · 时长 37:51

字幕摘录

时间	英文	中文
0:00	There's been a lot of misinformation about AI.	对AI的误导很多.
0:03	This is Andrew.	这是安德鲁。
0:04	他共同创办了Google Brain和Coursera.
0:07	His machine learning course has reached millions of learners	他的机器学习课程 已经达到了数百万的学习者
0:09	and he is one of the most influential voices in AI today.	他是今天AI最有影响力的声音之一。
0:13	So a handful of leading AI companies	所以,少数主要的AI公司
0:14	have been very loud voices of fear-mongering around AI	在AI周围发出恐惧的声音
0:18	to try to get regulations passed.	试图让 条例通过。
0:20	This drumbeat of fear-based messaging	恐怖讯息的鼓声
0:22	has skewed societal's perception	扭曲了社会观念

展开字幕全文（947 条）

序号	英文	中文
1	There's been a lot of misinformation about AI.	对AI的误导很多.
2	This is Andrew.	这是安德鲁。
3	他共同创办了Google Brain和Coursera.
4	His machine learning course has reached millions of learners	他的机器学习课程 已经达到了数百万的学习者
5	and he is one of the most influential voices in AI today.	他是今天AI最有影响力的声音之一。
6	So a handful of leading AI companies	所以,少数主要的AI公司
7	have been very loud voices of fear-mongering around AI	在AI周围发出恐惧的声音
8	to try to get regulations passed.	试图让 条例通过。
9	This drumbeat of fear-based messaging	恐怖讯息的鼓声
10	has skewed societal's perception	扭曲了社会观念
11	to be really negative on AI.	对AI非常负面。
12	People talk to me about data centers and job loss.	人们和我谈论数据中心和失业问题。
13	Maybe AI could do 30, 40% of many jobs.	也许AI能做30,40%的工作。
14	And what that means is, well, that 60% that the human does	也就是说 人类60%的工作
15	has become even more valuable.	已经变得更加宝贵。
16	What about loss of human control over AI?	对AI失去人的控制怎么办?
17	I think about something else that we can't control.	我想到别的东西,我们不能控制。
18	I think you're one of the voices in AI	我觉得你是AI里的声音之一
19	who comes with a huge background	有着巨大背景的人
20	and machine learning and teaching AI,	以及机器学习和教学
21	and also you're a positive voice	而且你是一个积极的声音
22	because this is something that I've been seeing,	因为这是我见过的
23	especially this summer, how poor lives the society has become,	特别是今年夏天 社会的贫穷生活
24	especially on social media when I'm posting about AI	尤其是在社交媒体上 当我发布AI
25	and people talk to me about data centers and job loss.	人们和我谈论数据中心和失业
26	Why do you think this wave started recently?	你觉得这波为什么是最近才开始的?
27	What do you think the causes are?	你觉得原因是什么?
28	There's been a lot of misinformation about AI	关于AI有很多错误信息
29	and the root cause of a lot of this	以及很多事情的根源
30	is an unfortunate attempt that started two, three years ago	这是一个不幸的尝试 开始于二,三年前
31	of, I think, PR and regulatory capture.	我认为,公关和监管捕捉。
32	It turns out that one of the most valuable things	原来最有价值的东西之一
33	in AI right now is the giant AI models,	在AI现在 是巨大的AI模型,
34	giant large language models that some companies have trained.	巨型大型语言模型 一些公司已经训练。
35	But if you spend billions of dollars training a model,	但如果你花几十亿美元训练一个模特儿
36	it's really inconvenient if someone else trains a model	如果别人训练一个模特儿 实在太不方便了
37	and wants to give it to anyone in the world to use for free.	并想把它给世界上的任何人免费使用.
38	So a handful of leading AI companies, I think, as you know,	所以,一些主要的AI公司,我认为,你知道,
39	have been very loud voices of fear-mongering around AI	在AI周围发出恐惧的声音
40	to try to get regulations passed,	试图通过条例,
41	to create an unfair playing field that favors incumbents	创造一个有利于现任者的不公平竞争环境
42	so that we all have to pay a high toll for use of AI	因此我们都得付出高昂的代价来使用AI
43	while stymieing the other teams,	一方面阻击其他队伍,
44	be it researchers or other companies	不管是研究者还是其他公司
45	that want to just give away open way to open source models	想要给开源模型开辟道路
46	that anyone can use much cheaper.	任何人都可以使用便宜得多。
47	Unfortunately, fear-mongering works.	不幸的是,恐惧的侵扰是有效的。
48	When you go and say AI is like nuclear weapons,	当你去说AI就像核武器,
49	which is an analogy that has no basis in fact,	这是没有根据的比喻,
50	what do they have to even do with each other?	他们之间有什么关系?
51	Or when you go around and cherry pick cases of AI,	或者当你绕圈子去 和樱桃挑AI的案件,
52	making a misstep and make it much bigger than it is,	做一个错误的步, 使它比它大得多,
53	or even spread misinformation about how AI uses data centers,	甚至散布关于AI如何使用数据中心的错误信息,
54	uses a lot more water than the actual reality.	用的水比现实多得多
55	This drumbeat of fear-based messaging	恐怖讯息的鼓声
56	has skewed societal's perception	扭曲了社会观念
57	to be really negative on AI,	对AI非常负面,
58	which is unfortunate because this is slowing down	这太不幸了,因为这样会减速
59	American adoption in AI.	AI中的美国收养.
60	This is making America less competitive	这让美国竞争力降低
61	and unless we get the truth about AI out there,	除非我们知道AI的真相
62	which is that it's fantastic benefit with some problems,	这对解决一些问题有好处
63	but not nearly to the degree to which they're blown up to be,	但没有达到爆炸的程度
64	it will hurt individuals.	它会伤害个人。
65	I'm gonna read out some of the problems	我要宣读一些问题
66	that people are highlighting, job loss and inequality.	人们正在强调,失业和不平等。
67	What do you think?	你觉得呢?
68	The job apocalypse, the job pocalypse,	工作启示录, 工作浮夸,
69	this idea that AI will take over 50% of jobs,	这个想法,AI将占据50%的工作,
70	people will be out of work, rotting in the streets,	人们将失业,在街头腐烂,
71	that's just not gonna happen.	那只是不会发生。
72	With every wave of technology, including AI,	随着每一波技术,包括AI,
73	the skills we need to do great work shifts	换工作需要的技能
74	and so AI is changing job professions.	所以AI正在改变职业
75	But boy, I wish AI just doesn't work well enough.	但是,孩子,我希望AI 只是工作不够好。
76	I know that handful of businesses want to hype up AI	我知道有几家企业想哄骗AI
77	to say, we have super intelligence	可以说,我们有超级智能
78	or we'll have artificial intelligence or whatever	否则我们会有人工智能之类的
79	and can do all the stuff that humans do.	并且可以做人类做的一切
80	I wish AI work better, which is not good enough	我希望人工智能工作做得更好 这还不够好
81	to make AI do everything a human does.	让人工智能做人的一切
82	And if you look at the analysis of jobs,	如果你看看工作的分析,
83	economists, like my friend Eric Brinovson at Stanford,	经济学家,像我的朋友埃里克·布里诺夫森 在斯坦福,
84	Andy McAfee at MIT,	麻省理工学院的安迪·麦卡菲
85	economists have analyzed many people's jobs	经济学家分析了很多人的工作
86	and by breaking it down into individual tasks	并把它分成个人任务
87	and maybe AI could do 30, 40% of many jobs.	也许AI能做30%,40%的工作。
88	And what that means is, well, that 60% that a human does	也就是说,人类60%的工作
89	has become even more valuable	越来越珍贵
90	because it's called an economic compliment	因为这叫做经济赞美
91	to the 30, 40% is now cheaper.	30,40%现在更便宜。
92	And so what will happen is people that use AI,	接下来是使用人工智能的人
93	maybe people that use AI will replace people	也许用人工智能的人会代替别人
94	that don't use AI, but AI is not in a position	但是AI没有地位
95	for the vast majority of jobs to replace people.	给绝大多数的工作来代替人。
96	Of all the different professions,	在不同的职业中
97	the one that's most affected by AI now is software engineering	现在受AI影响最大的是软件工程
98	because AI is actually fantastic at writing code.	因为人工智能在写代码方面的确很棒
99	And what we see is that the number of job openings	我们所看到的是 职位空缺的数量
100	in software engineering is up,	在软件工程上,
101	contrary to what the doom fear mongers would say, right?	和末日的恐惧相反 对吧?
102	AI is not actually able to replace software engineers	AI实际上无法替换软件工程师
103	and all the good software engineers I know	还有我认识的所有优秀的软件工程师
104	are busier than ever.	比以往任何时候都更忙。
105	Now the flip side of it is,	现在它的反面是,
106	if someone's still write code like this 2022	如果有人还写这样的代码2022年
107	before trying GPT, they're in trouble.	在尝试GPT之前 他们就有麻烦了
108	They need new skills, like don't do stuff	他们需要新技能,就像不做东西一样
109	that 30, 40% of AI can automate.	30,40%的人工智能可以自动化.
110	You've got to start doing that, let AI do that,	你必须开始这样做, 让AI做到这一点,
111	but then gain your skills to do the other 60, 70%	然后获得技能 做其他60,70%
112	that AI cannot do.	这是大赦国际不能做的。
113	What would your advice be to new graduates?	你给新毕业生的建议是什么?
114	Because I talked to Eric on this podcast	因为我跟艾瑞克谈过这个播客
115	and he was talking about that there's not really	他当时在说真的没有
116	a lot of impact on the job market,	对就业市场影响很大,
117	except for, I think he mentioned people from 18 to 25	除了,我想他提到了18到25岁的人
118	who just graduated.	谁刚毕业。
119	What would be your advice to those people	你对这些人的建议是什么?
120	who don't have the expertise	谁没有专长
121	maybe to strategize in their job yet?	也许他们的工作有策略吗?
122	They can only do manual work that AI can do as well.	他们只能做人工工作,而AI也可以做.
123	So one real challenge for fresh college grads	所以,一个真正的挑战 对于新鲜的大学毕业
124	is that the university system is slow to adapt.	大学系统适应性很慢
125	And so, you know, I love academia.	所以,你知道,我喜欢学术。
126	I think we should all support academia and universities.	我认为我们都应该支持学术界和大学。
127	And when AI comes and transforms the way software is written,	当AI来改变软件的写法时,
128	universities often take like a year or two	大学通常要花一两年时间
129	for the faculty to master the skills,	以便教职员掌握技能,
130	then create new courses,	然后创造新的课程,
131	get curriculum committee approval or whatever,	得到课程委员会的批准或什么的,
132	get the faculty senate to vote.	让参议院投票
133	It just takes years.	只是需要几年时间
134	And that speed of change in academia	学术界的变革速度
135	is very poorly matched to the speed of change in AI.	与AI的改变速度非常差。
136	So sadly, many universities are still teaching students	可惜许多大学还在教学生
137	to be ready for the jobs of 2022	为2022年的工作做好准备
138	when we shouldn't even be teaching them	我们不应该教他们
139	for the jobs of 2026,	为2026年的工作,
140	we should be teaching them for the jobs of 2028 and beyond.	我们应该教他们2028年及以后的工作
141	And what this means is the job openings are there.	这说明职位空缺就在那里。
142	Tons of employers I know just can't find enough skill,	我知道很多雇主都找不到足够的技能
143	you know, people at any level of seniority.	你知道,任何级别的人。
144	But it turns out in my office right now,	但是现在在我的办公室里
145	we have a lot of interns.	我们有很多实习生
146	There are current college students, fresh college grad,	有现在的大学生, 新鲜的大学毕业,
147	we also have one high school intern,	我们还有一个高中实习生,
148	and they're amazing and productive.	和它们惊人和生产。
149	But the key is they're all very AI native.	但关键是他们都是AI本地人
150	They all use AI tools to do the things that I could do,	他们都用人工智能工具来做我能做的事,
151	but then also, you know,	还有,你知道,
152	lean in to doing the things that humans can do	做人类能做的事
153	that AI can't for a long time.	AI不能在很长一段时间。
154	So there's plenty of work for people to do.	故众生有余功德.
155	But so my advice to fresh college grads	但我建议大学毕业
156	or the people currently in college is,	或者目前大学里的人
157	by all means, work hard in classes, you know,	在课堂上努力工作
158	get good grades, learn from the instructors,	学好成绩 向老师学习
159	but to the extent that there are still additional skills	但只要还有更多的技能
160	that the university has not yet adapted to teaching,	大学还没有适应教学,
161	then find other ways to learn online,	然后找其他方法在网上学习,
162	无论是来自Coursera还是深层学习.ai还是Udemy
163	or other places where you can gain the more cutting edge skills,	或其他可以获得更先进技能的地方,
164	especially AI skills,	特别是人工智能,
165	that universities have not yet worked in the curricula.	大学还没有在课程中工作。
166	Quick pause, because what Andrew just said about workflows	暂停一下,因为安德鲁刚才说的工作流程
167	connects to something that HubSpot just put out for free.	连接到HubSpot刚刚免费推出的东西.
168	The thing is, the real opportunity right now	问题是,真正的机会,现在
169	is not just in the models themselves,	不只是模型本身,
170	but in what you build on top of them,	但你们在他们的上面所创造的,
171	how you turn AI from a chat window	从聊天窗口打开AI
172	into workflows that actually do work for you.	进入工作流程 真正为你工作。
173	And the good news is,	好消息是,我们
174	you don't have to start by building a full agent.	你不必开始 建立一个完整的代理。
175	You can start much simpler with better prompts.	你可以用更好的提示开始简单得多。
176	HubSpot刚刚发布了高级聊天GPT
177	Prompt Engineering Playbook, and the idea is very simple.	即时工程游戏本 这个想法很简单
178	In seven days, it helps you move	7天之内,它能帮助你移动
179	from getting generic AI answers to building prompts	从获得通用AI解答到构建提示
180	that give you much more consistent and useful options.	那样你就能更一致和更有用的选择了
181	Because we've all had this moment,	因为我们都有这一刻
182	you type in one vague sentence,	你输入一个模糊的句子,
183	get something mediocre back,	拿些平庸的东西回来
184	and then spend the next 20 minutes fixing it yourself.	然后再花20分钟自己修
185	This playbook is basically built for that exact problem,	这本游戏本基本上是为这个问题而建的
186	and it walks you through a single progression.	它让你经历了一个单一的进化。
187	First, you learn how to structure prompts	首先,你学会如何构建提示
188	with things like role, context, format, and parameters.	包括角色、上下文、格式和参数。
189	Then it gets more advanced, few-shot examples,	然后,它得到了更先进, 几个镜头的例子,
190	chain of thought reasoning,	思想推理链,
191	more precise prompt structures,	更精确的快速结构,
192	and ways to make outputs more reliable.	以及如何使产出更加可靠。
193	What I like is that it doesn't stop at individual prompts.	我喜欢的是它不会停止 个别的提示。
194	It also shows you how to build reusable AI personas,	也告诉你如何建造可重复使用的AI人物,
195	modular prompt components,	模块快速组件,
196	and eventually your own signature prompt engineering system.	最后你自己的签名即时工程系统。
197	So instead of starting from scratch every time,	所以不是每次从零开始
198	you're building a repeatable way to work with AI,	你正在建立一个可以重复的方式 与AI合作,
199	whether you're creating content, analyzing data,	你是否正在创建内容 分析数据
200	or making business decisions.	或做出商业决定。
201	The playbook is free.	游戏本是免费的.
202	Links in the description.	描述中的链接.
203	Thanks to HubSpot for sponsoring this video.	感谢HubSpot赞助了这段视频.
204	And now back to our conversation with Andrew.	现在回到我们和安德鲁的谈话
205	I want to say one other thing.	我想说另外一件事。
206	It turns out, if you look at the skill map changes,	事实证明,如果你看 技能地图的变化,
207	one of the most important changes is	最重要的变化之一是
208	it's so much easier to build with AI than before.	和AI一起建起来比以前容易多了
209	When something becomes much easier,	当事情变得容易得多,
210	a lot more people should do it.	更多的人应该这样做。
211	And so now, not only should professional software engineers	所以现在,不仅专业的软件工程师
212	build software with AI,	与AI一起构建软件,
213	it's becoming much easier for everyone to build with AI,	对AI来说,每个人都更容易建起来
214	and people that embrace that and do so	和接受并接受的人
215	will be more productive and will accomplish more,	将会更加丰硕,并且更加成功,
216	and I think have more fun than the ones that don't.	我觉得比那些不开心的人更好玩
217	And AI lets you build really fast.	人工智能让你建得很快
218	So for people that are not just software engineers,	所以对于不仅仅是软件工程师的人来说
219	but marketers, recruiters, HR professionals,	但市场营销者、招聘者、人力资源专业人员,
220	operations specialists,	业务专家,
221	I think if they learn to build with AI,	我想如果他们学会和AI一起建设
222	they'll really just do much more, whatever the job role is.	他们真的会做的更多, 不管工作的角色是什么。
223	How do you, by the way, measure the increase in productivity	顺便说一句,你怎么衡量生产率的提高?
224	when you deploy AI?	当你部署AI时?
