Andrew Ng: The Biggest Opportunities in AI Aren't Where You Think

0:00There's been a lot of misinformation

0:01about AI.

0:03>> This is Andrew. He co-founded Google

0:05Brain and Corsera. His machine learning

0:07course has reached millions of learners

0:09and he is one of the most influential

0:11voices in AI today.

0:13>> So, a handful of leading AI companies

0:14have been very loud voices of fear

0:16monongering around AI to try to get

0:19regulations passed. This drum beat of

0:21fear-based messaging has skewed [music]

0:23societal perception to be really

0:25negative on AI. People talk to me about

0:28data centers and job loss.

0:30>> Maybe AI could do 30 40% of many jobs.

0:33And what that means is well that 60%

0:35that the human does has become even more

0:38valuable.

0:39>> What about loss of human control over

0:41AI?

0:42>> I think about something else that we

0:44can't control.

0:46I think you're one of the voices in AI

0:49who comes with a huge background in

0:51machine learning and teaching AI and

0:54also you're a positive voice because

0:55this is something that I've been seeing

0:58especially this summer how poor the

1:00society has become especially on social

1:02media when I'm posting about AI and

1:04people talk to me about data centers and

1:07job loss. Why do you think this wave

1:10started recently? What do you think the

1:12causes are? There's been a lot of

1:14misinformation about AI and the root

1:17cause of a lot of this is an unfortunate

1:20attempt that started two three years ago

1:22of I think PR and regret capture. It

1:25turns out that the most one of the most

1:27valuable things in AI right now is the

1:29giant AI models, giant large language

1:31models that some have trained. But if

1:34you spend billions of dollars training a

1:35model is really inconvenient if someone

1:37else trains a model and wants to give it

1:39to anyone in the world to use for free.

1:42So, a handful of leading AI companies, I

1:44think, as you know, have been very loud

1:46voices, fear-mongering around AI to try

1:49to get regulations passed to create an

1:51unfair playing field that favors

1:54incumbents so that we all have to pay a

1:57high toll for use of AI while stying the

2:01other teams, be it researchers or other

2:03companies that want to just give away

2:05open way to open source models that

2:07anyone could use much cheaper.

2:08Unfortunately, fear-mongering works. Um,

2:10when you go and say AI is like nuclear

2:13weapons, which is an analogy that has no

2:15basis in fact, what do they have to even

2:17do with each other? Or when you go

2:19around and cherrypick cases of AI, you

2:22know, making a misstep and making it

2:23much bigger than it is or even spread

2:26misinformation about how AI uses data

2:29centers, uses a lot more water than the

2:32actual reality. This drum beat of

2:34fear-based messaging has skewed societal

2:37perception to be really negative on AI,

2:40which is unfortunate because this is

2:42slowing down American adoption in AI.

2:44This is making America less competitive.

2:47Um, and unless we get the truth about AI

2:50out there, which is that it's fantastic

2:52benefit with some problems, but not

2:54nearly the TV which they're blown up to

2:56be, it will um will hurt individuals.

2:58>> I'm going to read out some of the

3:00problems that people are highlighting.

3:01Job loss and inequality. What do you

3:03think?

3:04>> The job apocalypse or job apocalypse,

3:06this idea that AI will take over 50% of

3:09jobs, people will be out of work, riding

3:11in the streets, that's just not going to

3:13happen. With every wave of technology,

3:15including AI, [gasps]

3:17the skills we need to do great work

3:20shifts. And so AI is changing job

3:22professions. But boy, I wish AI were AI

3:26just doesn't work well enough. I know

3:27that handful of businesses want to hype

3:29up AI to say we have super intelligence

3:32or we have artificial general

3:33intelligence or whatever and can do all

3:35the stuff that humans do. I wish AI work

3:38better. We're just not good enough to

3:39make AI do everything a human does. And

3:41if you look at the analysis of jobs,

3:44economists uh like my friend Eric

3:46Brennoffson at Stanford, Andy McAfee at

3:48MIT, um economists have analyzed many

3:52people's jobs and by break it down into

3:54individual tasks and maybe AI could do

3:57you know 30 40% of many jobs and what

4:00that means is well that 60% that a human

4:03does has become even more valuable

4:05because it's called an economic

4:07complement to the 30 40% that's now

4:09cheaper and So what will happen is

4:13people that use AI, maybe people that

4:16use AI will replace people that don't

4:18use AI, but AI is not in a position for

4:21the vast majority of jobs to replace

4:24people. Of all the different

4:25professions, the one that's most

4:27affected by AI now is software

4:28engineering because AI is actually

4:30fantastic at writing code. And

4:32[clears throat] what we see is that the

4:33number of job openings in software

4:35engineering is up contrary to what you

4:38know the doom fear-mongerers would say

4:41right AI is not actually able to replace

4:43software engineers and all the good

4:45software engineers I know are busier

4:47than ever now the flip side of it is if

4:50someone still write code like is 2022

4:53before chai GPT they're in trouble they

4:55need new skills like don't don't do

4:57stuff that that 30 40% AI can automate

5:00you got to stop doing that let AI do

5:01that but then gain your skills to do the

5:04other 60 70% that AI cannot do.

