YT YouTube2Text

YouTube2Text → Transcripts → 후회는 안할거야 ... (에릭 슈미트 구글 전 대표)

후회는 안할거야 ... (에릭 슈미트 구글 전 대표) — Transcript

by ARBTY | 아비트라리 · 4,495 words · 714 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Eric Schmidt has had a front-row seat to
0:02nearly every major technology shift of
0:04the last three decades. You know him
0:07from running Google, and since then he's
0:09become one of the leading voices shaping
0:11the conversation around artificial
0:13intelligence. He is CEO of Relativity
0:15Space, giving him yet another vantage
0:17point on how breakthrough technologies
0:20move from idea to reality. Eric, welcome
0:23to Inside Blackstone.
0:25>> Oh, thank you. Thank you very much.
0:26Well, of course I I I love the founder
0:28and the firm and everything you guys are
0:30doing, so.
0:30>> You're a technologist, you're a computer
0:32scientist. You give these incredible
0:34technical talks. I think your superpower
0:37is that you're able to break down
0:38complex ideas and make them more
0:41accessible. And that's ultimately what
0:42we're trying to do here.
0:44>> you.
0:44>> It's amazing the breadth of things that
0:46you've done in your career, that you're
0:47still doing.
0:48>> I'm not dead yet.
0:49>> You're not dead yet by any means, and
0:51that I guess that's my question is
0:52you're, you know, you're 71. When most
0:55people achieve the kind of success you
0:57have, or not even close, they spend the
0:59rest of their life sort of talking about
1:01it, what they accomplished. You keep
1:03reinventing yourself.
1:04>> I I I can't seem to stop. My own view is
1:07as you get older, you should take on
1:09more risk because you have less to lose.
1:12Right?
1:13>> Right.
1:13>> And so, you might as
1:19well try the things that you thought
1:20were impossible. of the ethics and the
1:23world that you care about, I care about
1:25democracy and I care about freedom.
1:27That's what I want to work on. At this
1:29point in in my own life, I'm interested
1:31in impact. Most of the problems in the
1:34world could be solved in Eric's little
1:36brain with an improvement or an
1:38adjustment in technology. If you look at
1:41at sort of improvement in our lives,
1:43improvement in our health, and so forth,
1:44it's all been related to technological
1:46progress. Starting with the invention of
1:47fire, which was extremely useful. Right,
1:50way back when.
1:51>> But also dangerous, right?
1:52>> And everything is dual use, and
1:54everything has to be managed. But just
1:56look at look at the dollar of value of
1:59labor over a thousand years or 500 years
2:02or 100 years, right? Human human output,
2:06human hours in terms of productivity are
2:08vastly improved. And because we've
2:10somehow collectively decided to have
2:12fewer children, not a good thing.
2:14>> I have four.
2:15>> Uh congratulations.
2:16>> Doing my part.
2:17>> Have as many as possible. [laughter]
2:19More children are good and more
2:20grandchildren are good and so forth. Um
2:23we're going to have to automate. And so,
2:25human wealth, human health, human
2:27well-being, one of the major
2:29determinants will be over the next
2:32hundred years, can we continue to
2:34accelerate progress in this? I
2:36profoundly believe we can. And I
2:39profoundly believe that long after I'm
2:40gone, humans will be much better.
2:43>> So, you tend to be way more optimistic
2:45on the benefits of AI than most.
2:48What gives you that confidence?
2:50>> Would you like most human diseases to be
2:52solved in the next 15 years?
2:54>> I would.
2:54>> Sign me up, especially given my age.
2:57Would you like solutions to climate
2:59change?
3:00>> I would love that.
3:01>> Absolutely. Maybe we could have safer
3:04products. Um maybe we could have better
3:06educational solutions. Maybe we could
3:08change the educational system so the
3:10kids are not bored, but rather they're
3:12engaged. Um the opportunities are
3:15significant. In the last month,
3:1810 major math problems were solved, uh
3:21including one called Navier-Stokes,
3:23which is historically important, by
3:25computers. Now, you sit there and go
3:26like, why do I care? Well, math is the
3:29foundation of that acceleration.
3:32>> For those on this listening to this
3:33podcast that are not technologists, what
3:35was the significance of that?
3:36>> Um Navier-Stokes is an equation that
3:38involves fluid flows.
3:39>> And why were they unsolvable math
3:41problems?
3:42>> The specific conjecture was had never
3:45been proven.