225	Do you have a KPI in your company?	你的公司有KPI吗?
226	I wish it was a simple answer.	我希望这是一个简单的答案。
227	I find that the business outcome of AI	我发现AI的商业结果
228	is more a function of the business	更属于企业的功能
229	than a function of the AI.	而不是大赦国际的职能。
230	For some, it may be increased customer growth and retention,	对一些人来说,这可能会增加客户的增长和保留,
231	or maybe faster to serve customers,	或者更快地为顾客服务,
232	or it may be increased accuracy in some tasks.	也可能提高某些任务的准确性。
233	So the KPI tends to be related to the business	因此,KPI倾向于与业务有关
234	rather than the AI.	而不是大赦国际。
235	So you can't directly measure just AI,	所以你不能直接测量AI
236	because it's an interesting thing to do,	因为它是一个有趣的事情做,
237	because we've been deploying AI actively in my company,	因为我们一直在积极部署AI在我的公司,
238	and I think for me as a media company,	我认为作为一个媒体公司,
239	it's probably the amount of views, the output.	可能是浏览量 输出量
240	It's just interesting how different people measure,	有趣的是不同的人如何衡量,
241	even revenue, like if you're becoming more effective	甚至收入,就像如果你越来越有效
242	with how you make money.	你如何赚钱。
243	Actually, say more.	其实,说更多。
244	How are you using AI in your business?	你怎么在你的生意中使用AI?
245	Oh my God.	天呐
246	First of all, we have Claude for all of us,	首先,我们有 克劳德为我们所有人,
247	and we have certain projects	我们有一些项目
248	for every social media that we're on.	我们所有的社交媒体。
249	So for example, for this podcast,	比如说,对于这个播客,
250	we have a project that's called Guests,	我们有个叫"客人"的项目
251	and it knows all the analytics from previous guests,	它知道所有分析 从前宾客,
252	and it has certain criteria	而且它有某些标准
253	on which we rank every single person	我们把每个人排在其中
254	who comes to the podcast,	他来见播客,
255	whether he or she is cited,	无论他或她被引用,
256	whether they have a certain opinion on AI,	他们是否对大赦国际持有某种意见,
257	whether they've been active with AI in their company,	他们是否一直活跃 在AI在他们的公司,
258	or if they're a recent founder in AI.	或者他们是AI的最近创始人
259	So it gives them different weights,	因此,它给他们不同的重量,
260	and it comes up with a grade based out of 40.	并得出一个等级 根据40。
261	40 meaning tier one, 30 meaning tier three, et cetera.	40表示一级,30表示三级等.
262	And then we have another one	礛и临Τ
263	that analyzes every single podcast,	分析每个播客,
264	and gives me tips on how to ask questions.	并给我如何提问的提示。
265	与Instagram相同,与LinkedIn相同.
266	It has my tone of voice, personal dossier,	它有我的声音,个人档案,
267	my business strategy.	我的业务策略
268	So whenever it writes something,	所以,每当它写的东西,
269	it knows all the facts about me, how I sound.	它知道关于我的所有事实 我的声音。
270	Every social media is run by a person.	每个社交媒体都由一个人管理.
271	So a person makes a strategic call.	所以一个人做出一个战略号召。
272	And by the way,	顺便说一句
273	if you can give me feedback on this, if I can improve.	如果你能给我反馈, 如果我能改进。
274	So what I'm working on right now is closing the loop,	所以,我现在的工作是 关闭循环,
275	because sometimes they send me a text,	因为有时候他们会给我发短信
276	and I'm like, oh, we need to change this, this, and that.	而我喜欢,哦,我们需要改变这个,这个,那个。
277	But that happens in a chat, in Telegram.	但是,这发生在聊天, 在Telegram。
278	And we have this feedback.	我们得到了这个反馈。
279	We have a bot that scans all of our chats.	我们有一个机器人 扫描我们所有的聊天。
280	But I really want AI to be able to learn continuously	但我真的希望AI能够不断学习
281	from this feedback to just know my taste better.	从这些反馈来了解我的品味
282	There's one thing I see a lot in AI,	我在AI里看到很多东西
283	which is it turns out for AI's data scientists,	对AI的数据科学家来说
284	or AI's brainstorming partner,	或AI的集思广益伙伴,
285	it often comes up with one or two good ideas,	它经常提出一个或两个好主意,
286	two or three mediocre ones, and four atrocious ones.	两三个平庸的,四个残暴的。
287	And sometimes you wonder,	有时候你会想
288	how could my AI have thought	我的人工智能怎么会想到
289	that could even be a plausible idea?	这甚至可能是一个可信的想法?
290	And to me, this relates to the job-pocalypse point of view,	对我来说,这涉及到 工作启示论的观点,
291	which is that for a long time, humans, you, me,	人类,你,我,
292	everyone watching this,	大家看这个
293	will have a significant context advantage over AI,	与大赦国际相比,在环境方面具有显著优势,
294	which is that you know something	就是你知道一些东西
295	that's incredibly obvious to you	这对你来说是显而易见的
296	that that was an awful idea, but the AI did not.	那是个糟糕的主意 但人工智能没有
297	And it turns out that one of the reasons	结果发现一个原因
298	why AI will not replace our jobs or whatever,	为什么AI不会取代我们的工作什么的,
299	large percentage not anytime soon,	比例很高,没有很快
300	is because humans have a massive context advantage	因为人类拥有巨大的环境优势
301	compared to AI.	与大赦国际相比。
302	We know so much that from our years of experience	我们从多年的经验中知道这么多
303	that we talked to customer,	我们和顾客谈过
304	we saw the funny facial expression that told us,	我们看到了有趣的面部表情 告诉我们,
305	ah, they don't like this.	啊,他们不喜欢这样。
306	Or we talked to business or our manager said,	或者我们跟生意谈了 或者我们的经理说
307	hey, blah, blah, blah, I really care about this.	嘿,嘿,嘿,我真的关心这个。
308	And so it turns out that almost all humans,	结果发现几乎所有的人
309	well, maybe all humans, just know a lot of the stuff	好吧,也许所有的人类, 只是知道很多的东西
310	that the plumbing does not exist,	使管道不存在,
311	and they don't think it exists for the foreseeable future.	他们认为它不存在 在可预见的未来。
312	And so that's why AI is so important for AI to get.	所以AI才这么重要
313	I know sometimes people talk about	我知道有时候人们会谈论
314	the importance of human judgment or human taste.	人类判断或品味的重要性。
315	And some people wonder, all right, what is taste?	有些人在想,什么是味道?
316	Is this fuzzy thing?	这是模糊的东西吗?
317	But to me, the technical thing that underlies	但对我来说, 技术问题的基础
318	why humans have better judgment and better taste than AI	为何人类比AI更有判断力和品味
319	is this context advantage.	这是上下文的优势。
320	And because this is a long-term advantage,	因为这是一个长期优势,
321	like no one's gonna solve this, you know, in a few years,	就像几年后 没人能解决这件事
322	this is why we just need a lot more humans	所以我们需要更多的人
323	with that judgment and taste to keep on complementing AI.	以判断和品味 继续补充AI。
324	And doesn't this make education even more important?	这难道不使教育更加重要吗?
325	Because education gives us context.	因为教育给我们提供了背景。
326	This is another thing I'm hearing about AI,	这是另一件事 我听到关于AI,
327	like you won't need education	就像你不需要教育一样
328	because all the information is at your fingertips.	因为所有信息都在你的指尖上
329	You just ask Chad GPT.	你只是问查德GPT。
330	But when you say context and taste,	但当你说上下文和品味时
331	for me, that's years of acquiring knowledge	对我来说,这是多年 获得知识
332	and learning from the best and seeing how they perform	向最好的人学习,看看他们的表现
333	versus just asking a chat.	而不是只是问话
334	I'm gonna say something that may be controversial.	我要说一些有争议的话
335	I don't know if I've said this publicly,	我不知道我是否公开说过
336	but I think it's true, which is frankly,	但我认为这是真的 老实说
337	AI models are terrible for learning.	AI模型对于学习来说是可怕的.
338	I know people think AI is wonderful at getting things done.	我知道人们觉得AI很擅长做事情
339	Use it all the time, love it.	常用之,爱之.
340	All the data that's coming out is that	所有的数据出来是
341	when, say, college students use AI, we know this.	当,比如,大学生使用AI, 我们知道这一点。
342	It's just the studies now back up as well,	只是现在的学习也有所恢复
343	so we also have numbers, but the data is very clear.	因此,我们也有数字, 但数据非常清楚。
344	Students go higher on homeworks when they use AI.	学生使用人工智能时,功课会更高.
345	Yay, higher homework schools.	耶,高等家庭作业学校。
346	But retention, their long-term performance is much worse	但保留,他们的长期表现更差
347	because their AI do the work for them.	因为他们的人工智能为他们工作。
348	More and more studies are coming out to back this up now	越来越多的研究 出来支持现在
349	that I think people think, oh, it turns out,	我认为人们会认为, 哦,结果,
350	I think Wikipedia is a wonderful tool.	我认为维基百科是一个很棒的工具。
351	It has tons of facts.	它有很多事实。
352	Web search is a wonderful tool.	网络搜索是一个很棒的工具。
353	It has tons of facts.	它有很多事实。
354	But it turns out that when you ask AI to do work for you,	但事实证明 当你要求AI为你工作时
355	you are cognitive offloading to AI,	你将认知卸载到AI,
356	which is great because that's how society moves forward	这是伟大的,因为 社会如何前进
357	when it gets work done.	当它完成。
358	But human retention is much worse.	但人类保留的情况更糟。
359	It's just so clear that LLMs, as they are most commonly used,	很明显,LLMS, 因为最常用的是,
360	are terrible for learning.	学习是可怕的。
361	I'm not saying there's no way to use it	我不是说没有办法用它
362	in a way that is good for learning.	以有利于学习的方式。
363	I think there are ways to use it that are good for learning.	我认为有方法使用它,对学习有好处.
364	But even for myself, there's so many things	但即使是我自己, 有很多事情
365	that are on the AI model over the last six months	过去6个月AI模式中
366	or whatever, like building some project,	或者什么的,像建设一些项目,
367	how does this front-end backend component work, whatever.	这个前端后端组件是如何工作的,什么的。
368	Give me the answer, get the job done.	给我答案 完成任务
369	It was fantastic.	简直太棒了
370	But six months later, I don't remember the answer	但半年后 我不记得答案了
371	when I need to redo that front-end backend component,	当我需要重做前端后端组件时,
372	I ask the AI again.	我又问AI。
373	So data is really clear.	所以数据非常清楚。
374	We should stop thinking of AI as helpful for learning.	我们不应该把人工智能视为有助于学习。
375	At least the vast majority of ways	至少绝大多数的方法
376	that the vast majority of people are using AI models today	绝大多数人今天都在使用AI模型
377	is absolutely terrible for learning.	这对学习是绝对可怕的。
378	But you're building a company helping solve that, right?	但你正在建立一个公司 帮助解决这个问题,对不对?
379	Because the one-to-one tutoring with AI	因为跟AI的一对一辅导
380	is that where you just announced	是你刚刚宣布的
381	和Coursera的一亿投资
382	Yes, so I'm excited about leading a new organization	是的,所以我很高兴领导一个新的组织
383	called Learn Vector that is focused	名为“ 学习矢量” 的焦点
384	on building new learning experiences	积累新的学习经验
385	that is much more one-to-one than one-to-many.	这比一对一多得多
386	So 15 years ago, I was privileged to participate	所以15年前,我有幸参加
387	in the online courses movement	在线课程运动
388	that I think changed the way a lot of people learn.	我认为改变了很多人的学习方式
389	But that was, and still remains,	但这曾经是,现在仍然是,
390	largely a one-to-many experience	主要是一对一的经验
391	where everyone kind of watches the same video,	所有人都看同样的视频
392	which is actually okay, it actually works well.	这其实是好的, 它实际上工作很好。
393	But the technology now exists	但技术已经存在
394	to create much more personalized,	创造更个性化,
395	customized one-to-one experiences.	定制的一对一体验.
396	And so our team is working hard on that.	因此,我们的团队正努力进行这项工作。
397	I think we'll have a lot more to show by early next year.	我想到明年初我们还有很多事情要做
398	When think about human skill development,	想想人类技能的发展
399	I feel like because AI has so heavily impacted	我觉得因为人工智能影响很大
400	software engineering, what we see happening	软件工程,我们看到的
401	in the job market for software engineering	在软件工程的就业市场上
402	is a harbinger or as a forerunner	是预兆还是预兆?
403	of what we'll see in other disciplines as well.	我们在其他学科也会看到的东西
404	And in software engineering,	在软件工程方面
405	people need to learn new skills.	人们需要学习新的技能。
406	But when they do, they are thriving	但是,当他们这样做, 他们蓬勃发展
407	and creating more value and frankly getting raises	创造更多价值 坦率地说 获得高薪
408	and doing even more exciting projects.	和做更令人兴奋的项目。
409	And what I'm seeing the early signs of	我所看到的早期迹象
410	in other disciplines as well,	其他学科,
411	for example, in software engineering,	例如,在软件工程方面,
412	most developers, like front-end, back-end developers	大多数开发者,如前端,后端开发者
413	have now become full-stack developers	现已成为全存储开发者
414	because of AI help, you could take on broader scope.	因为AI的帮助, 你可以承担更大的范围。
415	I'm seeing early signs of this in other disciplines as well	我在其他学科也发现了早期的迹象
416	where, for example, someone in marketing	例如,市场营销中的人
417	that did marketing coordination,	做营销协调,
418	coordinating marketing campaigns with AI help,	在大赦国际的帮助下协调营销活动,
419	they can now become more of a full-cycle marketer	他们现在更能成为全周期的营销者
420	and take on a broader scope.	并且扩大范围。
421	And I'm seeing, frankly, sources in recruiting	坦率地说,我看到 招募的来源
422	become more full-cycle to end-to-end recruiting.	成为更全周期的 端到端的招募。
423	So now the good news and bad news is	所以现在好消息和坏消息是
424	for people to step up to these broader roles,	让人民承担起这些更广泛的角色,
425	you do need to learn AI skills,	你需要学习人工智能技能,
426	but also it's not just learning AI.	不只是学习人工智能
427	You also need to learn these other skills,	你还需要学习这些其他技能,
428	like how do you do the other parts of marketing	就像你如何做营销的其他部分
429	of recruiting or software engineering or AI engineering.	招聘、软件工程或AI工程。
430	So I think this actually creates a heavy need,	所以我认为这实际上造成了一个沉重的需要,
431	a big need for people to gain new skills.	需要人们获得新的技能
432	But when they do, which is both AI skills	但是当他们这样做, 这既是AI的技能
433	but also disciplinary skills,	同时也有纪律技能,
434	then they can do much more, hopefully have more fun,	然后,他们可以做更多的, 希望有更多的乐趣,
435	work on more exciting projects,	在更刺激的项目上工作,
436	hopefully get paid more as well.	希望也能得到更多的报酬。
437	And one reason I kind of worry about the fear monitoring	还有一个原因 我担心恐惧的监控
438	is I got an email from someone	我收到别人的邮件
439	that was about to enter college.	即将进入大学。
440	And he emailed me saying,	他给我发邮件说,
441	hey, Andrew, taking online courses,	嘿,安德鲁,上在线课程
442	but I'm really struggling with	但我真的很挣扎
443	what I should major in college	我应该在大学里主修什么
444	because in four years, won't AI do all this	因为四年后,不会让AI做这一切
445	and everything I learn will be obsolete?	我学到的一切都会过时?
446	And the answer is no, of course it won't all be obsolete.	答案是否定的,当然不会全都过时。
447	But when we keep on pushing these fear messages,	但是,当我们继续推进这些恐惧信息,
448	we make people wonder if they will even be relevant	我们让大家怀疑他们是否相关
449	and it makes people not lean in to gain these skills.	这使得人们不能靠自己获得这些技能。
450	They'll put them in a much better position.	他们会让他们处于更好的位置
451	So I see very clearly that these fear monitoring messages	所以我非常清楚地看到 这些恐惧监测信息
452	are distorting how many people,	扭曲了多少人,
453	including high school students, college students,	包括高中生、大学生、
454	fresh grads think about the economy.	刚毕业的人会考虑经济
455	And frankly, making people give up	老实说,让人放弃
456	is one of the worst things we've been doing	是我们做过的最糟糕的事之一
457	in this era when people that lean in will thrive.	在这个时代, 人倾斜将繁荣。
458	Andrew has taught over 8 million people AI.	安德鲁教过超过800万人AI.
459	He started teaching machine learning online back in 2011,	他从2011年开始在网上学习机器,
460	years before the current AI boom.	距当前AI繁荣还有好几年
461	Now one of the companies he's building	现在他正在建设的公司之一
462	is focused on AI agents.	重点是人工智能特工
463	From the way that it sounds,	从听起来,
464	it can still feel way too technical.	感觉还是太技术了
465	So I put together a step-by-step guide	所以我编了一部"一步一步"的指南
466	to building your first AI agent with no coding required.	来建立第一个不需要编码的AI代理.