5:07>> What would your advice be to new

5:09graduates? Cuz I talked to Eric um on

5:11this podcast and uh he was talking about

5:14that there's not really a lot of impact

5:16on the job market except for I think he

5:19mentioned people from 18 to 25 who just

5:21graduated. What would be your advice to

5:23those people who don't have the

5:25expertise maybe to strategize in their

5:27job yet? uh they can only do manual work

5:30that AI can do as well.

5:31>> So one real challenge um uh for fresh

5:34college grads is that the university

5:37system is slow to adapt and so um you

5:42know I love academia. I think we should

5:44all support academia and universities.

5:46And when AI comes and transforms the way

5:49software is written, universities often

5:51take like a year or two for the faculty

5:53to master skills, then create new

5:56courses, get curriculum committee

5:57approval, whatever, get the faculty

5:59senate to vote. It just takes years. And

6:02that speed of change in academia is very

6:05poorly matched to the speed of change in

6:06AI. So sadly, many universities are

6:10still teaching students to be ready for

6:12the jobs of 2022 when we shouldn't even

6:15be teaching them for the jobs of 2026.

6:17We should be teaching them for the jobs

6:18of 2028 and beyond. And what this means

6:22is the job openings are there. Uh tons

6:25of employees I know just can't find

6:26enough skilled, you know, uh people at

6:29any level of seniority. But um it turns

6:32out in my office right now, we have a

6:34lot of interns. There are current

6:35college students, fresh college grad. We

6:37also had one high school intern and

6:39they're amazing and productive. But the

6:40key is they're all very AI native. They

6:42all use AI tools to do the things AI

6:45could do, but then also, you know, lean

6:46in to doing the things that uh humans

6:49can do that AI can't for the for for a

6:51long time. So there's plenty of work for

6:52people to do. But so my advice to fresh

6:55college grads or the people currently in

6:57college is um by all means work hard in

7:00classes. you know, get good grades,

7:02learn from the instructors, but to the

7:04extent that there's still additional

7:06skills that the university has not yet

7:08adapted to teaching, then find other

7:10ways to learn online, be it from Corsera

7:13or Deep Learning.AI or Udemy or other

7:16places where you can gain the more

7:18cutting edge skills, especially AI

7:20skills that um universities have not yet

7:23worked in the curriculara.

7:24>> Quick pause because what Andrew just

7:26said about workflows connects to

7:27something that HubSpot just put out for

7:29free. The thing is the real opportunity

7:32right now is not just in the models

7:33themselves, but in what you build on top

7:36of them. How you turn AI from a chat

7:38window into workflows that actually do

7:41work for you. And the good news is you

7:43don't have to start by building a full

7:44agent. You can start much simpler with

7:47better prompts. HubSpot just released

7:49the advanced chat GBT prompt engineering

7:52playbook. And the idea is very simple.

7:55in 7 days. It helps you move from

7:57getting generic AI answers to building

7:59prompts that give you much more

8:01consistent and useful options because

8:03we've all had this moment. You type in

8:05one vague sentence, get something

8:07mediocre back, and then spend the next

8:1020 minutes fixing it yourself. This

8:12playbook is basically built for that

8:14exact problem, and it walks you through

8:16a single progression. First, you learn

8:18how to structure prompts with things

8:20like role, context, format, and

8:22parameters. Then it gets more advanced.

8:25Few shot examples, chain of thought

8:27reasoning, more precise prompt

8:29structures, and ways to make outputs

8:30more reliable. What I like is that it

8:32doesn't stop at individual prompts. It

8:34also shows you how to build reusable AI

8:36personas, modular prompt components, and

8:39eventually your own signature prompt

8:41engineering system. So instead of

8:43starting from scratch every time, you're

8:45building a repeatable way to work with

8:47AI. Whether you're creating content,

8:49analyzing data, or making business

8:51decisions, the playbook is free. Links

8:53in the description. Thanks to Hopspot

8:55for sponsoring this video. And now back

8:58to our conversation with Andrew.

9:00>> I want to say one other thing. Um it

9:03turns out one if if look at the skill

9:06map changes. One of the most important

9:08changes is um it's so much easier to

9:11build with AI than before. When

9:13something becomes much easier, a lot

9:15more people should do it. And so now,

9:19not only should professional software

9:20engineers build software with AI, it's

9:22becoming much easier for everyone to

9:24build with AI and people that embrace

9:27that and do so will be more productive

9:30and will accomplish more and I think

9:31have more fun than the ones that don't.

9:33And AI lets you build really fast. So

9:36for people that not just software

9:38engineers but you know marketers,

9:40recruiters, uh HR professionals,

9:42operations specialists, I think if they

9:44learn to build with AI, uh they'll

9:47really just do much more whatever their

9:49job row is.