3:46And so, they use approximations for
3:48Navier-Stokes. Uh I indeed funded some
3:51physics work in this area, which I never
3:53understood, but they were really smart
3:54people. And the key thing is that that
3:57these fluid flows model, for example,
3:59lift of an airplane, how do air
4:01conditioners work? There are many, many
4:03examples.
4:04So, being able to solve the problem in a
4:07particular way indicates better
4:09algorithmic answers. We can get a better
4:12and more accurate answer. Why do you
4:14care? The airplane will fly faster. The
4:16airplane will use less gas. Right? I can
4:19go on. Uh the rocket will go faster. Uh
4:22the fuel will be less. You'll get to
4:24Mars quicker. You know, what have you.
4:26All of these fluid flow things matter in
4:30a way that's that we sort of take for
4:31granted.
4:32>> Let's go back to the internet, cuz I
4:33think it's so interesting. You've been
4:34around the rise of the internet and
4:37other platform, you know, shifts, the
4:39rise of enterprise software, search,
4:41cloud computing, mobile, all of these
4:43major technological innovations. Why
4:46does this moment feel different to you?
4:48>> It's not different. It's just bigger.
4:50>> Bigger.
4:51>> The internet felt exactly the same as
4:53this. I started with mainframes. So,
4:55that's how how long I've been in this
4:57field. I've been doing tech for 55
4:59years. And when the PC revolution came
5:03along, everyone said, "Oh my god, this
5:05is this huge wave." It created Microsoft
5:07and Apple and so on and so on.
5:09Uh this was 40 years ago, 45 years ago.
5:12So, these things take a while. And I
5:14think what another example, uh let's
5:17look at self-driving cars.
5:18>> Mhm.
5:18>> Did you arrive at work today in a
5:20self-driving car?
5:21>> I suspect not, nor did I. Um and so,
5:25what do we think about that?
5:26>> you think we will?
5:27>> Um well, it's beginning. Now,
5:29self-driving cars will
5:30>> York City?
5:31>> Uh New York City will be one of the last
5:32ones.
5:33>> Okay.
5:33>> But, it'll come.
5:34>> It'll come.
5:34>> But, it's coming in the different
5:35cities, uh including obviously Waymo and
5:37other competitors. But, why do I mention
5:40self-driving cars? They were largely
5:42they were designed in the '90s. The
5:44first real test of self-driving cars was
5:472004. So, that's 22 years and going.
5:51So, the diffusion rate is different from
5:53the invention rate.
5:55So, there the faster the diffusion rate
5:58of the sort of broad adoption, the
6:00quicker the societal change.
6:02Why does it occur in what I do first?
6:05It's because we're already connected.
6:08Right? You It's essentially a software
6:09problem. It's just connectivity. The
6:11marginal cost of connectivity, given
6:13we're already so connected, is very low.
6:14The incremental cost of connecting you
6:16and me is essentially zero. Boom, it
6:18just happens. It's straightforward
6:20economics. Whereas getting self-driving
6:23cars is a great deal of capital. It's
6:24hard and so forth. People are working on
6:26it. Why 20 years versus zero years? And
6:29that's the answer. You see the same, by
6:30the way, in AI. All of the gains right
6:33now are occurring in what is called
6:35essentially scale-free. And scale-free
6:38means that you can just keep doing it
6:40and get smarter. So, in math, I'm not a
6:42mathematician, but as best I can
6:44understand, they just invent stuff. And
6:46they just keep inventing things. And
6:48they keep inventing things. And they're
6:49so clever. Well, now a computer can take
6:52all of those ideas, and if you just give
6:54it enough juice, it can do the same
6:55thing.
6:56Because it doesn't need data It's It's
6:58self-contained.
6:59>> Unlocking all these solutions.
7:00>> Unlocking these solutions. In software,
7:03it turns out that once you start writing
7:05code, and you start learning how well
7:08the code works, you can make it better.
7:10And then you can make it better and
7:11better. And there's a what is called a
7:13recursive self-improvement loop, where
7:15it's getting smarter on its own.
7:17But this progress that I'm describing is
7:21happening faster than I have ever seen.
7:23Right? What it is true is each wave is
7:27bigger, but it's also faster. And we are
7:30not ready as a society.
7:31>> Eric, you gave me like 10 things to
7:33unpack there, but
7:33>> [laughter]
7:34>> let's try to do this. So, recursive
7:36self-improvement, right? This is the
7:37thing that had the Anthropic
7:39researchers, two of them making
7:40comments, one leaving, making comments,
7:42the other is sort of validating that
7:44there are serious concerns about the
7:46speed with which AI is learning, right?