467	It walks you through what to automate,	它让你通过什么 自动化,
468	how to set it up and how to make it actually useful.	如何设置,如何使其实际有用。
469	It is in my newsletter this week.	我这周的通讯里有这个
470	The newsletter is called Future Proof.	该通讯名为"未来证明".
471	It's free.	自由无碍.
472	Link is in the description.	链接在描述中.
473	What would you reply back to that email	你对那封邮件有什么反应?
474	that somebody sent you?	有人派你来的?
475	What would you say is the best major to study now,	你会说什么是最好的主 现在学习,
476	to thrive in AI era?	在AI时代蓬勃发展?
477	Do you think it's like going deep into a niche	你觉得这就像深入到一个小地方
478	or just broader computer science	或者只是更广泛的计算机科学
479	so that you can acquire AI skills really fast?	这样你才能很快获得人工智能技能?
480	You know, I don't know what's the best major.	你知道,我不知道什么是最好的主修课。
481	There are awful lot of great majors.	有很多伟大的专业。
482	It's like, I kind of feel like	我觉得...
483	what's the best job in the world?	什么是世界上最好的工作吗?
484	It's like, what's the best major in the world?	这就像, 什么是世界上最好的主修?
485	Something that you love, right?	你爱的东西,对不对?
486	Yeah, my daughter wants to be an astronaut.	对,我女儿想当宇航员
487	I don't know if she can major in becoming an astronaut.	我不知道她能否成为宇航员
488	I have to think about that.	我得考虑一下
489	When she gets older, she may change her mind.	她长大后可能会改变主意
490	I see so many opportunities across so many job roles.	我看到很多机会 贯穿这么多的工作角色。
491	It all seems very exciting to me.	这一切似乎都让我兴奋。
492	But do learn AI, do learn to build with AI.	但要学AI,要学与AI一起建设.
493	The other thing that my team's been working	还有一件事,我的团队一直在工作
494	on an AI engineering skills map to try to map out	在AI的工程技能地图上尝试映射
495	the most important skills for AI engineering.	AI工程最重要的技能.
496	One thing that I felt intuitively,	有件事我直觉地觉得
497	but I was surprised to see it show up in the data	但我惊讶地看到它出现在数据中
498	was that a lot more job descriptions seem to be saying	更多的职务说明似乎在说
499	they want people that demonstrate	他们希望人们展示
500	a very high sense of agency.	一种很高的代理意识。
501	Because it turns out with AI,	因为有AI的发现
502	there are a lot more opportunities	还有更多的机会
503	for individuals to spot problems	让个人发现问题
504	and go build something or do something to go solve it.	并去做一些事情或做一些事情去解决它。
505	So I think we're really evolving.	所以,我认为我们真的在进化。
506	Well, we've long been evolving,	我们早就在进化了
507	but we're accelerating past the era	但我们正在加速穿越时代
508	where people sit around and wait for their boss	在那里人们坐着等待他们的老板
509	to tell them what to do.	告诉他们该怎么做。
510	This is what I've been feeling a lot,	这就是我一直以来的感受,
511	especially when we started doing remote work.	特别是我们开始做远程工作的时候
512	I want people to be entrepreneurs within their niche.	我希望人们在自己的专长内成为企业家.
513	Like if you're helping me with LinkedIn,	就像你帮我处理LinkedIn一样
514	you're an entrepreneur there.	你是一个企业家在那里。
515	You can hire more contractors.	你可以聘请更多的承包商。
516	You can deploy different tools.	你可以部署不同的工具。
517	You make the strategic decision	你来决定战略
518	whether this topic is good or not.	不论这个议题是否良好。
519	Shall we proceed with it?	我们继续吧?
520	I really think, and tell me if you agree with me,	我真的觉得 如果你同意的话 告诉我
521	we're moving into that job market	我们要进入这个职业市场
522	where everyone is kind of independent in their workplace.	在那里,每个人都在他们的工作场所是独立的。
523	I think people will have much more autonomy and creativity.	我认为人们会有更多的自主权和创造力.
524	So I agree with that.	所以我同意这一点。
525	And I'd even go one step further,	我甚至更进一步,
526	which is I talk to a lot of people,	我和很多人谈过
527	engineers and others in large companies	大公司的工程师和其他人
528	that tell me that the manager	那告诉我经理
529	tells them to stay in their swim lane.	叫他们待在游泳道上
530	They'll say, oh, I have this creative idea,	他们会说,哦,我有一个创意的想法,
531	but the manager says, no,	但经理说,不,
532	I need you to focus on this one thing, frankly,	坦白地说,我需要你集中注意这个
533	often because the manager's career depends on it.	经常是因为经理的职业生涯取决于此.
534	But I feel like the number of opportunities	但我觉得机会之多
535	for people to spot things outside their swim lane	在游泳道外发现东西
536	and then in a responsible way explore how to get it done,	然后以负责任的方式探索如何完成它,
537	that feels very exciting to me.	感觉很刺激
538	And I think that in the future,	我认为在未来,
539	the businesses that serve a culture	为文化服务的企业
540	that encourage people to learn AI built fast, responsibly,	鼓励人们快速、负责任地学习AI,
541	talk to customers would drive a lot more value	和顾客聊天会吸引更多的价值
542	than the more hierarchical silo organizations.	而不是更分级的组织。
543	Yeah, it starts with hiring the right people	是啊,它从雇用合适的人开始
544	and then nurturing this in your organization.	然后在你的组织里培育这个
545	When you say learn how to use AI	当你说学习如何使用AI
546	and become proficient with AI,	并且精通AI,
547	can you give me some benchmarks,	你能给我一些基准,
548	like of a person who's like say a marketer,	就像说市场营销员一样
549	knowledge worker, advanced with AI,	知识工作者,高级AI,
550	what are you looking for	你们在找什么?
551	when you're interviewing this person?	当你采访这个人时?
552	I'm pretty sure my team's ahead of the curve.	我敢肯定,我的团队 领先曲线。
553	All of my marketers know how to code.	我所有的推销员都知道如何编码
554	So as part of how I interview marketers,	因此,作为我如何采访营销员的一部分,
555	we ask them what they've built	我们问他们建造了什么
556	and if they have not built any software.	如果他们没有建造任何软件,
557	If it's a dashboard, is it good or bad?	如果是仪表板,是好是坏?
558	Like, is it too basic or?	像,这是太基本还是?
559	A dashboard, again, my team's pretty,	仪表板,再次, 我的团队很漂亮,
560	somewhat ahead of the curve,	有点在曲线前,
561	but all of my-	但我的所有的... ... . .
562	Mine was good to hear like that.	我很高兴听到这样的声音。
563	All of my marketers have built	我所有的推销员都造了
564	much more sophisticated things than dashboards.	比仪表板更精密的东西
565	Likewise, to me.	同样,对我来说。
566	Oh, I feel like, I don't know.	哦,我觉得,我不知道。
567	The other day, one of our,	前几天,我们的一个
568	someone on the marketing team	有人在营销团队
569	was talking about the tools that he had built.	他正在谈论他建造的工具。
570	So when he's considering writing an article on something,	所以当他考虑写一篇关于某事的文章时
571	it will crawl the web, find related work,	它会爬网 找到相关的工作
572	has a custom desktop app,	有自定义桌面应用程序,
573	actually build a desktop app that runs on his Mac,	实际建造一个运行在Mac上的桌面应用程序,
574	to highlight related articles for him.	为他突出相关文章.
575	Then he can chat to the whole system,	然后他可以和整个系统聊天
576	navigate, you know, the thing he's writing,	导航,你知道, 东西他写,
577	as well as the related work.	以及相关的工作。
578	And he had a large dashboard for crawling the internet	他有一个大仪表板 爬上网
579	to highlight to him exciting things that are popping up.	告诉他一些激动人心的事
580	Now, even on my team,	现在,即使在我的团队,
581	I think that market is ahead of the curve.	我认为市场领先于曲线。
582	But that's great to hear.	但这是伟大的听到。
583	Any other interesting use cases	任何其他有趣的使用案例
584	that will inspire people to build something similar?	那会激励人们 建立类似的东西?
585	Let's see, maybe my finance team uses AI extensively.	让我们看看,也许我的金融团队 广泛使用AI。
586	So I think my, one of my CFOs realized that,	所以我想,我的一个CFO意识到,
587	you know, her team was spending hours every week	你知道,她的团队 每周花几个小时
588	clicking through documents, open this,	点击文档,打开此选项,
589	copy, paste this number here.	收到 把这个号码贴在这里
590	And so she started building automation strips	于是她开始建造自动化带
591	that runs on a routine that automatically opens files,	运行在自动打开文件的常规上,
592	checks what's in there, checks for consistency,	检查里面有什么, 检查一致性,
593	highlights to her team if there's something,	如果有什么事情的话,向她的团队突出
594	if there's something they need to be paying attention to	如果有什么需要他们注意的
595	of a new document that showed up.	显示的新文件。
596	So I find that rather than waiting around	所以,我发现这 宁可等待周围
597	for an engineer to do the work for them,	一个工程师为他们工作,
598	the team's ability to kind of,	团队的能力 有点,
599	not just build dashboards,	不只是建造仪表板,
600	but build kind of a data management infrastructures.	但建立某种数据管理基础设施。
601	They can ingest data, alert them if something's happening.	他们可以摄入数据, 提醒他们如果有事发生。
602	I think my finance and marketing teams are doing that.	我认为我的金融和营销团队正在这样做。
603	Oh, my recruiting team.	哦,我的招募团队。
604	Well, we actually have recruiting engineers,	我们实际上已经招聘了工程师
605	which are really professional engineers	他们是真正的专业工程师
606	that sit in a recruiting team.	在一个招募团队。
607	They're the building very sophisticated tools for recruiting.	他们是大楼非常精密的招募工具。
608	And this is actually the other trend.	这是另一个趋势。
609	I think marketers, recruiters, HR professionals,	我认为市场营销者,招聘者,人力资源专业人士,
610	ops people should all learn AI.	行动人员应该学会人工智能
611	But the other thing is when you take an engineer	但另外一件事是当你接受工程师
612	and embed them in these teams,	把他们安置在这些队伍中,
613	then that further accelerates what you can do.	然后,这进一步加快了你能做的。
614	We do the same.	我们也是
615	We start with something basic, build it ourselves.	我们从基本的东西开始, 建造它自己。
616	Then we hit the wall and engineer comes in.	然后我们撞上了墙 工程师进来了
617	We build it further.	我们进一步建造它。
618	Frankly, when you look at not just software engineers,	老实说,当你看到 不只是软件工程师,
619	but recruiting engineers, marketing engineers,	但招聘工程师,营销工程师,
620	HR engineers, I think there's so much valuable	HR工程师,我觉得有很多有价值的东西
621	engineering work that can now be done.	现在可以做的工程工作。
622	I'm just, you know, not worried about running out of,	我只是,你知道, 不担心跑出去,
623	frankly, all my friends were so busy.	坦白说,我的朋友都很忙
624	We think, boy, how could we run out of engineering jobs?	我们想,孩子,我们怎么能用完工程工作?
625	Yeah, there are so many cool ideas you can experiment on.	是啊,有很多很酷的想法 你可以进行实验。
626	But you touched upon something	但你碰了个东西
627	that is actually one of the fears.	这其实就是其中之一的恐惧。
628	When we talk about like financial information,	当我们谈论像财务信息,
629	how much you're giving to AI.	你要给AI多少钱?
630	So I gave my perplexity permission	所以我才允许我感到困惑
631	to scan my fidelity account so it can track my portfolio,	扫描我的忠诚账户 这样它就可以追踪我的投资组合
632	tell me when to rebalance.	告诉我什么时候再平衡
633	It doesn't do anything on my behalf, but it has access.	它不代表我做任何事,但它有权限。
634	Do you think there is any problem with that?	你觉得有什么问题吗?
635	This is complicated.	这很复杂
636	I think AI and privacy is a complex area.	我认为AI和隐私是一个复杂的领域.
637	And it depends a lot on the company	这要靠公司
638	that you are sharing your data with.	您正在分享您的数据。
639	So for example, I trust all the hyperscalers	比如说,我信任所有的超标器
640	to really 100%, you know, follow the terms of service	百分之百遵守服务规定
641	and to do what they say.	并按他们说的做。
642	My personal opinion, not giving legal business advice,	我个人的意见,不提供 法律业务建议,
643	but I'd be shocked if the largest hyperscalers	但是,如果最大的超标,我会震惊
644	publish the terms of service with some privacy notice	发布一些隐私通知的服务条款
645	and if they reach that, because that'd be not the culture,	如果他们达到这一点, 因为这不是文化,
646	it'd be so damaging to the long-term business model.	这会对长期商业模式造成如此大的伤害。
647	Now that's on the largest hyperscaler side.	现在,这是最大的超规模。
648	If you look at the AI company side,	如果你看看AI公司那边
649	there's been at least one company that I won't name	至少有一家公司我不点名
650	that seems to occasionally change the terms of service.	这似乎偶尔会改变服务条件。
651	And if you're using it, you go to the website,	如果你用它,你去网站,
652	so you pop up, hey, we changed the terms of service	所以,你出现了,嘿, 我们改变了服务条件
653	to retain your data or train your data.	保留数据或训练数据。
654	And if you aren't paying attention	如果你不注意的话
655	and click the wrong button,	点击错误的按钮,
656	then they suddenly gave themselves permission	然后,他们忽然答应了。
657	to access your data in a way	以某种方式访问您的数据
658	that I'm not that comfortable with.	说我不太舒服
659	I feel like I handle some sensitive information.	我觉得我处理一些敏感的信息。
660	So then I tend to be very careful with the businesses	所以,我总是非常小心 与企业
661	that I just don't feel the culture and the DNA	我只是感觉不到文化和DNA
662	and frankly, the long-term business model	坦率地说,长期商业模式
663	is as tied to protecting individual user privacy	与保护个人用户隐私一样
664	than the hyperscalers.	比超标者还要大
665	And I see businesses get this as well.	我认为企业也得到了这个。
666	例如,我的一个团队,AI Aspire,
667	we work with very large corporations,	我们和大公司合作
668	including banks with incredibly sensitive financial data.	包括拥有极其敏感金融数据的银行。
669	And as you can imagine, AI Aspire and our clients	你可以想象,AI Aspire和我们的客户
670	do not willingly share really sensitive,	不愿分享真正的敏感,
671	often material non-public information, NMPI,	经常是非公开资料,NMPI,
672	with frontier labs without really careful thinking	与前沿实验室 没有真正仔细的思考
673	about the guardrails in privacy.	私底下的护栏
674	So I think it's complicated.	所以我觉得这很复杂。
675	So trusting hyperscalers,	如此信任超标者,
676	but also another thing that you can do,	但还有一件事你可以做,
677	you can download an open source model	您可以下载开源模式
678	and just run it on your computer	在你的电脑上运行
679	and then it just stays on your computer, right?	然后它只是留在你的电脑上,对不对?