9:50>> How do you by the way measure uh the

9:52increase in productivity uh when you

9:54deploy AI? Do you have like a KPI in

9:57your company?

9:58>> I wish a simple answer. I find that the

10:01business outcome of AI is more a

10:03function of the business than a function

10:05of the AI. For for some it may be um

10:09increase you know uh customer growth and

10:11retention or maybe faster to serve

10:13customers or increase accuracy in some

10:15tasks. So the KPIs tend to be related to

10:17the business rather than the AI. M so

10:19you can't like directly measure just AI

10:21because it's it's an interesting thing

10:22to do because we've been deploying AI

10:24actively in my company and I think for

10:25me as a media company it's probably the

10:27amount of views the output [snorts] uh

10:29it's just interesting how yeah it's just

10:31interesting how different people measure

10:32even revenue like if you're becoming

10:34more effective

10:35>> uh with um how you make money

10:37>> actually same how are you using AI in

10:39your business

10:39>> oh my god I have the so first of all we

10:42have claude for all of us and we have

10:44certain projects for every social media

10:46that we're on so for example for this

10:48podcast. We [snorts] have a project

10:49that's called guests and it knows all

10:52the analytics from previous guests and

10:54it has certain criteria on which we rank

10:56every single person who comes to the

10:57podcast whether he's he or she's cited

11:01whether they have a certain opinion on

11:03AI whether they've been active with AI

11:05in their company or if they're a recent

11:07founder in AI. So it gives them

11:09different weights and it comes up with a

11:11grade based out of 40. 40 meaning tier

11:15one, 30 meaning tier three, etc. And

11:17then we have another one that analyzes

11:20every single podcast and gives me tips

11:21on how to ask questions.

11:23>> Oh wow.

11:24>> Same for Instagram, same for LinkedIn.

11:26It has my tone of voice, personal

11:28dossier, my business strategy. So it

11:30whenever it writes something, it knows

11:32all the facts about me, how I sound.

11:34Every social media is run by a person.

11:36So a person makes a strategic call and

11:38by the way if you can give me feedback

11:40on this if I can improve. So what I'm

11:42working on right now is closing the loop

11:44because sometimes they send me a text.

11:46I'm like oh we need to change this this

11:48and that. But that happens in a chat in

11:50Telegram and we have this feedback. We

11:53we have a bot that scans all of our

11:54chats. But I really want AI to be able

11:58to learn continuously from this feedback

12:00to just know my taste better. There's

12:02one thing I see a lot in AI which is um

12:04it turns out for AI as data scientists

12:06or AI brainstorming partner it often

12:09comes up with you know one or two good

12:10ideas two or three mediocre ones and

12:12like you know four atrocious ones and

12:16sometimes you wonder how could my AI

12:18have thought you know like that could

12:19even be a plausible idea and to me this

12:22relates to the job apocalypse point of

12:24view which is that for a long time

12:27humans you me everyone watching this

12:30will have a significant context

12:32advantage over AI, which is that you

12:34know something that's incredibly obvious

12:36to you that you know that was an awful

12:37idea but the AI did not and it turns out

12:40that one of the reasons why AI will not

12:43replace our jobs or whatever of large

12:45business anytime soon is because humans

12:48have a massive context advantage

12:51compared to AI. We know so much that you

12:54know from our years of experience that

12:55we talked to customer we saw the funny

12:57facial expression that told us ah they

12:59don't like this or we talked to business

13:01or you know our manager said hey blah

13:03blah blah I really care about this and

13:05so it turns out that almost all humans

13:07well maybe all humans just know a lot of

13:10stuff that the plumbing does not exist

13:12and I don't think exists for the

13:13foreseeable future for AI to get I know

13:16sometimes people talk about the

13:18importance of human judgment or human

13:20taste and Some people wonder all right

13:23what is taste is this fuzzy thing but to

13:25me the technical thing that underlies

13:27why humans have better judgment and

13:29better taste than AI is this context

13:31advantage and because this is a

13:33long-term advantage like no one's going

13:35to solve this you know in a few years

13:37this is why we just need a lot more

13:39humans with that judgment and taste to

13:41keep on complimenting AI

13:43>> and doesn't this make education even

13:45more important because education gives

13:47us context because it's another another

13:48thing I'm hearing about AI like you

13:50won't need education because all the

13:51information is at your fingertips. You

13:52just ask Chad GPT. But when you say

13:55context and taste, for me, that's years

13:58of acquiring knowledge and learning from

14:00the best and seeing how they perform

14:02versus just asking a chat.

14:04>> I'm going to say something that may be

14:06controversial. I don't know I've said

14:07this publicly, but I think it's true,

14:09which is frankly AI models are terrible

14:12for learning. Um, I know people think

14:17AI is AI is wonderful at getting things

14:20done. use it all the time, love it. But

14:23all the data that's coming out is that

14:25when say college students use AI, we

14:28know this. It's just a study now back up

14:30as well. So we also have numbers. But

14:32the data is very clear. Students score

14:34higher on homeworks when they use AI.