7:49>> So So, by the way, every month there is
7:51a group of people who quit over this
7:53this issue.
7:54>> topic, exactly.
7:55>> Well, Silicon Valley, and in particular
7:57San Francisco, they're absolutely
7:59convinced that that this is going to
8:02happen so quickly that it will threaten
8:04humanity in one way or the other. I
8:06personally do not agree with this.
8:07>> Okay, why?
8:08>> Well, let me explain their argument.
8:11Um
8:12is a relatively straightforward
8:13argument. As humans, we learn at a
8:16certain rate. If you have enough
8:17hardware and enough electricity and so
8:19forth, and enough scale-free learning,
8:22which is what I'm talking about, it can
8:24just get faster and learn faster than
8:26humans can. And the way it works, um
8:29I'll give you a simple example.
8:31I ask it to write a chapter, and then it
8:33writes the chapter, and I feed that
8:34chapter and say, "Write the next
8:35chapter." And so forth. Well, if I keep
8:38iterating like that, eventually it
8:41learns what I wanted with a process
8:43called reinforce- reinforcement
8:44learning.
8:45>> Easier to write the 12th chapter than
8:46the first.
8:47>> And it gets smarter.
8:48>> Mhm.
8:49>> And once it learns how to do it, it
8:51knows. It's like the way we learn. Uh
8:53but it can learn a lot faster.
8:55So So, think of it as humans versus
8:58computers. I think the the actual
9:00technical answer about this learning is
9:02it's not human learning, it's a
9:04different kind of intelligence. And so,
9:06the the best way to state the fear is if
9:09we're inventing a different kind of
9:11intelligence that we cannot necessarily
9:14constrain. This is called the alignment
9:16problem.
9:17Um also, in the past weeks, there was a
9:19very famous
9:21situation where there was an internal
9:23test going on at OpenAI, um and this
9:26internal test, which was trying to check
9:28something, managed to create a set of
9:30bots. They managed which are agents, if
9:33you will. They managed to collaborate to
9:35violate certain norms and laws. They
9:37managed to break into things, they
9:38managed to allegedly do a little bit of
9:41harm. Now, nobody was hurt. But, it's an
9:43example of what is possible.
9:45So, we must develop both recursive
9:49self-improvement, which is the
9:50self-learning, along with solving the
9:52alignment problem to human values. So, a
9:55simple way would be I say to you,
9:58um, learn everything, and you're a
10:00computer.
10:01That's not what I should say. I should
10:02say learn everything, but
10:04don't harm any humans, follow all laws,
10:07follow all laws, follow the US
10:08Constitution, make sure you don't get
10:10sued. You know, in other words, there's
10:11a set of assumptions
10:12>> that you build into the model.
10:14>> These systems don't have a fear of the
10:15police. They don't have feelings. And
10:18so, they have to be constrained.
10:20>> Come back to your optimism. It's in
10:21everything that you, you know, you've
10:23written and you've said. What gives you
10:25optimism that we will be able to sort
10:28out a future where humanity wins?
10:31>> What's the alternative?
10:33I mean,
10:34the best way to get a headline is to
10:35say, "In 10 years, we're all dead." I I
10:37was on a panel with a guy who wrote a
10:39book which is entitled, "If we invent
10:42superintelligence, we're all dead." It's
10:44very hard to argue with someone in that.
10:47The only problem with his argument is
10:48he's wrong, right? Because part of
10:50superintelligence is getting the super
10:52intel- super alignment problem solved.
10:55>> I would guess that your social network
10:57might be a little bit different than
10:58mine. What's the buzziest thing that you
10:59were hearing about this summer in the AI
11:01circles?
11:02>> The biggest and most important thing
11:04that has occurred in the last year is
11:06the arrival of long reasoning, deep
11:09reasoning.
11:10And when I say long reasoning, I mean 8
11:12hours of reasoning.
11:14So, what would happen is that you would
11:15start a year ago or two, and it could
11:18kind of begin thinking, and then it
11:21would start thinking about something
11:22else.
11:22>> Get distracted.
11:24>> Like I do.
11:24>> We can anthrop- we we can make it feel
11:26like human, but somehow it it takes a
11:29wrong path. It decides that we're
11:30instead of talking about math, we're
11:31going to talk about French literature.
11:33You know, something
11:34>> That's hilarious.