680	Yes, I think, yes, it turns out a lot of banks	是的,我想,是的,结果很多银行
681	will actually run the things in a virtual private cloud	将在虚拟的私人云中运行
682	or on-prem, so it never even leaves their control.	或Prem,所以它甚至从来没有离开他们的控制。
683	But I think for individuals,	但我认为对于个人来说,
684	for the very sensitive things,	对于非常敏感的事情,
685	I sometimes run a local model	有时我经营本地模特儿
686	and it's been interesting with the open-way models.	和它的有趣 与开放模式。
687	So the latest open-way models	因此,最新的开放模式
688	are approaching frontier capability	正在接近边境能力
689	and that are actually small enough.	这其实足够小。
690	They're actually really good models now	他们其实是真正的好模特
691	that you can run on them.	你可以运行在他们身上。
692	Yeah, the one from Meta, right?	是啊,梅塔的那个,对吧?
693	The recent one.	最近那个
694	Oh yes, Metamuse is a good model	哦,是的,梅塔缪斯是个好模特
695	and I'm thinking also the latest version of Quen	我也在考虑最新版的Quen
696	is also very good.	也很不错。
697	But I think frankly, these models change every other week.	但坦率地说,这些模型每隔一周就改变一次。
698	So I think the best practice is to not get stuck on one,	所以我认为最好的做法 就是不要被困在其中
699	but keep on trying new models.	但继续尝试新的模型。
700	So basically when there is a situation	所以,基本上当有情况
701	that you don't trust anyone, you run a local model	你不相信任何人,你经营一个本地模特
702	and this is how you keep your data safe.	这就是你保护数据安全的方法
703	I do trust the hyperscalers,	我确实相信那些超规模的人
704	but sometimes for literally NMPI,	但有时对NMPI来说
705	material non-public information,	* 非公开资料,
706	that I won't even send to that,	我甚至不会送去,
707	I just can't even send that to the cloud.	我甚至不能把那个寄到云里
708	So that, I'll either do it manually without AI help	所以,我要么手动做,没有AI帮助
709	or if I really need to use AI,	或者,如果我真的需要使用AI,
710	then really carefully only use a local model.	然后非常小心地使用一个本地模型。
711	Interesting.	有意思,这很有趣
712	Okay, this is an interesting one.	好吧,这是一个有趣的。
713	Okay, what about loss of human control over AI?	好吧,那人类失去对AI的控制呢?
714	因为我和Yoshua Benja谈过了
715	who is very negative when it comes to free AI	在自由AI方面,谁是非常消极的
716	without any regulation.	没有规定。
717	And he painted me some very scary pictures of AI	他给我画了几张很恐怖的AI照片
718	taking over control,	接管控制权,
719	because basically the whole scenario is	因为基本上整个情景
720	we can't control something that's smarter than us.	我们不能控制比我们聪明的东西
721	And if AI gets smarter and smarter, where do we end up?	如果AI变得聪明和聪明,我们最终会在哪里?
722	What do you think about that?	你觉得怎么样?
723	I think about something else that we can't control,	我想到别的东西,我们不能控制,
724	which is airplanes.	这是飞机。
725	No one can build an airplane that you can fly perfectly.	没人能造出能完美飞行的飞机
726	Winds will buff it around.	风会吹动它周围。
727	And then candidly, in the early days of developing airplanes,	然后坦率地说,在飞机发展初期,
728	some airplanes crashed and people died	有些飞机坠毁,有人丧生
729	and it was tragic and awful.	悲剧和可怕。
730	But through the early lessons learned,	但通过早期的教训,
731	we then learned to control airplanes better and better	然后我们学会了更好控制飞机
732	so that today we can mostly get in an airplane	这样今天我们就可以乘飞机了
733	and not fear too much for our lives.	也不要太害怕我们的生活。
734	And it's really like that too with AI.	对AI来说也是这样
735	No one can perfectly control AI	没有人能完全控制AI
736	because it generates tokens or outputs	因为它会产生符号或输出
737	that are a little bit random,	这有点随机,
738	so we don't really know what exactly it'll do.	所以我们不知道它到底能做什么
739	But as we run them,	但当我们运行他们,
740	and there's been a small number of mishaps,	发生了少量的意外
741	which is unfortunate,	这是不幸的,
742	and some number of mishaps have done some real damage,	和一些灾祸 已经做了一些真正的损害,
743	but the way we engineer almost any system	但是我们设计任何系统的方式
744	from an airplane to electric circuits to now AI	从飞机到电路到现在的AI
745	is carefully grow their capabilities	仔细地提高他们的能力
746	so that we can have a controlled environment	以便我们有一个控制的环境
747	in which to measure what's wrong	用来衡量什么是错的
748	and then to shape it,	然后造出来,
749	to make sure we can control it well enough	确保我们能很好地控制它
750	that it behaves responsibly and safely.	它的行为是负责和安全的。
751	And to this day,	时至今日,
752	we can't perfectly control any airplane	我们不能完全控制任何飞机
753	and we will never perfectly control AI either.	我们也不会完全控制AI
754	But I think we are certainly controlling them well enough	但我认为我们控制得够好了
755	that this loss of control doesn't feel like science,	这种失控的感觉不像是科学
756	it feels like science fiction.	感觉就像科幻小说
757	Yeah, yeah.	对 对
758	What about deep fakes?	那深层假货呢?
759	Deep fakes are a problem.	深假是烦恼.
760	Well, one of the most disgusting things I've ever seen	我见过最恶心的事情之一
761	or heard of is non-consensual,	或听而不闻,
762	intimate deep fake imagery.	亲密的深层假象。
763	I'm really glad that US Congress has been moving	我真的很高兴美国国会 一直在移动
764	to let's pass laws, get rid of that, penalties for that.	让我们通过法律, 摆脱它,惩罚它。
765	I think there's some really problematic uses of AI	我觉得人工智能的用法有点问题
766	that we should outlaw, heavily penalize,	我们应当严惩他们,
767	let's just get rid of that.	我们把那个扔掉吧
768	What do you think about children and social connection	你对孩子和社会联系有什么看法?
769	when it comes to AI, with kids using more of AI?	当涉及到AI, 孩子们使用更多的AI?
770	Because we've seen social media,	因为我们看过社交媒体
771	how there are people who are dumb-scrolling all day.	怎么会有人整天都傻傻地滚
772	And my daughter, who is five years old now,	还有我女儿 她现在五岁了
773	whenever I don't have an answer,	当我没有答案时
774	is like, ask Chad GPT.	就像问查德GPT
775	Who's that person?	那个人是谁?
776	I'm like, I don't know.	我很喜欢,我不知道。
777	问查德GPT.
778	She thinks Chad GPT knows everything.	她觉得Chad GPT什么都知道
779	What would you say about kids' future with AI?	你会怎么说 孩子们的未来与AI?
780	First, I think kids have a bright future.	首先,我觉得孩子有一个光明的未来。
781	It's such an exciting time to be a child,	小时候真是激动人心
782	to grow up in this environment	在这个环境中成长
783	with tools that none of us have ever had before.	拥有我们从未有过的工具。
784	At the same time, we've seen that social media,	同时,我们看到社交媒体,
785	I think social media has probably been blamed	我觉得社交媒体可能已经被指责了
786	a bit more than it deserves,	比它应得的要多一点
787	but it does deserve blame,	但它确实值得责备,
788	has kind of not been great for kids.	有点对孩子不好
789	I actually worry a lot about,	我其实很担心
790	it's a wonderful tool,	这是一个很好的工具,
791	but AI damaging learning is something I worry a lot about.	但是,AI破坏的学习 是我担心的很多。
792	So it turns out, I have a five-year-old and a seven-year-old.	因此,结果,我有一个5岁和7岁的孩子。
793	When I teach them math, they're so young enough	当我教他们数学时,他们就足够年轻了
794	that I can basically not let them use the calculator.	我不能让他们用计算器
795	I can say, how do you multiply these numbers?	我可以说,你如何乘以这些数字?
796	And I don't give them a calculator	我不给他们一个计算器
797	and practice that with them.	和他们一起练习。
798	But as they're a little bit older,	但是,由于他们有点老,
799	I worry a lot about students	我担心很多学生
800	using cognitive offloading to AI	使用认知卸载到AI
801	in a way that damages the long-term learning retention.	以损害长期学习保留的方式。
802	But then at the same time,	但与此同时,
803	oh, I actually built an app.	哦,其实我造了一个应用程序。
804	I did not like any of the free online learning	我不喜欢任何免费的网上学习
805	to type types of things,	和类型的东西,
806	so I actually built my own	所以我实际上建造我自己
807	to have my daughter learn to type.	让我女儿学会打字
808	And I'm hoping that,	我希望这样
809	she's actually getting pretty decent now	她现在变得很体面了
810	for a seven-year-old.	给一个七岁的孩子。
811	Oh, so she's seven already?	哦,那她已经七岁了?
812	Oh yeah, she actually typed all the lowercase letters.	哦,是的,她实际上打 所有小写字母。
813	She's a little bit,	她有点
814	her shift uppercase letters a little bit not quite there.	她的变换大写字母有点不太对劲
815	But I think that this unlocks	但我认为这个解锁
816	responsible adult-supervised use of online tools.	由成年人负责监督在线工具的使用。
817	And I think it's really tricky, you know?	我觉得这很棘手 你知道吗?
818	I think adult-supervised use of digital tools	我认为是成年人监督使用的数字工具
819	seems a great thing for kids,	似乎对孩子们来说是件好事
820	but too many adults don't have time	但太多的成年人没有时间
821	to supervise the use of the tools	监督工具的使用;
822	and the incentives of, say, social media, right?	以及社交媒体的激励,对吧?
823	To do funny things.	做一些有趣的事情。 做一些有趣的事情。
824	Yeah, has to be the right incentive	是啊,一定有正确的激励
825	when it comes to AI.	当涉及到AI时.
826	Okay, you mentioned, we talked about the fears,	好吧,你提到, 我们谈论了恐惧,
827	we talked about how you can improve your work with AI.	我们谈过如何改善你和AI的工作
828	Can you name some of the biggest opportunities	你能说出最大的机会吗?
829	in AI in 2026 for people who want to build?	在2026年的AI中,对于想要建设的人?
830	For an individual that wants to build,	对于一个想要建设的人来说,
831	I don't think it's one size fits all.	我不认为它是一个尺寸适合所有。
832	But because the cost of building has plummeted,	但由于建筑成本暴跌,
833	I encourage people to learn AI,	我鼓励人们学习AI,
834	build fast, and talk to customers.	快速建设,并与客户交谈。
835	I find myself building things, I don't know,	我发现自己建造的东西, 我不知道,
836	every week, every weekend,	每个礼拜,每个周末,
837	because I or someone on our team will have some problem	因为我或我们团队中有人会有问题
838	and I have some idea for building some AI thing	我有点想法 建造一些AI的东西
839	to automate it.	把它自动化。
840	Last weekend, I had, really, I was using a frontier model	上个周末,我真的在用边疆模型
841	to analyze a lot of our key business metrics	来分析我们许多关键的商业标准
842	because I didn't have time to do it myself,	因为我没有时间自己做,
843	but it was kind of measuring, you know,	但它是一种测量, 你知道,
844	deviantized key business metrics,	偏离关键商业标准,
845	and I didn't have time to go find a data scientist	我没有时间去找数据科学家
846	to go work me on it,	去工作我做到这一点,
847	so I just did a variety of frontier models,	所以我刚刚做了各种前沿模型,
848	being really careful on their data retention policies.	对数据保存政策非常小心
849	I did not use models with data retention policies	我没有使用有数据保存政策的模型
850	that I don't like in order to analyze data.	我不喜欢分析数据
851	And then I find that what's happened with AI	然后我发现AI出了什么事
852	is the cost of building has plummeted,	建筑成本暴跌,
853	and so the challenge is shifting	因此,挑战正在改变
854	to deciding what to build,	决定建造什么,
855	which I've been calling the product management bottleneck.	我称之为产品管理瓶颈
856	And so people, founders, engineers, product managers,	于是人们,创始人,工程师,产品经理,
857	that can talk to customers,	可以和顾客交谈,
858	get a sense for the taste of judgment on what to build,	体会到判断如何构建的滋味
859	and then build with AI and iterate quickly,	然后用人工智能迅速建造,
860	I think that's just a ton of exciting things to do.	我想这只是一吨 令人兴奋的事情做。
861	And you've been starting so many companies.	你开了这么多公司
862	You're like, when I looked at your portfolio,	你就像,当我看着你的组合,
863	do you think for beginners,	你认为初学者,
864	when you said you built something during the weekend,	你说你在周末造东西的时候
865	how do you decide what to focus on,	如何决定什么是重点,
866	or you can pursue multiple ideas because of AI now,	或者你可以追求多种想法 因为AI现在,
867	and you can just be, you know,	你可以只是,你知道吗,
868	playing in different companies at the same time?	在同一时间在不同的公司演出?
869	It turns out building a company is still really, really hard,	原来建公司还是很辛苦
870	and so there's a lot to be said	所以有很多话要说
871	for a single-threaded leadership	单线领导
872	or someone that's fully focused on just one thing.	或者某个完全专注于一件事的人
873	I find that, you know, over a weekend,	我发现,你知道,在一个周末,
874	I can often build an own wrapper,	我常常可以自己做包装
875	build a simple application,	建立一个简单的应用程序,
876	but I wish it was that easy to build a large company.	但我希望有那么容易 建立一个大公司。
877	I find that building something meaningful	我发现建造一些有意义的东西
878	often takes either real technical depth	往往需要真正的技术深度
879	and or deep customer insight and integration of customers.	或客户的深刻见解和融入。
880	And yes, we can now, you know,	是的,我们现在可以,你知道,
881	use AI to code something in a few hours,	几小时后用人工智能编码
882	but that's a small piece of the puzzle.	但这是谜题的一小块
883	So spend time understanding the technical complexity	所以要花时间了解技术的复杂性
884	and building the really complex software	并建造非常复杂的软件
885	that takes us like months, you know, maybe years,	这需要我们像几个月, 你知道,也许几年,
886	or having that deep customer insight to decide what to build	或拥有深刻的客户洞察力来决定如何构建
887	that also just takes a lot.	这也只是需要很多。
888	Talking to people, reading facial expressions,	与人交谈,阅读面部表情,
889	surveys, doing that over and over	调查,反复调查
890	until we figure out what to build.	直到我们想出 建造什么。
891	And so I think sometimes there's a lot of value	所以我觉得有时有很多价值
892	to sampling widely,	广泛采样,
893	but then having that focus for an individual	但后来又以个人为重点
894	to go really deep in the couple sectors,	真正深入到夫妇部门,
895	that still seems important for building a business.	这对建立企业来说仍然很重要。
896	My last question, I know we don't have much time,	我最后一个问题,我知道我们时间不多了
897	but I wanted to ask you about AGI,	但我想问你关于AGI的事
898	just because people use this word so much,	只是因为人们使用这个词这么多,
899	and some people say, I think Jensen Huang said,	有人说,我想黄詹森说,
900	we already reached AGI.	我们已经到达AGI。
901	You said it's decades away.	你说过还有几十年
902	What's the one criteria when you're gonna say we reach AGI?	当你要说我们到达AGI时 唯一的标准是什么?
903	So different people say we reach AGI at different times	不同的人说我们在不同的时间到达AGI
904	because of different definitions of AGI.	因为AGI的定义不同.
905	The definition I'm most familiar with	我最熟悉的定义
906	is AI that could do any intellectual tasks	是AI,可以做任何智力任务
907	that a human can.	人类可以。
908	But so the human brain can take, say, five years	但人类大脑需要五年时间
909	to study and do a PhD thesis,	研究并做博士论文,
910	and so can AI write a PhD thesis,	AI也可以写博士论文,
911	or a human can learn to drive a truck	或者人类可以学会开车
912	through a dense rainforest	穿过茂密的雨林
913	with tens of minutes of practice.	和数十分钟的练习。
914	So when can AI do that,	所以,什么时候AI可以做到这一点,
915	to drive a new environment	驱动新环境
916	with tens of minutes of practice?	练习数十分钟?
917	It feels like there's a long list of these things	感觉有一大堆这些东西
918	that AI cannot do for what feels to me decades.	对我而言,大赦国际不能做几十年的感受。
919	I hope it's only decades.	惟愿数十载.
920	Maybe it'll turn out to be longer.	也许它会变成更长。
921	So that's what I think for that definition of AI or AGI.	所以这就是我对AI或AGI的定义的看法.
922	AGI还很远
923	But it turns out, because of economic incentives,	但事实证明,由于经济刺激,
924	I think OpenAI and Microsoft had an agreement.	我认为OpenAI和微软达成了协议.
925	There's actually been renegotiated now,	已经重新谈判过了
926	so that's gone away.	所以这已经消失了。
927	但OpenAI有经济刺激
928	to try to declare reaching AGI earlier.	试图更早地宣布到达AGI。
929	And so it turns out that if you come up	结果发现如果你上来
930	with other definitions of AI,	与大赦国际的其他定义,
931	depending on how far you lower the bar,	看你把酒吧放低多远
932	then you could totally have reached AGI already,	那你本可以联系AGI的
933	or even 30 years ago,	甚至在30年前,
934	depending on how you want to define it.	取决于您要如何定义它。
935	Yeah, true.	是的,没错。
936	Andrew, thank you so much for this	安德鲁,谢谢你
937	positive conversation, very applicable.	积极的谈话,非常适用。
938	I like when you watch something,	我喜欢你看东西
939	and then you go and you measure yourself	然后你去测量自己
940	against what people are doing with AI,	与人们对AI的所做所为相对应,
941	look at your process, and maybe expand it.	看看你的流程,也许扩展它。
942	So thank you so much for showing what your team is doing,	所以,非常感谢你展示 你的团队在做什么,
943	and thank you for your insights.	谢谢你的洞察力
944	Yeah, I think given the huge benefits of AI to come,	是啊,我认为考虑到AI的巨大好处 来,
945	I hope whoever's watching this is motivated	希望看这个的人有动机
946	to really go learn AI, apply it,	真正去学习AI,应用它,
947	and even to go build some things.	甚至去制造一些东西。

该视频共有字幕 947 条。解锁更多字幕为会员功能，请移动到 价格

P2 · 【AI趋势】吴恩达：AI最大的机会不在你以为的地方 · 打破AI恐惧 p02 无字幕 (P2)

【AI趋势】吴恩达：AI最大的机会不在你以为的地方 · 打破AI恐惧 p02 无字幕 (P2)

发布于 2026-09-02 · 时长 37:51

字幕摘录

时间	英文	中文
0:00	There's been a lot of misinformation about AI.	对AI的误导很多.