14:37Yay, higher homework scores. But

14:39retention, their long-term performance

14:41is much worse because their AI do the

14:43work for them. more and more studies are

14:44coming out to back this up now that I

14:48think people think oh it turns out you

14:50know I think Wikipedia is a wonderful

14:52tool has tons of facts web search is a

14:55wonderful tool has tons of facts but it

14:57turns out that when you ask AI to do

15:00work for you you're cognitive offloading

15:04to AI which is great because that's how

15:06society moves forward and gets work done

15:08but human retention is much worse it's

15:11just so clear that LMS MS as they are

15:14most commonly used are terrible for

15:16learning. I'm not saying there's no way

15:17to use it in a way that is good for

15:19learning. I think there are ways to use

15:20that good for learning but even for

15:22myself there's so many things on the AI

15:24model over the last you know 6 months or

15:25whatever like I don't know [gasps]

15:27building some project how does this

15:28front end backend component work

15:30whatever give me the answer get the job

15:32done it was fantastic but 6 months later

15:34I don't remember the answer when I need

15:36to redo that front end backend component

15:39I ask AI again so data is really clear

15:42we should stop thinking of AI as helpful

15:45for learning at least the vast majority

15:48of ways that the vast majority majority

15:49of your people are using AI models

15:51today. It's absolutely terrible for

15:52learning.

15:53>> But you're building a company helping

15:54solve that, right? Because the one

15:56onetoone tutoring with AI is that where

15:58you just announced with a 100 million

16:00investment from Corsera.

16:02>> Yes. So I'm excited about leading a new

16:04organization called Learn Vector that is

16:07focused on um building new learning

16:09experiences that is much more onetoone

16:12than one to many. So you know 15 years

16:14ago I I was privileged to participate in

16:17the online courses movement that I think

16:20changed the way a lot of people learn

16:22but that was and still remains largely a

16:25one to many experience where everyone

16:27you know kind of watches the same video

16:28which is actually okay it actually works

16:30well but the technology now exists to

16:32create much more personalized customized

16:35onetoone experiences and so our team is

16:38working hard on that I think we'll have

16:39a lot more to show by early next year

16:42when think about human skill

16:43development. I I feel like because AI

16:45has so heavily impacted software

16:47engineering, um what we see happening in

16:50the job market for software engineering

16:52is a harbinger as a forerunner of what

16:54we'll see in other disciplines as well.

16:56And in software engineering, um people

16:59need to learn new skills, but when they

17:01do, they are thriving and creating more

17:03value and frankly getting raises and

17:05doing even more exciting projects. And

17:07what I've seen the early signs of in

17:09other disciplines as well for example in

17:11software engineering you know most

17:12developers like front end backend

17:14developers have now become full stack

17:16developers because of AI hub you could

17:19take on broader scope I'm seeing early

17:21signs of this in other disciplines as

17:23well where for example someone that in

17:25marketing that did marketing

17:26coordination uh coordinating marketing

17:28campaigns with AI help they can now

17:30become more of a full cycle marketing

17:32take on a broader scope and I'm seeing

17:35you know frankly sources in recruiting

17:37become more full cycle do endto-end

17:39recruiting. So now the good news and bad

17:42news is for people to step up to these

17:43broader roles. You do need to learn AI

17:45skills but also it's not just learning

17:47AI you also need to learn these other

17:49skills uh like how do you do the other

17:50parts of marketing of recruiting or

17:52software engineering or AI engineering.

17:54So I think this actually creates a heavy

17:56need, a big need for people to gain new

17:58skills. But when they do, which is both

18:00AI skills, but also disciplinary skills,

18:03then they can do much more, hopefully

18:05have more fun, work on more exciting

18:08projects, hopefully get paid more as

18:10well. And one reason I kind of worry

18:12about the fear mongering is um I got an

18:14email from someone that was about to

18:16enter college and you know he emailed me

18:18saying hey Andrew taking online courses

18:21but I'm really struggling with what I

18:22should major in college because in four

18:24years won't AI do all this and

18:26everything I learn will be obsolete and

18:28the answer is no of course it won't all

18:30be obsolete but when we keep on pushing

18:32these fear messages uh we make people

18:36wonder if they will even be relevant and

18:38it makes people not lean in to gain

18:41these skills, they'll put them in much

18:43better position. So, I see very clearly

18:45that these fear-mongering messages are

18:48distorting how many people, including,

18:51you know, high school students, college

18:52students, fresh grads, think about the

18:54economy. And frankly, making people give

18:57up is one of the worst things we'll be

18:58doing in this era when people that lean

19:01in will thrive.

19:03>> Andrew has taught over 8 million people

19:05AI. He started teaching machine learning

19:08online back in 2011, years before the

19:11current AI boom. Now, one of the

19:13companies he's building is focused on AI

19:15agents. From the way that it sounds, it

19:17can still feel way too technical. So, I

19:19put together a step-by-step guide to

19:21building your first AI agent with no

19:23coding required. It walks you through

19:25what to automate, how to set it up, and

19:27how to make it actually useful. It is in

19:29my newsletter this week. The newsletter

19:31is called Future Proof. It's free. Link

19:33is in the description. What would you

19:36reply back to that email that somebody

19:39sent you? What would you say is the best

19:40major to study now to thrive in AI era?