11:34>> Something that doesn't make any sense,
11:35right? Um that problem was solved using
11:39essentially technical guardrails for
11:41what it kept thinking. So, they said
11:43stay focused. Stay focused. Stay
11:45focused. Is the way to understand it.
11:47And that produced the ability to ask
11:49very They're called chain of thought
11:50trains. Where you can basically see it's
11:53thinking, how it solved a problem. And
11:56these chains can be a thousand steps,
11:592,000 steps.
12:00Now, why am I mentioning this? I don't
12:02know about you. I can't think a thousand
12:04steps. I just don't I'm not It's not
12:06that smart, but it can. So, it looks
12:09like deep reasoning will allow you to
12:11see deeper as a human that than a human.
12:15The best AI is in service of humans
12:17doing things that humans are not so good
12:19at.
12:20>> So, artificial superintelligence, how
12:22will we know we've achieved it and what
12:24do you think the timeline is?
12:25>> So, most people believe right now that
12:29we're at the beginning of
12:30superintelligence, which is broadly
12:32defined as humans plus super smart
12:35computers working together, right? There
12:37are more technical definitions. Now,
12:40today vision is better,
12:43um data storage is better, retrieval is
12:45better, and analytical stuff should be
12:48better, and proofs are getting better.
12:50So, we're at the beginning of that. What
12:52is the thing that is not obvious how to
12:54do? And the answer is real discovery.
12:57So, the test that's used in the industry
12:59is
13:00could you
13:01take all the information that's known in
13:031902, put it into a computer,
13:06and have it generate um general
13:08relativity, special relativity,
13:10Einstein, 1902-1904.
13:12Um the consensus right now is no.
13:15And the consensus is no because it
13:16required a leap that was not obvious
13:20from any knowledge that existed at the
13:22time.
13:23Okay?
13:24So, in the industry, we say, well, how
13:26would we solve that?"
13:28Well, there's an obvious way, which is
13:30you take the computer and you just say,
13:33"Keep trying." You know, just it's the
13:34it's the monkey with the keyboard
13:36eventually produces Mozart, but it takes
13:37a long time.
13:38>> Mhm.
13:39>> I don't think that's going to work
13:39because it's computationally not
13:41realistic.
13:42A better idea is to invent something
13:45which allows for analytical reasoning in
13:48non-coherent spaces. So, think of it as
13:51a leap. I happen to understand something
13:54in physics and then you give me a
13:56completely unrelated problem and somehow
13:59I
14:00>> I something clicks, you know, it's like
14:02in my case, you're in the morning you're
14:03in the shower and you have this idea
14:05>> Right.
14:05>> that just showed up. Can we invent that?
14:09>> So, do you think these leaps are coming
14:11faster than we realize?
14:13>> The kind of leaps I'm talking to have
14:14not occurred yet. There's something
14:16called the San Francisco consensus by me
14:18anyway, and everyone in San Francisco
14:20thinks this will occur in 2 years. I
14:22don't.
14:23For one thing, I don't think there's
14:24enough computers, I don't think there's
14:25enough people, I don't think there's
14:26enough algorithms. I think maybe in my
14:28lifetime, maybe in a decade. That would
14:30be a big deal.
14:31>> Mhm.
14:32>> But each of these moments are they're
14:33happening so fast now.
14:36And and by the way, it's fast because of
14:38the internet. It's because of what is
14:40called basically scalable computing.
14:43Um the supercomputers that are being
14:45used. So, typical model takes three or
14:47four months, costs $100 million to
14:49train, takes a few months to to test.
14:52Right?
14:53Thank God for the American financial
14:55industry that would be willing to
14:56finance these things.
14:58>> And we're seeing the creation of an
14:59entirely new financing model for this
15:01industry, right?
15:02>> If you want to achieve the vision I am
15:04describing, you have to do it at a world
15:07level. America is particularly good at
15:09building world standards and
15:11world-changing impacts at this scale.
15:13You're going to have to do it in the US
15:15for regulatory reasons and you're going
15:17to have to find a way to finance it.
15:18You're also going to have to wait to
15:20find a way to solve the energy problem.
15:22The best energy solution is fusion. They
15:24look like they're going to get to fusion
15:26within two two to three years. When that
15:29occurs, that's another massive moment in
15:31human history. I think it will occur in,
15:34you know, three to five years.
15:35>> You've said you've sort of invested in
15:37everything when it comes to AI, but if
15:39you could only invest in one thing
15:41related to artificial intelligence, like
15:43one theme, not a single company, but one
15:45theme, what would it be?