0:03	This is Andrew.	这是安德鲁。
0:04	他共同创办了Google Brain和Coursera.
0:07	His machine learning course has reached millions of learners	他的机器学习课程 已经达到了数百万的学习者
0:09	and he is one of the most influential voices in AI today.	他是今天AI最有影响力的声音之一。
0:13	So a handful of leading AI companies	所以,少数主要的AI公司
0:14	have been very loud voices of fear-mongering around AI	在AI周围发出恐惧的声音
0:18	to try to get regulations passed.	试图让 条例通过。
0:20	This drumbeat of fear-based messaging	恐怖讯息的鼓声
0:22	has skewed societal's perception	扭曲了社会观念

展开字幕全文（949 条）

序号	英文	中文
1	There's been a lot of misinformation about AI.	对AI的误导很多.
2	This is Andrew.	这是安德鲁。
3	他共同创办了Google Brain和Coursera.
4	His machine learning course has reached millions of learners	他的机器学习课程 已经达到了数百万的学习者
5	and he is one of the most influential voices in AI today.	他是今天AI最有影响力的声音之一。
6	So a handful of leading AI companies	所以,少数主要的AI公司
7	have been very loud voices of fear-mongering around AI	在AI周围发出恐惧的声音
8	to try to get regulations passed.	试图让 条例通过。
9	This drumbeat of fear-based messaging	恐怖讯息的鼓声
10	has skewed societal's perception	扭曲了社会观念
11	to be really negative on AI.	对AI非常负面。
12	People talk to me about data centers and job loss.	人们和我谈论数据中心和失业问题。
13	Maybe AI could do 30, 40% of many jobs.	也许AI能做30,40%的工作。
14	And what that means is, well, that 60% that the human does	也就是说 人类60%的工作
15	has become even more valuable.	已经变得更加宝贵。
16	What about loss of human control over AI?	对AI失去人的控制怎么办?
17	I think about something else that we can't control.	我想到别的东西,我们不能控制。
18	I think you're one of the voices in AI	我觉得你是AI里的声音之一
19	who comes with a huge background	有着巨大背景的人
20	and machine learning and teaching AI,	以及机器学习和教学
21	and also you're a positive voice	而且你是一个积极的声音
22	because this is something that I've been seeing,	因为这是我见过的
23	especially this summer,	特别是今年夏天
24	how polarized the society has become,	社会是如何两极分化的,
25	especially on social media when I'm posting about AI	尤其是在社交媒体上 当我发布AI
26	and people talk to me about data centers and job loss.	人们和我谈论数据中心和失业
27	Why do you think this wave started recently?	你觉得这波为什么是最近才开始的?
28	What do you think the causes are?	你觉得原因是什么?
29	There's been a lot of misinformation about AI	关于AI有很多错误信息
30	and the root cause of a lot of this	以及很多事情的根源
31	is an unfortunate attempt that started two, three years ago	这是一个不幸的尝试 开始于二,三年前
32	of I think PR and registry capture.	我认为公关和注册记录记录。
33	It turns out that the most,	事实证明,最
34	one of the most valuable things in AI right now	现在AI最有价值的东西之一
35	is the giant AI models,	是巨大的AI模型,
36	giant large language models that some companies have trained.	巨型大型语言模型 一些公司已经训练。
37	But if you spend billions of dollars training a model	但如果你花了数十亿美元训练一个模特
38	is really inconvenient if someone else trains a model	如果别人训练一个模型,那真的很不方便
39	and wants to give it to anyone in the world to use for free.	并想把它给世界上的任何人免费使用.
40	So a handful of leading AI companies, I think as you know,	因此,少数主要的AI公司,我想你知道,
41	have been very loud voices, fear-mongering around AI	在AI周围发出非常响亮的声音 恐惧的恐惧
42	to try to get regulations passed	试图通过法规
43	to create an unfair playing field that favors incumbents	创造一个有利于现任者的不公平竞争环境
44	so that we all have to pay a high toll for use of AI	因此我们都得付出高昂的代价来使用AI
45	while stymieing the other teams,	一方面阻击其他队伍,
46	be it researchers or other companies	不管是研究者还是其他公司
47	that wanna just give away open way to open source models	给开源模型开辟道路
48	that anyone could use much cheaper.	任何人都可以使用便宜得多。
49	Unfortunately, fear-mongering works.	不幸的是,恐惧的侵扰是有效的。
50	When you go and say AI is like nuclear weapons,	当你去说AI就像核武器,
51	which is an analogy that has no basis in fact,	这是没有根据的比喻,
52	what did they have to even do with each other?	他们之间有什么关系?
53	Or when you go around and cherry pick cases of AI,	或者当你绕圈子去 和樱桃挑AI的案件,
54	making a misstep and make it much bigger than it is,	做一个错误的步, 使它比它大得多,
55	or even spread misinformation about how AI uses data centers,	甚至散布关于AI如何使用数据中心的错误信息,
56	uses a lot more water than the actual reality.	用的水比现实多得多
57	This drum beat of fear-based messaging	这鼓拍基于恐惧的信息
58	has skewed societal's perception	扭曲了社会观念
59	to be really negative on AI, which is unfortunate	对AI非常负面,这是不幸的
60	because this is slowing down American adoption in AI.	因为这减缓了美国在AI的收养速度.
61	This is making America less competitive.	这使得美国的竞争力降低.
62	And unless we get the truth about AI out there,	除非我们知道AI的真相
63	which is that it's fantastic benefit with some problems,	这对解决一些问题有好处
64	but not nearly the degree to which they're blown up to be,	但不是被炸得那么严重
65	it will hurt individuals.	它会伤害个人。
66	I'm gonna read out some of the problems	我要宣读一些问题
67	that people are highlighting, job loss and inequality.	人们正在强调,失业和不平等。
68	What do you think?	你觉得呢?
69	The job apocalypse, the job pocalypse,	工作启示录, 工作浮夸,
70	this idea that AI will take over 50% of jobs,	这个想法,AI将占据50%的工作,
71	people will be out of work, rioting in the streets,	人们将失业,在街头暴动,
72	that's just not gonna happen.	那只是不会发生。
73	With every wave of technology, including AI,	随着每一波技术,包括AI,
74	the skills we need to do great work shifts,	我们所需要的技术 做伟大的工作转变,
75	and so AI is changing job professions.	所以AI正在改变职业
76	But boy, I wish AI just doesn't work well enough.	但是,孩子,我希望AI 只是工作不够好。
77	I know that handful of businesses want to hype up AI	我知道有几家企业想哄骗AI
78	to say, we have super intelligence,	说,我们有超级智能,
79	or we'll have artificial intelligence or whatever,	或者我们有人工智能之类的
80	and can do all the stuff that humans do.	并且可以做人类做的一切
81	I wish AI work better, which is not good enough	我希望人工智能工作做得更好 这还不够好
82	to make AI do everything a human does.	让人工智能做人的一切
83	And if you look at the analysis of jobs,	如果你看看工作的分析,
84	economists, like my friend Eric Brinovson at Stanford,	经济学家,像我的朋友埃里克·布里诺夫森 在斯坦福,
85	Andy McAfee at MIT,	麻省理工学院的安迪·麦卡菲
86	economists have analyzed many people's jobs	经济学家分析了很多人的工作
87	and by breaking it down into individual tasks,	并把它分成个人任务,
88	and maybe AI could do 30, 40% of many jobs.	也许AI能做30%,40%的工作。
89	And what that means is, well, that 60% that a human does	也就是说,人类60%的工作
90	has become even more valuable	越来越珍贵
91	because it's called an economic compliment,	因为这叫做经济赞美,
92	so the 30, 40% is now cheaper.	因此,30%,40% 现在更便宜。
93	And so what will happen is people that use AI,	接下来是使用人工智能的人
94	maybe people that use AI will replace people	也许用人工智能的人会代替别人
95	that don't use AI, but AI is not in a position	但是AI没有地位
96	for the vast majority of jobs to replace people.	给绝大多数的工作来代替人。
97	Of all the different professions,	在不同的职业中
98	the one that's most affected by AI now is software engineering	现在受AI影响最大的是软件工程
99	because AI is actually fantastic at writing code.	因为人工智能在写代码方面的确很棒
100	And what we see is that the number of job openings	我们所看到的是 职位空缺的数量
101	in software engineering is up,	在软件工程上,
102	contrary to what the doom theorist would say, right?	与末日论者的说法相反,对不对?
103	AI is not actually able to replace software engineers,	AI实际上无法替换软件工程师,
104	and all the good software engineers I know	还有我认识的所有优秀的软件工程师
105	are busier than ever.	比以往任何时候都更忙。
106	Now, the flip side of it is,	现在,它的反面是,
107	if someone's still writing code,	如果有人还在写密码
108	like this 2022 before CHI GPT, they're in trouble.	像这个2022年前的CHI GPT, 他们有麻烦了。
109	They need new skills, like don't do stuff	他们需要新技能,就像不做东西一样
110	that 30, 40% of AI can automate.	30,40%的人工智能可以自动化.
111	You've got to stop doing that, let AI do that,	你必须停止这样做, 让AI做到这一点,
112	but then gain new skills to do the other 60, 70%	然后获得新技能 做其他60,70%
113	that AI cannot do.	这是大赦国际不能做的。
114	What would your advice be to new graduates?	你给新毕业生的建议是什么?
115	Because I talked to Eric on this podcast,	因为我和艾瑞克谈过了
116	and he was talking about that there's not really	他当时在说真的没有
117	a lot of impact on the job market,	对就业市场影响很大,
118	except for, I think he mentioned people from 18 to 25	除了,我想他提到了18到25岁的人
119	who just graduated.	谁刚毕业。
120	What would be your advice to those people	你对这些人的建议是什么?
121	who don't have the expertise	谁没有专长
122	maybe to strategize in their job yet?	也许他们的工作有策略吗?
123	They can only do manual work that AI can do as well.	他们只能做人工工作,而AI也可以做.
124	So one real challenge for fresh college grads	所以,一个真正的挑战 对于新鲜的大学毕业
125	is that the university system is slow to adapt.	大学系统适应性很慢
126	And so, I love academia,	因此,我喜欢学术界,
127	I think we should all support academia and universities.	我认为我们都应该支持学术界和大学。
128	And when AI comes and transforms the way software is written,	当AI来改变软件的写法时,
129	universities often take like a year or two	大学通常要花一两年时间
130	for the faculty to master the skills,	以便教职员掌握技能,
131	then create new courses,	然后创造新的课程,
132	get curriculum committee approval or whatever,	得到课程委员会的批准或什么的,
133	get the faculty senate to vote.	让参议院投票
134	It just takes years, and that speed of change in academia	需要很多年 学术界的变革速度
135	is very poorly matched to the speed of change in AI.	与AI的改变速度非常差。
136	So sadly, many universities are still teaching students	可惜许多大学还在教学生
137	to be ready for the jobs of 2022,	准备好2022年的工作,
138	when we shouldn't even be teaching them	我们不应该教他们
139	for the jobs of 2026,	为2026年的工作,
140	we should be teaching them for the jobs of 2028 and beyond.	我们应该教他们2028年及以后的工作
141	And what this means is the job openings are there.	这说明职位空缺就在那里。
142	Tons of employers I know just can't find enough skill,	我知道很多雇主都找不到足够的技能
143	you know, people at any level of seniority.	你知道,任何级别的人。
144	But it turns out in my office right now,	但是现在在我的办公室里
145	we have a lot of interns,	我们有很多实习生
146	there are current college students, fresh college grad,	有现在的大学生, 新鲜的大学毕业,
147	we're also one high school intern,	我们也是高中实习生
148	and they're amazing and productive.	和它们惊人和生产。
149	But the key is they're all very AI native,	但关键是他们都是AI本地人
150	they all use AI tools to do the things that I could do,	他们都用人工智能工具来做我能做的事,
151	but then also, you know,	还有,你知道,
152	lean in to doing the things that humans can do	做人类能做的事
153	that AI can't for a long time.	AI不能在很长一段时间。
154	So there's plenty of work for people to do.	故众生有余功德.
155	But so my advice to fresh college grads,	所以我建议大学毕业
156	or the people currently in college is,	或者目前大学里的人
157	by all means, work hard in classes, you know,	在课堂上努力工作
158	get good grades, learn from the instructors,	学好成绩 向老师学习
159	but to the extent that there's still additional skills	但只要还有其他技能
160	that the university has not yet adapted to teaching,	大学还没有适应教学,
161	then find other ways to learn online,	然后找其他方法在网上学习,
162	无论是来自Coursera还是深层学习.ai还是Udemy
163	or other places where you can gain the more cutting edge skills,	或其他可以获得更先进技能的地方,
164	especially AI skills,	特别是人工智能,
165	that universities have not yet worked on the curricula.	大学尚未制定课程。
166	Quick pause, because what Andrew just said about workflows	暂停一下,因为安德鲁刚才说的工作流程
167	connects to something that HubSpot just put out for free.	连接到HubSpot刚刚免费推出的东西.
168	The thing is, the real opportunity right now	问题是,真正的机会,现在
169	is not just in the models themselves,	不只是模型本身,
170	but in what you build on top of them,	但你们在他们的上面所创造的,
171	how you turn AI from a chat window	从聊天窗口打开AI
172	into workflows that actually do work for you.	进入工作流程 真正为你工作。
173	And the good news is,	好消息是,我们
174	you don't have to start by building a full agent.	你不必开始 建立一个完整的代理。
175	You can start much simpler with better prompts.	你可以用更好的提示开始简单得多。
176	HubSpot刚刚发布了高级聊天GPT
177	Prompt Engineering Playbook, and the idea is very simple.	即时工程游戏本 这个想法很简单
178	In seven days, it helps you move	7天之内,它能帮助你移动
179	from getting generic AI answers	从获得通用AI的答案
180	{\fn黑体\fs22\bord1\shad0\3aHBE\4aH00\fscx67\fscy66\2cHFFFFFF\3cH808080}你更能保持一贯 {\fn黑体\fs22\bord1\shad0\3aHBE\4aH00\fscx67\fscy66\2cHFFFFFF\3cH808080}我很高兴你
181	and useful options.	和有用的选择。
182	Because we've all had this moment,	因为我们都有这一刻
183	you type in one vague sentence,	你输入一个模糊的句子,
184	get something mediocre back,	拿些平庸的东西回来
185	and then spend the next 20 minutes fixing it yourself.	然后再花20分钟自己修
186	This playbook is basically built for that exact problem,	这本游戏本基本上是为这个问题而建的
187	and it walks you through a single progression.	它让你经历了一个单一的进化。
188	First, you learn how to structure prompts	首先,你学会如何构建提示
189	with things like role, context, format, and parameters.	包括角色、上下文、格式和参数。
190	Then it gets more advanced.	然后它会更先进。
191	Few shot examples, chain of thought reasoning,	很少有例子 思想推理链
192	more precise prompt structures,	更精确的快速结构,
193	and ways to make outputs more reliable.	以及如何使产出更加可靠。
194	What I like is that it doesn't stop at individual prompts.	我喜欢的是它不会停止 个别的提示。
195	It also shows you how to build reusable AI personas,	也告诉你如何建造可重复使用的AI人物,
196	modular prompt components,	模块快速组件,
197	and eventually your own signature prompt engineering system.	最后你自己的签名即时工程系统。
198	So instead of starting from scratch every time,	所以不是每次从零开始
199	you're building a repeatable way to work with AI,	你正在建立一个可以重复的方式 与AI合作,
200	whether you're creating content, analyzing data,	你是否正在创建内容 分析数据
201	or making business decisions.	或做出商业决定。
202	The playbook is free, links in the description.	游戏本是免费的,在描述中的链接.
203	Thanks to HubSpot for sponsoring this video.	感谢HubSpot赞助了这段视频.
204	And now back to our conversation with Andrew.	现在回到我们和安德鲁的谈话
205	I want to say one other thing.	我想说另外一件事。
206	It turns out, if you look at the skill map changes,	事实证明,如果你看 技能地图的变化,
207	one of the most important changes is	最重要的变化之一是
208	it's so much easier to build with AI than before.	和AI一起建起来比以前容易多了
209	When something becomes much easier,	当事情变得容易得多,
210	a lot more people should do it.	更多的人应该这样做。
211	And so now, not only should professional software engineers	所以现在,不仅专业的软件工程师
212	build software with AI,	与AI一起构建软件,
213	it's becoming much easier for everyone to build with AI,	对AI来说,每个人都更容易建起来
214	and people that embrace that and do so	和接受并接受的人
215	will be more productive and will accomplish more,	将会更加丰硕,并且更加成功,
216	and I think have more fun than the ones that don't.	我觉得比那些不开心的人更好玩
217	And AI lets you build really fast.	人工智能让你建得很快
218	So for people that are not just software engineers,	所以对于不仅仅是软件工程师的人来说
219	but marketers, recruiters, HR professionals,	但市场营销者、招聘者、人力资源专业人员,
220	operations specialists,	业务专家,
221	I think if they learn to build with AI,	我想如果他们学会和AI一起建设
222	they really just do much more, whatever the job role is.	他们只是做更多的事情, 不管工作的角色是什么。
223	How do you, by the way, measure the increase in productivity	顺便说一句,你怎么衡量生产率的提高?
224	when you deploy AI?	当你部署AI时?