19:42Do you think it's like going deep into a

19:45niche or just broader computer science

19:47so that you can acquire AI skills really

19:49fast?

19:50>> You know, I don't know what's the best

19:51major. There are awful lot of great

19:52majors. It's is like um I kind of feel

19:55like what's the best job in the world is

19:57like what's the best major in the world?

19:58>> Oh, something that you love, right?

20:00>> Yeah. My daughter wants to be an

20:01astronaut. I don't know if she can major

20:02in becoming an astronaut. I have to

20:04think about that. When she get older

20:05though, she may change your mind. I see

20:06so many opportunities um across across

20:09so many job roles. It all seems very

20:11exciting to me. But do learn AI, do

20:13learn to build with AI. The other thing

20:15that my team's been working on AI

20:17engineing skills map to try to map out,

20:19you know, the most important skills for

20:20AI engineering. One thing that I felt

20:22intuitively, but I was surprised to see

20:24it show up in the data was that a lot

20:26more job descriptions seem to be saying

20:28they want people that demonstrate a very

20:30high sense of agency. Because it turns

20:32out with AI there a lot more

20:34opportunities for individuals to spot

20:36problems and go build something or do

20:40something to go solve it. So I think

20:42we're really evolving. Well, we've long

20:44been evolving but we're accelerating

20:46positive era where people sit around and

20:48wait for their boss to tell them what to

20:50do.

20:50>> This is what I've been feeling a lot

20:52especially when we started doing remote

20:53work. I want people to be entrepreneurs

20:56within their niche. Like if you're

20:58helping me with LinkedIn, you're an

21:00entrepreneur there. You can hire more

21:01contractors. You can deploy different

21:03tools. You make the strategic decision

21:05whether this topic is good or not. Shall

21:07we proceed with it? I really think and

21:09tell me if you agree with me, we're

21:11moving into that job market where

21:12everyone is kind of independent in their

21:15workplace.

21:15>> I think people will have much more

21:17autonomy and creativity. So I agree with

21:19that. And I'd even go one step further

21:21which is I talked to a lot of people is

21:23know engineers and others in large

21:24companies that tell me that their

21:26manager tells them to stay in their swim

21:28lane. They'll say, "Oh, I have this

21:29creative idea, but the manager says,

21:31"No, I need you to focus on this one

21:33thing, frankly, often because their

21:35manager's career depends on it." But I

21:36feel like the number of opportunities

21:38for people to spot things outside the

21:40swim lane. Um, and then in a responsible

21:43way explore how to get it done, that

21:46feels very exciting to me. And I think

21:47that in the future the businesses that

21:50set up a culture that encourage people

21:53to learn AI build fast responsibly um

21:57talk to customers would drive a lot more

21:59value than the more hierarchical silo

22:02organizations.

22:03>> Yeah, it it starts with hiring the right

22:05people and then nurturing this in your

22:07organization. When you say learn how to

22:09use AI and become proficient with AI,

22:11can you give me some benchmarks like of

22:13a person who's like say a marketer,

22:15knowledge worker, advanced with AI, what

22:17are you looking for when you're

22:19interviewing this person?

22:20>> I'm pretty sure my team's ahead of the

22:21curve. All of my marketers know how to

22:23code. So, as part of how I interview

22:25marketers, we ask them what they've

22:27built and uh if they have not built any

22:30software.

22:31>> If it's a dashboard, is it good or bad?

22:33Like, is it too basic or

22:34>> a dashboard? Again, my team's probably,

22:36you know, somewhat ahead of the curve,

22:38but uh

22:39>> was good to hear like that.

22:40>> All of my marketers have built much

22:41specific things in

22:43>> Oh, I I feel like um I don't know the

22:46other day uh one of our someone on the

22:47marketing team was talking about the

22:48tools that he had built to uh when he's

22:51considering writing an article on

22:53something, it will um crawl the web,

22:55find related work, has a custom desktop

22:57app, actually built a desktop app that

22:59runs on his Mac to um highlight related

23:01articles for him. then you can chat to

23:03the whole system navigate you know the

23:05thing he's writing as well as the

23:07related work and he had a large

23:09dashboard for trolling the internet to

23:11highlight to him exciting things that

23:13are popping up um now even on my team I

23:16think that marketers is ahead of the

23:18curve

23:18>> but that's that's great to hear any

23:20other interesting use cases uh that will

23:23inspire people to build something

23:24similar

23:26>> let's see maybe u uh my finance team

23:29uses AI extensively

23:32So I think uh uh my um one of my CFOs uh

23:37realized that you know her team was

23:39spending hours every week clicking

23:40through documents open this copy paste

23:42this number here and so um she started

23:45building uh automation scripts that runs

23:47on a routine that um automatically opens

23:50files checks what's in there checks for

23:52consistency highlights to her team if

23:54there's something uh if there's

23:56something they need to be paying

23:57attention to if a new document has

23:58showed up. So I find that rather than

24:00waiting around for an engineer to do the

24:02work for them, the team's ability to to

24:05kind of a not just build dashboards but

24:08build kind of a data management

24:10infrastructures. They can ingest data,

24:12alert them if something's happening.