15:46>> Uh long reasoning.
15:49I don't think we appreciate when we have
15:51a thought partner
15:53that can do can think longer than we do.
15:56We can't think about how So, I'll give
15:58you an example.
15:59You're a CEO. You have 24-hour agents.
16:02You have one that's running looking at
16:04security, another one that's looking at
16:05power, another one that's looking at
16:06cash, another one that's looking at
16:07products, another one that's looking at
16:09global macro, and they're just running.
16:12And then your AI CEO every day when you
16:14get up says,
16:15"Does Eric need to know anything? Does
16:17Christine need to know anything?" And
16:18they say, "Nah, nothing special." But
16:21it's always running. So, you want to
16:23think about that in running a business.
16:26That those things are You have to sleep,
16:28it doesn't. They can be disciplined into
16:31focus on this and not something else. Is
16:34that important? Absolutely.
16:35>> So, do you think there's a productivity
16:37boom
16:38that's coming?
16:39>> Well, I think it's already underway.
16:40>> It's already underway.
16:41>> An example would be my entire life, I
16:43would according to myself, I was a very
16:45very good programmer when I was 22,
16:47right? I was one of the best, according
16:49to myself. My field is over, right?
16:53>> And yet you still tell people to learn
16:55coding.
16:55>> I do because I want them to be
16:57architects. So, I went from be I had to
16:59build the house, now I can architect the
17:02house using the equivalent of Claude
17:04code or OpenAI Codex or future products
17:07from Google, uh which they're working
17:09on, all of which allow me to say what I
17:13want and the code gets written. I look
17:15at the code it writes now. I can't write
17:17that code. I couldn't when I was 22.
17:19Right? So, I got beaten. Right? Hard. Um
17:23it's okay. Right? I did I did fine.
17:25The next generation of people like me
17:27will be the architects, and they'll be
17:30very very smart, and they'll know how to
17:31say this is what I want. Um one of my
17:34friends Some of my friends are not so
17:36friendly, and they're not so social.
17:37They're kind of So, we say asocial.
17:40They're nice people.
17:40>> No naming names.
17:41>> I'm not going to name anybody. Um so, I
17:43said, "What are you doing?" And he said,
17:44"I'm deleting names from my address
17:46book." And I said, "Why?" He said, "I
17:49don't want to talk to them anymore." I
17:50said, "What do you do all day?" Well, I
17:52get up in the morning, I have coffee,
17:54and then I check with my friends. And I
17:55said, "Which friends?" He said, "My
17:57agent friends."
17:58>> Oh, god.
17:59>> Okay. So, you So, what do you do? And he
18:01spends all morning
18:03plugging and playing with his agents who
18:05are solving really interesting problems
18:07to him.
18:08And then he gives them a 1-hour project
18:10for lunch. He has lunch, and he comes
18:12back and he interacts with his agents.
18:15Now, he's got a wife and a family and so
18:16forth. So, at 5:00, he gives it an
18:19overnight, you know, task.
18:21>> Right. And it goes to work.
18:22>> And it goes to work, and then it
18:24typically completes between 2:00 and
18:254:00 in the morning. And I said, "Do you
18:26wake up for that?" He said, "No, no, no,
18:28no. I wake up naturally. I know I with
18:31certainty when I wake up, I'll have a
18:32result."
18:33>> I was with you on everything other than
18:35that these are his friends. You right? I
18:37mean, I think they're amazing
18:38productivity tools. But, I'm going to
18:40take us back now to data centers. The
18:42case for data centers was made long
18:44before artificial intelligence, for
18:46mobile and cloud computing and
18:49cybersecurity, all these things that it
18:51helps enable. Yet, now there's so much
18:53concern around data centers. What's your
18:55take?
18:56>> Well, if you're running one of these AI
18:58companies,
19:00your revenue is completely determined by
19:01your data centers. So, this is a massive
19:04change in what I do. It It used to be
19:06that software was pretty high gross
19:08margin. You could sort of sell it.
19:09Didn't have a lot of capital.
19:11Because you need the hardware.
19:13That's what's changed the economics. So,
19:16in the short term, it's the same thing
19:18as in the gold rush, the people who made
19:21all the money were the the pickaxes and
19:23so forth and so on. The infrastructure
19:24>> shovels. We have a whole
19:25We have a whole thing on this, Eric.