225	Do you have a KPI in your company?	你的公司有KPI吗?
226	I wish it was a simple answer.	我希望这是一个简单的答案。
227	I find that the business outcome of AI	我发现AI的商业结果
228	is more a function of the business	更属于企业的功能
229	than a function of the AI.	而不是大赦国际的职能。
230	For some, it may be increased customer growth and retention,	对一些人来说,这可能会增加客户的增长和保留,
231	or maybe faster to serve customers,	或者更快地为顾客服务,
232	or it may be increased accuracy in some tasks.	也可能提高某些任务的准确性。
233	So the KPI tends to be related to the business	因此,KPI倾向于与业务有关
234	rather than the AI.	而不是大赦国际。
235	So you can't directly measure just AI,	所以你不能直接测量AI
236	because it's an interesting thing to do,	因为它是一个有趣的事情做,
237	because we've been deploying AI actively in my company,	因为我们一直在积极部署AI在我的公司,
238	and I think for me as a media company,	我认为作为一个媒体公司,
239	it's probably the amount of views, the output.	可能是浏览量 输出量
240	It's just interesting how different people measure,	有趣的是不同的人如何衡量,
241	even revenue, like if you're becoming more effective	甚至收入,就像如果你越来越有效
242	with how you make money.	你如何赚钱。
243	Actually, say more.	其实,说更多。
244	How are you using AI in your business?	你怎么在你的生意中使用AI?
245	Oh my God.	天呐
246	So first of all, we have Claude for all of us,	所以首先,我们有克劳德为我们所有人,
247	and we have certain projects	我们有一些项目
248	for every social media that we're on.	我们所有的社交媒体。
249	So for example, for this podcast,	比如说,对于这个播客,
250	we have a project that's called Guests,	我们有个叫"客人"的项目
251	and it knows all the analytics from previous guests,	它知道所有分析 从前宾客,
252	and it has certain criteria	而且它有某些标准
253	on which we rank every single person	我们把每个人排在其中
254	who comes to the podcast,	他来见播客,
255	whether he or she's cited,	无论他或她被引用,
256	whether they have a certain opinion on AI,	他们是否对大赦国际持有某种意见,
257	whether they've been active with AI in their company,	他们是否一直活跃 在AI在他们的公司,
258	or if they're a recent founder in AI.	或者他们是AI的最近创始人
259	So it gives them different weights,	因此,它给他们不同的重量,
260	and it comes up with a grade based out of 40.	并得出一个等级 根据40。
261	40 meaning tier one, 30 meaning tier three, et cetera.	40表示一级,30表示三级等.
262	And then we have another one	礛и临Τ
263	that analyzes every single podcast	分析每一个播客
264	and gives me tips on how to ask questions.	并给我如何提问的提示。
265	与Instagram相同,与LinkedIn相同.
266	It has my tone of voice, personal dossier,	它有我的声音,个人档案,
267	my business strategy.	我的业务策略
268	So whenever it writes something,	所以,每当它写的东西,
269	it knows all the facts about me, how I sound.	它知道关于我的所有事实 我的声音。
270	Every social media is run by a person.	每个社交媒体都由一个人管理.
271	So a person makes a strategic call.	所以一个人做出一个战略号召。
272	And by the way,	顺便说一句
273	if you can give me feedback on this, if I can improve.	如果你能给我反馈, 如果我能改进。
274	So what I'm working on right now is closing the loop,	所以,我现在的工作是 关闭循环,
275	because sometimes they send me a text.	因为有时候他们会给我发短信
276	I'm like, oh, we need to change this, this, and that.	我想,哦,我们需要改变这个,这个,那个。
277	But that happens in a chat, in Telegram.	但是,这发生在聊天, 在Telegram。
278	And we have this feedback.	我们得到了这个反馈。
279	We have a bot that scans all of our chats.	我们有一个机器人 扫描我们所有的聊天。
280	But I really want AI to be able to learn continuously	但我真的希望AI能够不断学习
281	from this feedback to just know my taste better.	从这些反馈来了解我的品味
282	There's one thing I see a lot in AI,	我在AI里看到很多东西
283	which is it turns out for AI's data scientists	对AI的数据科学家来说
284	or AI's brainstorming partner,	或AI的集思广益伙伴,
285	it often comes up with one or two good ideas,	它经常提出一个或两个好主意,
286	two or three mediocre ones,	两三个平庸的,
287	and four atrocious ones.	和四个残暴的。
288	And sometimes you wonder,	有时候你会想
289	how could my AI have thought	我的人工智能怎么会想到
290	that could even be a plausible idea?	这甚至可能是一个可信的想法?
291	And to me, this relates to the job-pocalypse point of view,	对我来说,这涉及到 工作启示论的观点,
292	which is that for a long time,	长久以来,
293	humans, you, me, everyone watching this,	人类,你,我,每个人都看这个
294	will have a significant context advantage over AI,	与大赦国际相比,在环境方面具有显著优势,
295	which is that you know something	就是你知道一些东西
296	that's incredibly obvious to you	这对你来说是显而易见的
297	that that was an awful idea, but the AI did not.	那是个糟糕的主意 但人工智能没有
298	And it turns out that one of the reasons	结果发现一个原因
299	why AI will not replace our jobs or whatever,	为什么AI不会取代我们的工作什么的,
300	but perhaps not anytime soon,	但也许不会很快
301	is because humans have a massive context advantage	因为人类拥有巨大的环境优势
302	compared to AI.	与大赦国际相比。
303	We know so much that from our years of experience	我们从多年的经验中知道这么多
304	that we talked to customer,	我们和顾客谈过
305	we saw the funny facial expression that told us,	我们看到了有趣的面部表情 告诉我们,
306	ah, they don't like this.	啊,他们不喜欢这样。
307	Or we talked to business or our manager said,	或者我们跟生意谈了 或者我们的经理说
308	hey, blah, blah, blah, I really care about this.	嘿,嘿,嘿,我真的关心这个。
309	And so it turns out that almost all humans,	结果发现几乎所有的人
310	well, maybe all humans, just know a lot of stuff	好吧,也许所有的人类, 只是知道很多的东西
311	that the plumbing does not exist	水管不存在
312	and they don't think exists for the foreseeable future.	他们认为在可预见的未来不存在
313	So AI to get, I know sometimes people talk about	所以,AI得到,我知道有时 人们谈论
314	the importance of human judgment or human taste.	人类判断或品味的重要性。
315	And sometimes you wonder, all right, what is taste?	有时候你会想,好吧,什么是味道?
316	Is this fuzzy thing?	这是模糊的东西吗?
317	But to me, the technical thing that underlies	但对我来说, 技术问题的基础
318	why humans have better judgment and better taste than AI,	为什么人类比AI有更好的判断和品味,
319	is this context advantage.	这是上下文的优势。
320	And because this is a long-term advantage,	因为这是一个长期优势,
321	like no one's gonna solve this, you know, in a few years,	就像几年后 没人能解决这件事
322	this is why we just need a lot more humans	所以我们需要更多的人
323	with that judgment and taste to keep on complimenting AI.	以判断和品味 继续称赞AI。
324	And doesn't this make education even more important?	这难道不使教育更加重要吗?
325	Because education gives us context.	因为教育给我们提供了背景。
326	This is another thing I'm hearing about AI,	这是另一件事 我听到关于AI,
327	like you won't need education	就像你不需要教育一样
328	because all the information is at your fingertips.	因为所有信息都在你的指尖上
329	You just ask, Chad GPT.	你只是问,乍得GPT。
330	But when you say context and taste,	但当你说上下文和品味时
331	for me, that's years of acquiring knowledge	对我来说,这是多年 获得知识
332	and learning from the best and seeing how they perform	向最好的人学习,看看他们的表现
333	versus just asking a chat.	而不是只是问话
334	I'm gonna say something that may be controversial.	我要说一些有争议的话
335	I don't know if I've said this publicly,	我不知道我是否公开说过
336	but I think it's true, which is frankly,	但我认为这是真的 老实说
337	AI models are terrible for learning.	AI模型对于学习来说是可怕的.
338	I know people think AI is wonderful at getting things done,	我知道人们觉得AI很擅长做事情
339	use it all the time, love it.	总是用它,爱它。
340	All the data that's coming out is that	所有的数据出来是
341	when say college students use AI, we know this.	当说大学生使用AI, 我们知道这一点。
342	It's just the studies now back up as well,	只是现在的学习也有所恢复
343	so we also have numbers, but the data is very clear.	因此,我们也有数字, 但数据非常清楚。
344	Students go higher on homeworks when they use AI,	学生用人工智能做功课会更高
345	yay, higher homework schools.	是的,高等家庭作业学校。
346	But retention, their long-term performance is much worse	但保留,他们的长期表现更差
347	because their AI do the work for them.	因为他们的人工智能为他们工作。
348	More and more studies are coming out to back this up now	越来越多的研究 出来支持现在
349	that I think people think, oh, it turns out,	我认为人们会认为, 哦,结果,
350	I think Wikipedia is a wonderful tool,	我认为维基百科是一个很棒的工具,
351	it has tons of facts.	它有很多事实。
352	Web search is a wonderful tool, it has tons of facts.	网络搜索是一个很棒的工具,它有很多事实。
353	But it turns out that when you ask AI to do work for you,	但事实证明 当你要求AI为你工作时
354	you are cognitive offloading to AI,	你将认知卸载到AI,
355	which is great because that's how society moves forward	这是伟大的,因为 社会如何前进
356	and it gets work done.	并且它得到了工作完成。
357	But human retention is much worse.	但人类保留的情况更糟。
358	It's just so clear that LLMs, as they are most commonly used,	很明显,LLMS, 因为最常用的是,
359	are terrible for learning.	学习是可怕的。
360	I'm not saying there's no way to use it	我不是说没有办法用它
361	in a way that is good for learning.	以有利于学习的方式。
362	I think there are ways to use it that are good for learning.	我认为有方法使用它,对学习有好处.
363	But even for myself, there's so many things	但即使是我自己, 有很多事情
364	that are also AI modeled over the last six months	过去六个月的人工智能模型
365	or whatever, like building some project,	或者什么的,像建设一些项目,
366	how does this front-end backend component work, whatever.	这个前端后端组件是如何工作的,什么的。
367	Give me the answer, get the job done, it was fantastic.	给我答案,把工作做好,真是太棒了
368	But six months later, I don't remember the answer	但半年后 我不记得答案了
369	when I need to redo that front-end backend component,	当我需要重做前端后端组件时,
370	I ask the AI again.	我又问AI。
371	So data's really clear.	所以数据非常清晰。
372	We should stop thinking of AI as helpful for learning.	我们不应该把人工智能视为有助于学习。
373	At least the vast majority of ways	至少绝大多数的方法
374	that the vast majority of people are using AI models today	绝大多数人今天都在使用AI模型
375	is absolutely terrible for learning.	这对学习是绝对可怕的。
376	But you're building a company helping solve that, right?	但你正在建立一个公司 帮助解决这个问题,对不对?
377	Because the one-to-one tutoring with AI	因为跟AI的一对一辅导
378	is that where you just announced	是你刚刚宣布的
379	和Coursera的一亿投资
380	Yes, so we started about leading a new organization	是的,所以我们开始领导一个新的组织
381	called Learn Vector that is focused	名为“ 学习矢量” 的焦点
382	on building new learning experiences	积累新的学习经验
383	that is much more one-to-one than one-to-many.	这比一对一多得多
384	So 15 years ago, I was privileged to participate	所以15年前,我有幸参加
385	in the online causes movement	在线原因运动
386	that I think changed the way a lot of people learn.	我认为改变了很多人的学习方式
387	But that was and still remains largely	但它过去和现在 基本上
388	a one-to-many experience where everyone	一次到多次的体验 每个人
389	kind of watches the same video, which is actually okay.	那种看同一个视频, 这其实是好的。
390	It actually works well.	实际上效果不错。
391	But the technology now exists to create much more	但现在已有技术 创造更多
392	personalized, customized one-to-one experiences.	个性化,定制的一对一体验.
393	And so our team is working hard on that.	因此,我们的团队正努力进行这项工作。
394	I think we'll have a lot more to show by early next year.	我想到明年初我们还有很多事情要做
395	When think about human skill development,	想想人类技能的发展
396	I feel like because AI has so heavily impacted	我觉得因为人工智能影响很大
397	software engineering, what we see happening	软件工程,我们看到的
398	in the job market for software engineering	在软件工程的就业市场上
399	is a harbinger or as a forerunner	是预兆还是预兆?
400	of what we'll see in other disciplines as well.	我们在其他学科也会看到的东西
401	And in software engineering,	在软件工程方面
402	people need to learn new skills, but when they do,	人们需要学习新的技能,但当他们学习时,
403	they are thriving and creating more value	他们正在繁荣,创造更多的价值
404	and frankly getting raises	坦率地说,我得到了高薪
405	and doing even more exciting projects.	和做更令人兴奋的项目。
406	And what I've seen the early signs of	我所看到的早期迹象
407	in other disciplines as well,	其他学科,
408	for example, in software engineering,	例如,在软件工程方面,
409	most developers like front-end, back-end developers	大多数开发者喜欢前端,后端开发者
410	have now become full stack developers	现已成为完整的堆栈开发者
411	because of AI help, you could take on broader scope.	因为AI的帮助, 你可以承担更大的范围。
412	I'm seeing early signs of this in other disciplines as well	我在其他学科也发现了早期的迹象
413	where, for example, someone that in marketing,	比如说,在营销中的人,
414	that did marketing coordination,	做营销协调,
415	coordinating marketing campaigns with AI help,	在大赦国际的帮助下协调营销活动,
416	they can now become more of a full cycle marketer	现在他们可以更成为 整个周期的营销者
417	and take on a broader scope.	并且扩大范围。
418	And I'm seeing, you know, frankly,	而我看到,你知道,坦率地说,
419	sources in recruiting become more full cycle	招募来源更全面
420	to end to end recruiting.	以结束招募。
421	So now the good news and bad news	所以现在好消息和坏消息
422	is for people to step up to these broader roles,	也就是人们去发挥这些更广泛的作用,
423	you do need to learn AI skills,	你需要学习人工智能技能,
424	but also it's not just learning AI,	而且不只是学习AI,
425	you also need to learn these other skills,	您还需要学习这些其他技能,
426	like how do you do the other parts of marketing,	就像你如何做市场营销的其他部分,
427	of recruiting or software engineering or AI engineering.	招聘、软件工程或AI工程。
428	So I think this actually creates a heavy need,	所以我认为这实际上造成了一个沉重的需要,
429	a big need for people to gain new skills,	需要人们获得新的技能,
430	but when they do, which is both AI skills,	但当他们这样做, 这既是AI的技能,
431	but also disciplinary skills,	同时也有纪律技能,
432	then they can do much more, hopefully have more fun,	然后,他们可以做更多的, 希望有更多的乐趣,
433	work on more exciting projects,	在更刺激的项目上工作,
434	hopefully get paid more as well.	希望也能得到更多的报酬。
435	And one reason I kind of worry about the fear monitoring	还有一个原因 我担心恐惧的监控
436	is I got an email from someone	我收到别人的邮件
437	that was about to enter college,	即将进入大学,
438	and, you know, he emailed me saying,	他给我发了邮件说
439	hey, Andrew, taking online courses,	嘿,安德鲁,上在线课程
440	but I'm really struggling with	但我真的很挣扎
441	what I should major in college,	我应该在大学主修,
442	because in four years, won't AI do all this	因为四年后,不会让AI做这一切
443	and everything I learn will be obsolete?	我学到的一切都会过时?
444	And the answer is no, of course it won't all be obsolete,	答案是没有,当然不会都是过时的
445	but when we keep on pushing these fear messages,	但是当我们不断推波助澜时
446	we make people wonder if they will even be relevant,	我们让人们怀疑他们是否相关
447	and it makes people not lean in to gain these skills,	它使得人们不 倚靠获得这些技能,
448	they'll put them in a much better position.	他们会让他们处于更好的地位。
449	So I see very clearly that these fear monitoring messages	所以我非常清楚地看到 这些恐惧监测信息
450	are distorting how many people,	扭曲了多少人,
451	including, you know, high school students,	包括,你知道,高中学生,
452	college students, fresh grads, think about the economy,	大学生, 新生, 思考经济,
453	and frankly, making people give up	坦白地说,让人们放弃
454	is one of the worst things we've been doing	是我们做过的最糟糕的事之一
455	in this era when people that lean in will thrive.	在这个时代, 人倾斜将繁荣。
456	Andrew has taught over 8 million people AI.	安德鲁教过超过800万人AI.
457	He started teaching machine learning online back in 2011,	他从2011年开始在网上学习机器,
458	years before the current AI boom.	距当前AI繁荣还有好几年
459	Now one of the companies he's building	现在他正在建设的公司之一
460	is focused on AI agents.	重点是人工智能特工
461	From the way that it sounds,	从听起来,
462	it can still feel way too technical.	感觉还是太技术了
463	So I put together a step-by-step guide	所以我编了一部"一步一步"的指南
464	to building your first AI agent with no coding required.	来建立第一个不需要编码的AI代理.