24:14>> I think uh my finance and marketing

24:16teams are doing that. Oh my recruiting

24:18team um well we actually have recruiting

24:20engineers which are really professional

24:22engineers that sit in a recruiting team

24:25that are building very sophisticated

24:27tools for recruiting. And this is

24:28actually the other trend. I think

24:29marketers, recruiters, HR professional,

24:31ops people should all learn AI. But the

24:33other thing is when you take an engineer

24:35and embed them in these teams, then that

24:37further accelerates what you can do.

24:38>> We do the same. We we start with

24:40something basic, build it ourselves,

24:41then we hit the wall, an engineer comes

24:43in, we build it further.

24:45>> Frankly, when you look at not just

24:46software engineers, but recruiting

24:47engineers, marketing engineers, HR

24:49engineers, I think there's so much

24:50valuable engineering work that can now

24:52be done. I'm just, you know, not worried

24:55about running out of, [laughter]

24:56frankly, all my friends were so busy. We

24:58think, boy, how could we run out of

25:00engineering jobs?

25:01>> Yeah. Yeah. There are so many cool ideas

25:02you can experiment on. But you touched

25:03up on something that is actually one of

25:05the fears when when we talk about like

25:07financial information, how much you're

25:09giving to AI. So, I gave my perplexity

25:11permission to scan my Fidelity account

25:13so it can track my portfolio, tell me

25:16when to rebalance. It doesn't do

25:17anything on my behalf, but it has

25:19access. Do you think there is any

25:22problem with that?

25:23>> This is complicated. I think AI and

25:25privacy is a complex area. Um, and it

25:29depends a lot on the company that you

25:31are sharing your data with. So, for

25:33example, I trust all the hyperscalers to

25:37really 100%, you know, follow their

25:40terms of service and to do what they

25:41say. my personal opinion not not giving

25:43legal business advice but I'd be shocked

25:44if you know the largest hyperscalers

25:47publish the terms of service with some

25:48privacy notice and if they breach that

25:50because that would be not the culture be

25:52so damaging of the long-term business

25:54model now that's on the largest

25:55hyperscaler side if you look at AI

25:58company's side there's been you know at

26:00least one company that I won't name that

26:03seems to occasionally change the terms

26:05of service and if you're using it you go

26:08to the website so you pop up hey we

26:10changed the terms of service to retain

26:11your data or train your data and if

26:13you're aren't paying attention and click

26:14the wrong button then they suddenly gave

26:17themselves permission to access your

26:19data in a way that I'm not that

26:20comfortable with. I feel like I handle

26:22you know some sensitive information. So

26:24then I tend to be very careful with the

26:27businesses that I just don't feel that

26:30culture and the DNA and frankly the

26:32long-term business model is as tied to

26:35protecting individual user privacy than

26:37the hyperscalers. And um I see

26:40businesses, you know, get this as well.

26:43For example, one of my teams, AI Aspire,

26:45we work with very large corporations,

26:47including banks, with incredibly

26:50sensitive financial data. And as you can

26:53imagine, AI Aspire and our and our

26:56clients do not willy-nilly share, you

26:59know, really sensitive I often material

27:02nonpublic information, right? NPI uh

27:05with Frontier with with Frontier Labs

27:07without really careful thinking about

27:09the guardrails and privacy. So I think

27:11it's complicated.

27:12>> So trusting hypers scale, but also u

27:14another thing that you can do, you can

27:16download an open source model and just

27:17run it on your computer and then it just

27:19stays on your computer, right?

27:21>> Yes. I think yes I it turns out a lot of

27:23banks will actually run um the things in

27:25uh you know a virtual private car or on

27:27prem so they so it never even leaves

27:29their control but I think for

27:31individuals it's true for for the really

27:32sensitive things um uh I sometimes run a

27:36local model and it's been interesting

27:37with the open way models some of the

27:39latest open way models are approaching

27:40frontier capability and that are you

27:43know actually small enough they're

27:45actually really good models now they can

27:47run on it

27:47>> yeah the one from Meta right the recent

27:48one

27:49>> oh yes Metamuse is a good model and I'm

27:50thinking Also the latest version of Quen

27:52is also very good but I think frankly

27:54these models change every other week. So

27:56I think the best practice is to not get

27:59stuck on one but keep on trying new

28:01models.

28:01>> So basically when there is a situation

28:03that you don't trust anyone you run a

28:05local model and this is how you keep

28:06your data safe.