19:27>> And and this picks and shovels thing is
19:29real. Um so, from the standpoint of the
19:31economy,
19:32that's those are huge they're real
19:34businesses. There's an estimate that 11%
19:37of US electricity demand in 2030 will be
19:41AI data center related. That's an
19:43extraordinary transition by any measure.
19:45Historically, it was 2 or 3%. So, and
19:48that buildout is one of the primary
19:49drivers of economic growth in America
19:52today. So, thank goodness for that.
19:54>> If you could leave CEOs with one
19:55takeaway today, what would it be?
19:57>> The the term that I've been using is
19:59being AI native CEO. So, what I'm trying
20:02to do in all my things is I'm trying to
20:04fully automate the execution of the
20:06businesses um in the factory floor in
20:08Relativity,
20:10I was walking through it one day and I
20:11thought, "How many computers are here?"
20:13So, I said, "Why don't we connect them
20:14all?" So, they just connected all the
20:16computers. And so, the immediate thing
20:18that they can do using AI is they can
20:21look at sort of utilization cross
20:22utilization times in new ways.
20:24It occurs as a natural byproduct of
20:26getting the digital systems connected.
20:29Furthermore, if you take generic data,
20:31right, unstructured data, and you
20:33connect it all, these systems are smart
20:35enough that they can interpolate between
20:37one data normal seeing another. In other
20:39words, they can combine disparate data
20:41sets and give you business insights. So,
20:44how do I make more money? How do I
20:46increase uh customer retention? How do I
20:48make sure I don't lose a customer? You
20:49can
20:50ask these questions.
20:52What I've been struck by is that when
20:54you take open-source data, that is
20:56public data,
20:58plus the proprietary data in your
21:00business, and you put it into a system,
21:02and you start asking questions, I go
21:05crazy over it. You can now talk across
21:08all of the data. You can get data
21:09fusion. You can actually figure stuff
21:11out.
21:11>> It drives me crazy a little bit though
21:13when I do have content given to me that
21:15is generated by the computer.
21:17>> Is it correct?
21:18>> Some is and some isn't. Sometimes I
21:20worry about the loss of original human
21:23thought.
21:24>> There's a whole bunch of concerns.
21:26I think the most one the one that I've
21:28always been worried about is the loss of
21:30deep reading. Which I blame not on AI
21:33but on social media. I used to read a
21:36book a week and now I'm too interrupt
21:39driven to do that. And my attention span
21:42is so much shorter.
21:43>> I've been watching young scientists. I
21:45fund a whole bunch of young scientists.
21:47How do they do it? And the answer is
21:49they turn off the phone. They turn off
21:51and they just do it. But it requires
21:54this huge strength to turn off the drug.
21:58Right? That interrupt drug, the
21:59serotonin that we have. That's bad. In
22:02my view, that's bad for society.
22:04>> So I have I have a bunch of young kids.
22:06I have you know, two that are nearing
22:08their college years. What would you tell
22:10young people to study today?
22:12>> Well, I used to say biology. Now I say
22:13deep reasoning. If you're a
22:15non-technical person, you should figure
22:18out how to use these tools to make your
22:22dreams and your realities
22:24extraordinarily scaled. You want to be a
22:27global star, a global influencer, a
22:30global impactor, or a global discoverer,
22:32a global singer. You want to use these
22:35tools. Whatever it is that you you want,
22:38figure out a way to use them to amplify
22:40you and what you care about and your
22:43innate goodness. If you're a technical
22:45person, use the same tools to invent
22:49stuff. And to invent stuff that changes
22:51the world. I've never seen the the cost
22:55of entry to be so low and the
22:57availability of these ideas so great.
22:59The only thing that limits you is your
23:02curiosity, your willingness to take
23:04risks, and so forth. So, get over it and
23:07say, "I want to dream. I want to use
23:09these tools to have this enormous
23:11impact." Right? And some of you, not
23:14everybody, will have a huge impact. And
23:17it won't just be coming being
23:19financially successful. You might become
23:21famous or important in something that
23:23you didn't even know was important, but
23:25you'll be so proud of yourself.
23:26>> I love the optimistic future that you've
23:28painted and all that you've taught us
23:30today. I really appreciate [music] it.
23:31Thank you so so much.

About this transcript

This page contains the full transcript of 후회는 안할거야 ... (에릭 슈미트 구글 전 대표) by ARBTY | 아비트라리, generated from the public captions YouTube serves with the video. The transcript has 4,495 words across 714 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.