465	It walks you through what to automate, how to set it up,	它让你通过什么 自动化,如何设置它,
466	and how to make it actually useful.	如何让它真正有用。
467	It is in my newsletter this week.	我这周的通讯里有这个
468	The newsletter is called Future Proof.	该通讯名为"未来证明".
469	It's free.	自由无碍.
470	Link is in the description.	链接在描述中.
471	What would you reply back to that email	你对那封邮件有什么反应?
472	that somebody sent you?	有人派你来的?
473	What would you say is the best major to study now,	你会说什么是最好的主 现在学习,
474	to thrive in AI era?	在AI时代蓬勃发展?
475	Do you think it's like going deep into a niche	你觉得这就像深入到一个小地方
476	or just broader computer science	或者只是更广泛的计算机科学
477	so that you can acquire AI skills really fast?	这样你才能很快获得人工智能技能?
478	You know, I don't know what's the best major.	你知道,我不知道什么是最好的主修课。
479	There are an awful lot of great majors.	有很多伟大的专业
480	It's like, I kind of feel like	我觉得...
481	what's the best job in the world?	什么是世界上最好的工作吗?
482	It's like, what's the best major in the world?	这就像, 什么是世界上最好的主修?
483	Something that you love, right?	你爱的东西,对不对?
484	Yeah, my daughter wants to be an astronaut.	对,我女儿想当宇航员
485	I don't know if she can major in becoming an astronaut.	我不知道她能否成为宇航员
486	I have to think about that.	我得考虑一下
487	When she gets older, she may change her mind.	她长大后可能会改变主意
488	I see so many opportunities across so many job roles.	我看到很多机会 贯穿这么多的工作角色。
489	It all seems very exciting to me.	这一切似乎都让我兴奋。
490	But do learn AI, do learn to build with AI.	但要学AI,要学与AI一起建设.
491	The other thing that my team's been working	还有一件事,我的团队一直在工作
492	on an AI engineering skills map to try to map out	在AI的工程技能地图上尝试映射
493	the most important skills for AI engineering.	AI工程最重要的技能.
494	One thing that I felt intuitively,	有件事我直觉地觉得
495	but I was surprised to see it show up in the data	但我惊讶地看到它出现在数据中
496	was that a lot more job descriptions seem to be saying	更多的职务说明似乎在说
497	they want people that demonstrate	他们希望人们展示
498	a very high sense of agency.	一种很高的代理意识。
499	Because it turns out with AI,	因为有AI的发现
500	there are a lot more opportunities	还有更多的机会
501	for individuals to spot problems	让个人发现问题
502	and go build something or do something to go solve it.	并去做一些事情或做一些事情去解决它。
503	So I think we're really evolving.	所以,我认为我们真的在进化。
504	Well, we've long been evolving,	我们早就在进化了
505	but we're accelerating past the era	但我们正在加速穿越时代
506	where people sit around and wait for their boss	在那里人们坐着等待他们的老板
507	to tell them what to do.	告诉他们该怎么做。
508	This is what I've been feeling a lot,	这就是我一直以来的感受,
509	especially when we started doing remote work.	特别是我们开始做远程工作的时候
510	I want people to be entrepreneurs within their niche.	我希望人们在自己的专长内成为企业家.
511	Like if you're helping me with LinkedIn,	就像你帮我处理LinkedIn一样
512	you're an entrepreneur there.	你是一个企业家在那里。
513	You can hire more contractors,	你可以聘请更多的承包商,
514	you can deploy different tools,	你可以部署不同的工具,
515	you make the strategic decision	你做出战略决定
516	whether this topic is good or not.	不论这个议题是否良好。
517	Shall we proceed with it?	我们继续吧?
518	I really think, and tell me if you agree with me,	我真的觉得 如果你同意的话 告诉我
519	we're moving into that job market	我们要进入这个职业市场
520	where everyone is kind of independent in their workplace.	在那里,每个人都在他们的工作场所是独立的。
521	I think people will have much more autonomy and creativity.	我认为人们会有更多的自主权和创造力.
522	So I agree with that.	所以我同意这一点。
523	And I'd even go one step further,	我甚至更进一步,
524	which is I talk to a lot of people,	我和很多人谈过
525	engineers and others in large companies	大公司的工程师和其他人
526	that tell me that the manager tells them	这告诉我,经理告诉他们
527	to stay in their swim lane.	留在他们的游泳道。
528	They'll say, oh, I have this creative idea,	他们会说,哦,我有一个创意的想法,
529	but the manager says, no,	但经理说,不,
530	I need you to focus on this one thing, frankly,	坦白地说,我需要你集中注意这个
531	often because the manager's career depends on it.	经常是因为经理的职业生涯取决于此.
532	But I feel like the number of opportunities	但我觉得机会之多
533	for people to spot things outside their swim lane,	人们在游泳道外发现东西,
534	and then in a responsible way,	然后以负责的方式,
535	explore how to get it done,	探索如何做到这一点,
536	that feels very exciting to me.	感觉很刺激
537	And I think that in the future,	我认为在未来,
538	the businesses that serve a culture	为文化服务的企业
539	that encourage people to learn AI,	鼓励人们学习AI,
540	build fast, responsibly,	快速、负责地建设
541	talk to customers would drive a lot more value	和顾客聊天会吸引更多的价值
542	than the more hierarchical silo organizations.	而不是更分级的组织。
543	Yeah, it starts with hiring the right people	是啊,它从雇用合适的人开始
544	and then nurturing this in your organization.	然后在你的组织里培育这个
545	When you say learn how to use AI	当你说学习如何使用AI
546	and become proficient with AI,	并且精通AI,
547	can you give me some benchmarks?	你能给我一些基准吗?
548	Like of a person who's like say a marketer,	就像一个像说营销员的人,
549	knowledge worker, advanced with AI.	知识工作者,高级AI.
550	What are you looking for	你们在找什么?
551	when you're interviewing this person?	当你采访这个人时?
552	I'm pretty sure my team's ahead of the curve.	我敢肯定,我的团队 领先曲线。
553	All of my marketers know how to code.	我所有的推销员都知道如何编码
554	So as part of how I interview marketers,	因此,作为我如何采访营销员的一部分,
555	we ask them what they've built,	我们问他们建造了什么
556	and if they have not built any software.	如果他们没有建造任何软件,
557	If it's a dashboard, is it good or bad?	如果是仪表板,是好是坏?
558	Like is it too basic or?	这太基本了还是?
559	A dashboard, again, my team's pretty,	仪表板,再次, 我的团队很漂亮,
560	you know, somewhat ahead of the curve,	你知道,有些在曲线前,
561	but all of my-	但我的所有的... ... . .
562	Mine was good to hear like that.	我很高兴听到这样的声音。
563	All of my marketers have built	我所有的推销员都造了
564	much more sophisticated things than dashboards.	比仪表板更精密的东西
565	Like what?	比如说?
566	Oh, I feel like, I don't know,	哦,我觉得,我不知道,
567	the other day, one of our,	那天 我们中的一个
568	someone on the marketing team	有人在营销团队
569	was talking about the tools that he had built.	他正在谈论他建造的工具。
570	When he's considering writing an article on something,	当他考虑写一篇关于某事的文章时,
571	it will crawl the web, find related work,	它会爬网 找到相关的工作
572	has a custom desktop app,	有自定义桌面应用程序,
573	actually build a desktop app that runs on his Mac,	实际建造一个运行在Mac上的桌面应用程序,
574	to highlight related articles for him.	为他突出相关文章.
575	Then he can chat to the whole system,	然后他可以和整个系统聊天
576	navigate, you know, the thing he's writing,	导航,你知道, 东西他写,
577	as well as the related work,	以及相关工作,
578	and he had a large dashboard for crawling the internet	他有一个巨大的仪表板 爬上网
579	to highlight to him exciting things that are popping up.	告诉他一些激动人心的事
580	Now, even on my team,	现在,即使在我的团队,
581	I think that marketplace is ahead of the curve.	我认为市场领先于曲线。
582	But that's great to hear.	但这是伟大的听到。
583	Any other interesting use cases	任何其他有趣的使用案例
584	that will inspire people to build something similar?	那会激励人们 建立类似的东西?
585	Let's see, maybe my finance team uses AI extensively.	让我们看看,也许我的金融团队 广泛使用AI。
586	So I think one of my CFOs realized that,	所以我认为我的首席财务干事意识到,
587	you know, her team was spending hours every week	你知道,她的团队 每周花几个小时
588	clicking through documents,	点击文档,
589	open this copy, paste this number here.	打开此副本, 在此粘贴此号码 。
590	And so she started building automation scripts	于是她开始制作自动化脚本
591	that runs on a routine,	运行在常规,
592	that automatically opens files, checks what's in there,	自动打开文件 检查里面有什么
593	checks for consistency, highlights her team	检查一致性,突出她的团队
594	if there's something they need to be paying attention to,	如果有什么需要他们注意的
595	if a new document has showed up.	如果新文档已经出现。
596	So I find that rather than waiting around	所以,我发现这 宁可等待周围
597	for an engineer to do the work for them,	一个工程师为他们工作,
598	the team's ability to kind of not just build dashboards,	团队的能力 那种不只是建造仪表板,
599	but build kind of a data management infrastructures.	但建立某种数据管理基础设施。
600	They can ingest data,	他们可以吸收数据,
601	alert them if something's happening.	发生什么时通知他们
602	I think my finance and marketing teams are doing that.	我认为我的金融和营销团队正在这样做。
603	Oh, my recruiting team.	哦,我的招募团队。
604	Well, we actually have recruiting engineers,	我们实际上已经招聘了工程师
605	which are really professional engineers	他们是真正的专业工程师
606	that sit in a recruiting team.	在一个招募团队。
607	They're the building very sophisticated tools for recruiting.	他们是大楼非常精密的招募工具。
608	And this is actually the other trend.	这是另一个趋势。
609	I think marketers, recruiters,	我认为市场商,招聘商,
610	HR professionals, ops people should all learn AI.	人力资源专业人士,行动人员都应该学习AI.
611	But the other thing is when you take an engineer	但另外一件事是当你接受工程师
612	and embed them in these teams,	把他们安置在这些队伍中,
613	then that further accelerates what you can do.	然后,这进一步加快了你能做的。
614	We do the same.	我们也是
615	We start with something basic, build it ourselves,	我们从基础开始,自己造
616	then we hit the wall and engineer comes in,	然后我们撞上了墙 工程师进来了
617	we build it further.	我们把它建设得更进一步。
618	Frankly, when you look at not just software engineers,	老实说,当你看到 不只是软件工程师,
619	but recruiting engineers, marketing engineers,	但招聘工程师,营销工程师,
620	HR engineers, I think there's so much valuable	HR工程师,我觉得有很多有价值的东西
621	engineering work that can now be done.	现在可以做的工程工作。
622	I'm just not worried about running out of...	我只是不担心...
623	Frankly, all my friends were so busy.	老实说,我的朋友都这么忙
624	We think, boy, how could we run out of engineering jobs?	我们想,孩子,我们怎么能用完工程工作?
625	Yeah, there are so many cool ideas you can experiment on,	是啊,有很多很酷的想法 你可以实验,
626	but you touched upon something	但你碰了个东西
627	that is actually one of the fears.	这其实就是其中之一的恐惧。
628	When we talk about like financial information,	当我们谈论像财务信息,
629	how much you're giving to AI.	你要给AI多少钱?
630	So I gave my perplexity permission	所以我才允许我感到困惑
631	to scan my fidelity account so it can track my portfolio,	扫描我的忠诚账户 这样它就可以追踪我的投资组合
632	tell me when to rebalance.	告诉我什么时候再平衡
633	It doesn't do anything on my behalf, but it has access.	它不代表我做任何事,但它有权限。
634	Do you think there is any problem with that?	你觉得有什么问题吗?
635	This is complicated.	这很复杂
636	I think AI and privacy is a complex area.	我认为AI和隐私是一个复杂的领域.
637	And it depends a lot on the company	这要靠公司
638	that you are sharing your data with.	您正在分享您的数据。
639	So for example, I trust all the hyperscalers	比如说,我信任所有的超标器
640	to really 100% follow their terms of service	百分之百遵守他们的服务条款
641	and to do what they say.	并按他们说的做。
642	My personal opinion, not giving legal business advice,	我个人的意见,不提供 法律业务建议,
643	but I'd be shocked if the largest hyperscalers	但是,如果最大的超标,我会震惊
644	publish the terms of service with some privacy notice	发布一些隐私通知的服务条款
645	and if they reach that.	如果他们到了那里。
646	Because that'd be not the culture,	因为那不是文化
647	it'd be so damaging of the long-term business model.	这会对长期商业模式造成如此大的破坏
648	Now that's on the largest hyperscaler side.	现在,这是最大的超规模。
649	If you look at the AI company side,	如果你看看AI公司那边
650	there's been at least one company that I won't name	至少有一家公司我不点名
651	that seems to occasionally change the terms of service.	这似乎偶尔会改变服务条件。
652	And if you're using it, you go to the website,	如果你用它,你去网站,
653	so you pop up, hey, we changed the terms of service	所以,你出现了,嘿, 我们改变了服务条件
654	to retain your data or train your data.	保留数据或训练数据。
655	And if you aren't paying attention	如果你不注意的话
656	and click the wrong button,	点击错误的按钮,
657	then they suddenly gave themselves permission	然后,他们忽然答应了。
658	to access your data	访问您的数据
659	in a way that I'm not that comfortable with.	我不太喜欢这样
660	I feel like I handle some sensitive information.	我觉得我处理一些敏感的信息。
661	So then I tend to be very careful with the businesses	所以,我总是非常小心 与企业
662	that I just don't feel the culture and the DNA.	我只是感觉不到 文化和DNA。
663	And frankly, the long-term business model	坦白说,长期商业模式
664	is as tied to protecting individual user privacy	与保护个人用户隐私一样
665	than the hyperscalers.	比超标者还要大
666	And I see businesses get this as well.	我认为企业也得到了这个。
667	例如,我的一个团队,AI Aspire,
668	we work with very large corporations,	我们和大公司合作
669	including banks with incredibly sensitive financial data.	包括拥有极其敏感金融数据的银行。
670	And as you can imagine, AI Aspire and our clients	你可以想象,AI Aspire和我们的客户
671	do not willingly share really sensitive,	不愿分享真正的敏感,
672	often material non-public information, right, NMPI,	经常提供非公开资料,对,NMPI,
673	with frontier labs without really careful thinking	与前沿实验室 没有真正仔细的思考
674	about the guardrails in privacy.	私底下的护栏
675	So I think it's complicated.	所以我觉得这很复杂。
676	So trusting hyperscalers,	如此信任超标者,
677	but also another thing that you can do,	但还有一件事你可以做,
678	you can download an open source model	您可以下载开源模式
679	and just run it on your computer	在你的电脑上运行
680	and then it just stays on your computer, right?	然后它只是留在你的电脑上,对不对?
681	Yes, I think, yes.	是的,我认为,是的。
682	It turns out a lot of banks will actually run the things	原来很多银行会管理这些东西
683	in the virtual private cloud or on-prem.	在虚拟的私密云中或prem上。
684	So they never even leave their control.	因此,他们甚至从来没有离开他们的控制。
685	But I think for individuals,	但我认为对于个人来说,
686	for the really sensitive things,	对于真正敏感的东西,
687	I sometimes run a local model.	我有时会经营本地模特儿
688	And it's been interesting with the open-way models.	并且对开通车型颇感兴趣.
689	So the latest open-way models	因此,最新的开放模式
690	are approaching frontier capability	正在接近边境能力
691	and that are actually small enough.	这其实足够小。
692	There are actually really good models now	现在真的有很好的模型了
693	that you can run on them.	你可以运行在他们身上。
694	Yeah, the one from Meta, right?	是啊,梅塔的那个,对吧?
695	The recent one.	最近那个
696	Oh yes, Metamuse is a good model.	哦,是的,Metamuse是一个很好的模型。
697	And I'm thinking also the latest version of Quen	我还在想最新版的Quen
698	is also very good.	也很不错。
699	But I think frankly, these models change every other week.	但坦率地说,这些模型每隔一周就改变一次。
700	So I think the best practice is to not get stuck on one,	所以我认为最好的做法 就是不要被困在其中
701	but to keep on trying new models.	但是要继续尝试新的模型。
702	So basically when there is a situation	所以,基本上当有情况
703	that you don't trust anyone,	你不信任任何人,
704	you run a local model	您运行本地模式
705	and this is how you keep your data safe.	这就是你保护数据安全的方法
706	I do trust the hyperscalers,	我确实相信那些超规模的人
707	but sometimes for literally NMPI,	但有时对NMPI来说
708	material non-public information,	* 非公开资料,
709	that I won't even send to that.	我甚至不会送去那个。
710	I just can't even send that to the cloud.	我甚至不能把那个寄到云里
711	So that, I'll either do it manually without AI help	所以,我要么手动做,没有AI帮助
712	or if I really need to use AI,	或者,如果我真的需要使用AI,
713	then really carefully only use a local model.	然后非常小心地使用一个本地模型。
714	Interesting.	有意思,这很有趣
715	Okay, this is an interesting one.	好吧,这是一个有趣的。
716	Okay, what about loss of human control over AI?	好吧,那人类失去对AI的控制呢?
717	因为我和Yoshua Benja谈过了
718	who is very negative when it comes to free AI	在自由AI方面,谁是非常消极的
719	without any regulation.	没有规定。
720	And he painted me some very scary pictures	他给我画了一些非常恐怖的照片
721	of AI taking over control because we,	由于我们,
722	basically the whole scenario is	基本上整个情景
723	we can't control something that's smarter than us.	我们不能控制比我们聪明的东西
724	And if AI gets smarter and smarter,	如果AI变聪明了
725	where do we end up?	我们到哪里去?