28:08>> I do trust the hyperscalers but

28:10sometimes for you know literally NMPI

28:13material nonpublic information that I

28:15won't even send to that I just can't

28:17even send that to the cloud. So that uh

28:19I'll either do it manually without AI

28:22help or if I really need to use AI then

28:24you know really carefully only use a

28:27local model.

28:28>> Interesting. Okay. This is this is this

28:30is an interesting one. Okay. What about

28:33loss of human control over AI? Because

28:36I've talked to I talked to Yoshua Benja

28:38who is very um negative when it comes to

28:42open free AI without any regulation and

28:46he painted me some very scary pictures

28:49of AI taking over control because we

28:52basically the the whole scenario is we

28:54can't control something that's smarter

28:56than us and if AI gets smarter and

28:58smarter where where do we end up? What

29:00do you think about that? I think about

29:01something else that we can't control

29:03which is um airplanes. No one can build

29:06an airplane that you can fly perfectly.

29:08Winds will buffet it around. And then

29:10candly in the early days of developing

29:12airplanes, some airplanes crash and

29:14people died and it was tragic and awful.

29:17But through the early lessons learned,

29:19we then learned to control airplanes

29:21better and better. So that today, you

29:23know, we can mostly get in an airplane

29:25and not fear too much for our lives. And

29:28it's really like that too of AI. No one

29:30can perfectly control AI because it

29:32generates tokens or outputs that a

29:34little bit random. So we don't really

29:36know what exactly it'll do. But as we

29:38run them and you know there's been a

29:40small number of mishaps which is

29:42unfortunate and some number of mishaps

29:44have done some real damage but the way

29:47we engineer almost any system from an

29:49airplane to electric circus to now AI is

29:53carefully grow their capabilities so

29:57that we can have a controlled

29:59environment in which to measure what's

30:01wrong and then to shape it to make sure

30:03we can control it well enough that it

30:05behaves responsibly and safely and to

30:07this day we can't perfectly control any

30:09airplane and we will never perfectly

30:11control AI either but I think um we are

30:14certainly controlling them well enough

30:16that this loss of control doesn't feel

30:19like science science

30:20>> fiction yeah what about deep fakes

30:24>> deep fakes are a problem well one of the

30:26most disgusting things I've ever seen or

30:28heard of is non-consensual

30:30intimate deep fake imagery

30:33>> I'm really glad that you know US

30:35Congress has been moving Right. Let's

30:37pass laws. Get rid of that. Penalties

30:39for that. I'm just I think there's some

30:40really problematic uses of AI that we

30:43should outlaw, heavily penalize. Let's

30:45just get rid of that.

30:46>> What do you think about children and

30:48social connection when it comes to AI

30:50with kids using more of AI? Because

30:52we've seen social media how, you know,

30:55there are people who are dumb scrolling

30:57all day and my daughter who is 5 years

30:59old now, whenever I don't have an

31:01answer, he's like, "Ask Chad GPT." And

31:03like, who's that person? I'm like, "I

31:05don't know. Ask Jajiv Viti like she

31:07thinks Jaji knows everything. What would

31:10you say about you know kids future with

31:14AI?

31:15>> First I think kids have a bright future.

31:16It's just such an exciting time to be

31:18child to grow up in this environment

31:19with tools that none of us ever had

31:21before. [gasps] At the same time we've

31:23seen that social media um I think social

31:26media has probably been blamed a bit

31:27more than it deserves. But it does

31:29deserve blame uh has kind of not been

31:32great for kids. I actually worry a lot

31:34about it's a wonderful tool, but AI

31:38damaging learning is something I worry a

31:40lot about. So, it turns out um I have a

31:435-year-old and a seven-year-old. When I

31:45teach them math, they're so young enough

31:47that I can basically, you know, not let

31:49them use a calculator, can say, "How do

31:52you multiply these numbers?" And I don't

31:54give them a calculator and practice that

31:55with them. But as they're a little bit

31:56older, I worry a lot about students

32:00using cognitive offloading to AI in a

32:02way that damages the long-term learning

32:04retention. Um, but then at the same

32:07time, oh, I actually built an app. I did

32:08not like any of the, you know, free

32:10online learning to type types of things.

32:12So, I actually built my own to um have

32:15my daughter learn to type. And I'm

32:17hoping that she's actually getting

32:19pretty decent now for a seven-year-old.

32:20>> Oh, she's Oh, yeah. She actually typed

32:22all the lowercase letters. she's a

32:24little bit fit, you know, not her shift

32:27uppercase letter is a little bit not

32:29quite there. But I think that this um

32:31unlocks, you know, responsible adult

32:34supervised use of online tools and I

32:37think it's really tricky. You know, I

32:40think um adult supervised use of digital

32:43tools seems a great thing for kids, but

32:45too many adults don't have time to

32:46supervise the use of the tools and then

32:48the incentives of [gasps] say social

32:50media, right, to do funny things.

32:53>> Yeah. has to be the right incentive when

32:55it comes to AI. Okay. [snorts] You

32:57mentioned we we talked about the fears.

32:59We talked about how you can improve your

33:02work with AI. Can you name some of the

33:04biggest opportunities in AI in 2026 for

33:07people who want to build? for an

33:08individual that wants to build. I don't

33:11think it's one size fits all, but

33:13because the cost of building has

33:16plummeted,

33:17um, [clears throat]

33:18I encourage people to learn AI, build

33:22fast, and talk to customers. I find

33:25myself building things, I don't know,

33:28every week, every weekend because I or

33:31someone on our team, we have some

33:32problem and I have some idea for

33:35building some AI thing. to automate it.

33:39Last weekend, I had really I was using a

33:41frontier model to analyze a lot of our

33:43key business metrics because I didn't

33:45have time to do it myself, but it was

33:46kind of measuring, you know, deandized

33:49key business metrics and I didn't have

33:51time to go find a data scientist to go

33:53work me on it. So, I just did variety of

33:55frontier models being really careful on

33:58their uh data retention policies. I did

34:00not use models with data retention

34:02policies I don't like uh in order to

34:04analyze data. But and then I find that

34:07um what's happen of AI is the cost of

34:10building has plummeted and so the

34:11challenge is shifting to deciding what

34:13to build which I was calling which I've

34:15been calling the product management

34:16bottleneck and so people you know

34:19founders engineers product managers that

34:21can talk to customers get a sense for

34:24the taste of judgment on what to build

34:26and then build with AI and iterate

34:28quickly. I think that's just a ton of

34:30exciting things to do

34:31>> and you've been starting so many

34:32companies. You're like when I looked at

34:34your portfolio, do you think for

34:36beginners when you said you built

34:37something during the weekend, how do you

34:39decide what to focus on or you can

34:41pursue multiple ideas because of AI now

34:43and you can just be, you know, playing

34:45in different companies at the same time.

34:47>> It turns out building a company is still

34:49really, really hard and so there's a lot

34:50to be said for a single threaded

34:52leadership or someone that's fully

34:54focused on just one thing. I find it,

34:56you know, over a weekend I can often

34:58build an Elm wrap, build a simple

34:59application, but I wish it was that easy

35:03to build a large company. Um, I find

35:07that building something meaningful often

35:09takes either real technical depth uh and

35:12or deep customer insight and integration

35:15with customers. And yes, we can now, you

35:18know, use AI to code something in a few

35:20hours, but that's a small piece of the

35:21puzzle. So spending time understanding

35:24the technical complexity and building

35:26the really complex software that takes

35:28us like months you know maybe years or

35:30having that deep custom insight to

35:32decide what to build that also just

35:33takes a lot talking to people reading

35:35facial expressions surveys doing that

35:37over and over until we figure out what

35:38to build. Um and so I think sometimes

35:41there's a lot of value to sampling

35:43widely but then having that focus for an

35:46individual to go really deep in a couple

35:48sectors that that still seems important

35:50for building a business. My last

35:51question, I know it's we don't have much

35:53time, but I wanted to ask you about AGI

35:55just because people use this word so

35:57much and some people say I think Jensen

36:00Hang said we already reached AGI. You

36:02said it's decades away. What's the one

36:05criteria when you're going to say we

36:06reach AGI?

36:08>> So different people say we reach AGI at

36:10different times because of different

36:11definitions of AGI. The definition I'm

36:14most familiar with is AI that could do

36:16any intellectual task that a human can.

36:18But so the human brain can take say five

36:22years to study and do a PhD thesis or or

36:25and so can AI write a PhD thesis or a

36:28human can learn to drive a truck through

36:30a dense rainforest with you know tens of

36:33minutes of practice. So when can AI do

36:35that to drive new environment with tens

36:37of minutes of practice. It feels like

36:38there's a long list of these things that

36:41AI cannot do uh for what feels to me

36:45decades. I hope it's only decades. maybe

36:48you turn out to be longer. So that's why

36:50I think for that definition of AI or

36:51AGI, AGI is still very far away.

36:55>> But um it turns out because of you know

36:57economic incentives, I think open

36:59Microsoft had an agreement that's

37:00actually been renegotiated now. So

37:02that's gone away. But open AI had an

37:04economic incentive to try to declare

37:06reaching AGI earlier. Uh and so it turns

37:10out that if you come up with other

37:11definitions of AI depending on how far

37:14you lower the bar then you could totally

37:16have reached AGI you know already or

37:18even 30 years ago depending on how you

37:20want to define it.

37:21>> Yeah true Andrew thank you so much for

37:23this positive conversation very

37:25applicable. I like when u you watch

37:28something and then you go and you

37:29measure yourself against what people are

37:31doing with AI look at your process and

37:34uh maybe expand it. So thank you so much

37:36for showing what your team is doing and

37:38thank you for your insights. Yeah, I

37:40think given the huge benefits of AI to

37:43come, I hope whoever was watching this

37:44is motivated to really go learn AI,

37:48apply it, um, and and even to go build

37:50some