726	What do you think about that?	你觉得怎么样?
727	I think about something else that we can't control,	我想到别的东西,我们不能控制,
728	which is airplanes.	这是飞机。
729	No one can build an airplane	没人能造飞机
730	that you can fly perfectly.	你可以完美飞行。
731	Winds will buff it around.	风会吹动它周围。
732	And then candidly,	然后坦率地说,
733	in the early days of developing airplanes,	在发展飞机的初期,
734	some airplanes crashed and people died	有些飞机坠毁,有人丧生
735	and it was tragic and awful.	悲剧和可怕。
736	But through the early lessons learned,	但通过早期的教训,
737	we then learned to control airplanes better and better	然后我们学会了更好控制飞机
738	so that today we can mostly get in an airplane	这样今天我们就可以乘飞机了
739	and not fear too much for lies.	不要害怕太多的谎言。
740	And it's really like that too with AI.	对AI来说也是这样
741	No one can perfectly control AI	没有人能完全控制AI
742	because it generates tokens or outputs	因为它会产生符号或输出
743	that are a little bit random.	这有点随机。
744	So we don't really know what exactly it'll do.	所以我们不知道它到底能做什么。
745	But as we run them,	但当我们运行他们,
746	and there's been a small number of mishaps,	发生了少量的意外
747	which is unfortunate.	这是不幸的。
748	And some number of mishaps have done some real damage.	也有一些灾祸造成了一些真正的破坏.
749	But the way we engineer almost any system	但是我们设计系统的方式
750	from an airplane to electric circuits to now AI	从飞机到电路到现在的AI
751	is carefully grow their capabilities	仔细地提高他们的能力
752	so that we can have a controlled environment	以便我们有一个控制的环境
753	in which to measure what's wrong	用来衡量什么是错的
754	and then to shape it,	然后造出来,
755	to make sure we can control it well enough	确保我们能很好地控制它
756	that it behaves responsibly and safely.	它的行为是负责和安全的。
757	And to this day,	时至今日,
758	we can't perfectly control any airplane.	我们控制不了飞机
759	And we will never perfectly control AI either.	我们也不会完全控制AI
760	But I think we are certainly controlling them well enough	但我认为我们控制得够好了
761	that this loss of control doesn't feel like science,	这种失控的感觉不像是科学
762	if it's a science fiction.	如果这是个科幻小说
763	Yeah, yeah.	对 对
764	What about deep fakes?	那深层假货呢?
765	Deep fakes are a problem.	深假是烦恼.
766	Well, one of the most disgusting things	最恶心的事情之一
767	I've ever seen or heard of	我见过或听说过
768	is non-consensual, intimate deep fake imagery.	是未经同意的 亲密的深层假象
769	I'm really glad that, you know,	我真的很高兴,你知道,
770	U.S. Congress has been moving to, right,	美国国会已经向着
771	let's pass laws, get rid of that.	让我们通过法律,摆脱它。
772	Penalties for that.	惩罚
773	I'm just, I think there's some really problematic uses	我只是,我觉得 有一些真正的问题 使用
774	of AI that we should outlaw, heavily penalize.	我们应严加惩罚,
775	Let's just get rid of that.	先释此义.
776	What do you think about children and social connection	你对孩子和社会联系有什么看法?
777	when it comes to AI, with kids using more of AI?	当涉及到AI, 孩子们使用更多的AI?
778	Because we've seen social media,	因为我们看过社交媒体
779	how, you know, there are people	怎么会有人
780	who are dumb-scrolling all day.	他们整天都是哑巴。
781	And my daughter, who is five years old now,	还有我女儿 她现在五岁了
782	whenever I don't have an answer,	当我没有答案时
783	is like, ask Chad GPT.	就像问查德GPT
784	Like, who's that person?	喜欢,是谁的人?
785	I'm like, I don't know.	我很喜欢,我不知道。
786	问查德GPT.
787	Like, she thinks Chad GPT knows everything.	她觉得Chad GPT什么都知道
788	What would you say about, you know,	你会怎么说,你知道,
789	kids' future with AI?	孩子们的未来和AI?
790	First, I think kids have a bright future.	首先,我觉得孩子有一个光明的未来。
791	It's such an exciting time to be a child,	小时候真是激动人心
792	to grow up in this environment with tools	在这种环境中用工具成长
793	that none of us ever had before.	我们从来没有有过。
794	At the same time, we've seen that social media,	同时,我们看到社交媒体,
795	I think social media has probably been blamed	我觉得社交媒体可能已经被指责了
796	a bit more than it deserves, but it does deserve blame.	比它应得的还要多,但它的确应该受到指责。
797	Has kind of not been great for kids.	有点对孩子不好
798	I actually worry a lot about, it's a wonderful tool,	其实我担心的很多, 这是一个很好的工具,
799	but AI damaging learning is something I worry a lot about.	但是,AI破坏的学习 是我担心的很多。
800	So it turns out, I have a five-year-old and a seven-year-old.	因此,结果,我有一个5岁和7岁的孩子。
801	When I teach them math, they're so young enough	当我教他们数学时,他们就足够年轻了
802	that I can basically, you know,	我可以基本上, 你知道,
803	not let them use the calculator.	不让他们使用计算器。
804	Can say, how do you multiply these numbers?	可以说,你是如何乘以这些数字的?
805	And I don't give them a calculator	我不给他们一个计算器
806	and practice that with them.	和他们一起练习。
807	But as they're a little bit older,	但是,由于他们有点老,
808	I worry a lot about students using cognitive offloading	我非常担心学生使用认知卸载
809	to AI in a way that damages the long-term learning retention.	以损害长期学习保留的方式向大赦国际提出。
810	But then at the same time, oh, I actually built an app.	但与此同时,哦,我实际上造了一个应用程序。
811	I did not like any of the, you know,	我不喜欢任何,你知道,
812	free online learning to type types of things.	免费在线学习,以类型方式进行。
813	So I actually built my own to have my daughter learn to type.	所以我自己造的,让我的女儿学会打字。
814	And I'm hoping that she's actually getting pretty decent now	我希望她现在变得相当体面
815	for a seven-year-old.	给一个七岁的孩子。
816	Oh, so she's typing already?	哦,所以她已经打字了?
817	Oh yeah, she actually typed all the lowercase letters.	哦,是的,她实际上打 所有小写字母。
818	She's a little bit, you know, not,	她有点,你知道,不是,
819	her shift uppercase letters a little bit, not quite there.	她的变身大写字母一点点, 并不完全在那里。
820	But I think that this unlocks, you know,	但我认为这个解锁,你知道,
821	responsible adult-supervised use of online tools.	由成年人负责监督在线工具的使用。
822	And I think it's really tricky, you know?	我觉得这很棘手 你知道吗?
823	I think adult-supervised use of digital tools	我认为是成年人监督使用的数字工具
824	seems a great thing for kids,	似乎对孩子们来说是件好事
825	but too many adults don't have time	但太多的成年人没有时间
826	to supervise the use of the tools.	监督工具的使用。
827	And then the incentives of, say, social media, right?	然后是社交媒体的激励,对吧?
828	To do funny things.	做一些有趣的事情。 做一些有趣的事情。
829	Yeah, has to be the right incentive when it comes to AI.	对,对AI来说,应该有正确的激励
830	Okay, you mentioned, we talked about the fears.	好吧,你提到,我们谈论了恐惧。
831	We talked about how you can improve your work with AI.	我们谈过如何改善你和AI的工作
832	Can you name some of the biggest opportunities in AI in 2026	你能说说2026年AI里最大的机会吗?
833	for people who want to build?	给那些想建楼的人?
834	For an individual that wants to build,	对于一个想要建设的人来说,
835	I don't think it's one size fits all.	我不认为它是一个尺寸适合所有。
836	But because the cost of building has plummeted,	但由于建筑成本暴跌,
837	I encourage people to learn AI, build fast,	我鼓励人们学习人工智能 快速建设
838	and talk to customers.	和客户交谈。
839	I find myself building things, I don't know,	我发现自己建造的东西, 我不知道,
840	every week, every weekend,	每个礼拜,每个周末,
841	because I or someone on our team will have some problem	因为我或我们团队中有人会有问题
842	and I have some idea for building some AI thing	我有点想法 建造一些AI的东西
843	to automate it.	把它自动化。
844	Last weekend, I had, really, I was using a Frontier model	上个周末,我真的在用边疆模型
845	to analyze a lot of our key business metrics	来分析我们许多关键的商业标准
846	because I didn't have time to do it myself.	因为我没有时间自己做
847	But it was kind of measuring, you know,	但它是一种测量,你知道,
848	deviantized key business metrics,	偏离关键商业标准,
849	and I didn't have time to go find a data scientist	我没有时间去找数据科学家
850	to go work me on it,	去工作我做到这一点,
851	so I just did a variety of Frontier models,	所以我刚刚做了各种边疆模型,
852	being really careful on their data retention policies.	对数据保存政策非常小心
853	I did not use models with data retention policies	我没有使用有数据保存政策的模型
854	that I don't like in order to analyze data.	我不喜欢分析数据
855	But, and then I find that what's happened with AI	但是,然后我发现 发生了什么事AI
856	is the cost of building has plummeted,	建筑成本暴跌,
857	and so the challenge is shifting to deciding what to build,	因此,挑战正在转向决定建设什么,
858	which I've been calling	我一直在打电话
859	the product management bottleneck.	产品管理瓶颈。
860	And so people, founders, engineers, product managers,	于是人们,创始人,工程师,产品经理,
861	they can talk to customers,	他们可以和顾客交谈,
862	get a sense for the taste of judgment on what to build,	体会到判断如何构建的滋味
863	and then build with AI and iterate quickly.	然后用人工智能快速建造
864	I think there's just a ton of exciting things to do.	我觉得只有一吨刺激的事情要做。
865	And you've been starting so many companies.	你开了这么多公司
866	You're like, when I looked at your portfolio,	你就像,当我看着你的组合,
867	do you think for beginners,	你认为初学者,
868	when you said you built something during the weekend,	你说你在周末造东西的时候
869	how do you decide what to focus on	如何决定什么是重点
870	or you can pursue multiple ideas because of AI now,	或者你可以追求多种想法 因为AI现在,
871	and you can just be playing in different companies	你可以在不同的公司里玩
872	at the same time?	同时呢?
873	It turns out building a company is still really, really hard.	事实证明,建立一家公司仍然非常困难。
874	And so there's a lot to be said	所以有很多话要说
875	for a single-threaded leadership	单线领导
876	or someone that's fully focused on just one thing.	或者某个完全专注于一件事的人
877	I find that over a weekend,	我周末发现,
878	I can often build an own wrapper, build a simple application,	我常常可以自己做包装 做简单的应用
879	but I wish it was that easy to build a large company.	但我希望有那么容易 建立一个大公司。
880	I find that building something meaningful	我发现建造一些有意义的东西
881	often takes either real technical depth	往往需要真正的技术深度
882	and or deep customer insight and integration of customers.	或客户的深刻见解和融入。
883	And yes, we can now use AI to code something	是的,我们现在可以用AI 来编码一些东西
884	in a few hours, but that's a small piece of the puzzle.	几小时后,但这是谜题的小作品。
885	So spending time understanding the technical complexity	所以花时间了解技术的复杂性
886	and building the really complex software	并建造非常复杂的软件
887	that takes us like months, maybe years,	我们花了几个月,也许几年,
888	or having that deep customer insight	或者是客户的深刻见解
889	to decide what to build, that also just takes a lot.	决定建造什么, 这同样需要很多。
890	Talking to people, reading facial expressions, surveys,	与人们交谈,阅读面部表情,调查,
891	it's doing that over and over	它的做 一遍又一遍
892	until we figure out what to build.	直到我们想出 建造什么。
893	And so I think sometimes there's a lot of value	所以我觉得有时有很多价值
894	to sampling widely, but then having that focus	广泛采样,但随后有重点
895	for an individual to go really deep in a couple of sectors,	一个人真正深入到 几个部门,
896	that still seems important for building a business.	这对建立企业来说仍然很重要。
897	My last question, I know we don't have much time,	我最后一个问题,我知道我们时间不多了
898	but I wanted to ask you about AGI,	但我想问你关于AGI的事
899	just because people use this word so much.	只是因为人们使用这个词这么多。
900	And some people say, I think Jensen Huang said,	有人说 黄詹森说
901	we already reached AGI.	我们已经到达AGI。
902	You said it's decades away.	你说过还有几十年
903	What's the one criteria when you're gonna say	当你要说,一个标准是什么?
904	we reach AGI?	我们找到AGI了吗?
905	So different people say we reach AGI at different times	不同的人说我们在不同的时间到达AGI
906	because of different definitions of AGI.	因为AGI的定义不同.
907	The definition I'm most familiar with	我最熟悉的定义
908	is AI that could do any intellectual tasks	是AI,可以做任何智力任务
909	that a human can.	人类可以。
910	But so the human brain can take, say, five years	但人类大脑需要五年时间
911	to study and do a PhD thesis.	学习和做博士论文。
912	And so can AI write a PhD thesis?	AI也能写博士论文吗?
913	Or a human can learn to drive a truck	或者人类可以学会开车
914	through a dense rainforest	穿过茂密的雨林
915	with tens of minutes of practice.	和数十分钟的练习。
916	So when can AI do that?	那么AI什么时候能做到这一点?
917	To drive a new environment	推动新的环境
918	with tens of minutes of practice.	和数十分钟的练习。
919	It feels like there's a long list of these things	感觉有一大堆这些东西
920	that AI cannot do for what feels to me decades.	对我而言,大赦国际不能做几十年的感受。
921	I hope it's only decades.	惟愿数十载.
922	Maybe it would turn out to be longer.	也许它会变成更长。
923	So that's what I think for that definition of AI or AGI.	所以这就是我对AI或AGI的定义的看法.
924	AGI还很远
925	But it turns out because of economic incentives,	但结果却因为经济刺激
926	I think open and Microsoft had an agreement.	我认为开放和微软有协议。
927	There's actually been renegotiated now,	已经重新谈判过了
928	so that's gone away.	所以这已经消失了。
929	但OpenAI有经济刺激
930	to try to declare reaching AGI earlier.	试图更早地宣布到达AGI。
931	And so it turns out that if you come up	结果发现如果你上来
932	with other definitions of AI,	与大赦国际的其他定义,
933	depending on how far you lower the bar,	看你把酒吧放低多远
934	then you could totally have reached AGI already	那你就完全可以联系到AGI了
935	or even 30 years ago,	甚至在30年前,
936	depending on how you want to define it.	取决于您要如何定义它。
937	Yeah, true.	是的,没错。
938	Andrew, thank you so much for this positive conversation.	安德鲁 非常感谢你这么积极的谈话
939	Very applicable.	非常适用。
940	I like when you watch something	我喜欢你看东西
941	and then you go and you measure yourself	然后你去测量自己
942	against what people are doing with AI,	与人们对AI的所做所为相对应,
943	look at your process and maybe expand it.	看看你的过程,也许可以扩展它。
944	So thank you so much for showing what your team is doing	所以,非常感谢你展示 你的团队在做什么
945	and thank you for your insights.	谢谢你的洞察力
946	Yeah, I think given the huge benefits of AI to come,	是啊,我认为考虑到AI的巨大好处 来,
947	I hope whoever is watching this is motivated	希望看这个的人有动机
948	to really go learn AI, apply it,	真正去学习AI,应用它,
949	and even to go build some things.	甚至去制造一些东西。

该视频共有字幕 949 条。解锁更多字幕为会员功能，请移动到 价格