YT YouTube2Text

YouTube2Text → Transcripts → A Top Mathematician's 9 Lessons for Anyone Who Feels Behind | Ken Ono, Axiom Math

A Top Mathematician's 9 Lessons for Anyone Who Feels Behind | Ken Ono, Axiom Math — Transcript

by EO · 6,205 words · 926 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Hi, my name is Ken Ono. I'm a
0:01mathematician. I work at Axiom Math and
0:04the University of Virginia.
0:06But in my life, I was not a good student
0:09in college. In fifth grade, we had a a
0:11math contest [music]
0:13and I got third. Now, third is a pretty
0:14good out of a fifth grade,
0:16but my father was a famous
0:18mathematician. They came to the
0:19competition and son of famous
0:21mathematician gets third in Hampton
0:23Elementary [music] School. I thought
0:26for 50 years of my life that I would
0:29have utterly failed. [music]
0:31But the reason I bring this up is that
0:33when my dad passed away in January, we
0:36were cleaning up his belongings and of
0:39all the things that he could have kept,
0:41and it was just in a closet now, was
0:43that plaque from my fifth grade contest.
0:48And I I thought, "Wow, I had
0:50misinterpreted
0:52that event my whole life."
0:55It actually meant something him to keep
0:57a plaque where I didn't win, but I got
0:59third place. What he saw, I'm sure, was
1:03that I wanted to do well.
1:05We live at a time where the world places
1:08so much emphasis on benchmarks. How
1:12these
1:12>> [music]
1:12>> AI firms and the state-of-the-art large
1:15language models are competing for these
1:17crazy scores. There's a lot of anxiety
1:19over benchmarks. [music] But when it
1:21comes to assessing intelligence, do we
1:24honestly believe that someone who gets a
1:25higher IQ score is somehow smarter?
1:29No, of course not. And it should put
1:30people at ease, right? If you have to
1:33live up to the standards set by someone
1:35else, then you're not living for
1:37yourself. You're not giving yourself
1:39credit. I consider that toxic. Because
1:42>> [music]
1:42>> you know what the brutal truth is?
1:44The brutal truth is if you're not LeBron
1:47James or Rafael Nadal or a Nobel
1:51Prize-winning scientist, the reality is
1:54you will always be able to find someone
1:56that looks better, achieve something
1:58that you cannot do, and you're not then
2:00giving yourself permission to live the
2:02life that was meant for you. It's
2:04important to give yourself permission to
2:07live your life.
2:23Almost exactly 1 year ago, I was part of
2:27a group of mathematicians hired by a
2:30company called Epoch AI
2:32to write very difficult math problems
2:35that would serve as a benchmark for
2:37state-of-the-art large language models.
2:40And
2:41I thought it would be easy money. We
2:43were paid.
2:44But last year
2:46I found it very difficult to write some
2:48of these problems. Strictly speaking,
2:50the models would make mistakes, but when
2:52you studied the reasoning traces, it was
2:55frightening how far these large language
2:57models have come. So, I think the right
2:59way to describe it is, well, is an
3:02identity crisis. Maybe it was something
3:04like being the sharecropper, the farmer
3:07in the early late 19th century who comes
3:10face-to-face with the first combustion
3:12engine tractor, recognizing that, well,
3:15maybe there's no future for my work as a
3:17sharecropper. What to do next? What's
3:20next for a mathematician?
3:22It was pretty devastating, honestly,
3:25seeing these models solve problems that
3:28were on my research program. Well, came
3:31to realize that technology has helped
3:34mankind over and over again. There was
3:38the invention of the wheel, and later
3:40there was the invention of the engine,
3:42and then calculators and computers, and
3:45somehow
3:46we adapted.
3:48What's surprising about this particular
3:50moment is that many of the techno
3:52technological advances were about
3:54lightening physical work. An elevator
3:57meant that you didn't have to climb all
3:58these stairs. Tractors can do a lot of
4:01work that humans shouldn't do. The
4:02difference now is the work is mental.
4:05That is the stuff of identities. And
4:08where are we now? We are now at a point
4:10where many of those skills can be done
4:12automatically. And AI companies are
4:15talking about what's called self-play.
4:18They want their AI systems to play with
4:20themselves. And so, where does that
4:22leave us?
4:23Well, mathematics, it's it is
4:25devastating. It would be dishonest to
4:27say that a student who's graduating from
4:30college now with a bachelor's degree in
4:32mathematics or who is in graduate school
4:34now isn't deeply worried about all the
4:37years of effort they put into learning a
4:39trade, learning a body of knowledge that
4:41now anybody who can type, if they have
4:44access to a large language model. So,
4:47there's no dancing around that fact.
4:50This is very disruptive. When I was a
4:52graduate student and a young assistant
4:55professor, I would have said that I was
4:58most proud of my works that depended on
5:02the accumulation of knowledge involving
5:05years of effort. I could solve this
5:07paper because a few years ago I learned
5:09this technique and last year I learned
5:10this technique. And here we are. I I've
5:13written this paper.
5:15My view has changed on that. And I hope
5:20that the viewers here think about this.
5:23There's actually something quite hollow
5:26about how I viewed myself as a
5:27mathematician that I only recognized
5:30recently. If my success as a
5:32mathematician relied only on my ability
5:35to learn techniques that somehow could
5:38be put together to prove a theorem,
5:41well, then maybe that was actually
5:43automatable. And maybe I mistook all of
5:47that hard effort for something maybe it
5:50wasn't, right? As hard as it was to
5:53master bodies of work, many papers,
5:56graduate texts, maybe at the end of the
5:58day there is some truth to that being an
6:01automated process. Now, make no mistake,
6:04that's not what we do for a living in
6:06mathematics and in most fields.
6:09In my work now, and this is how I think
6:11about what we do with AI at Axiom, is
6:14supported in a number of ways. The The
6:16cop out would be to say, "Can I ask an
6:19interesting question that I want an
6:20answer to?" But make no mistake, that's
6:22not what research is. Research begins
6:25with a question that you're probably not
6:26able to answer. You try to answer, and
6:29by failing, you learn a little bit more
6:31about that conjecture. And the work that
6:33you do sheds light on a path that might
6:35reveal a long list of questions that you
6:39one by one try to attack, and eventually
6:41you might prove a theorem. And if you
6:43prove that theorem, you backtrack and
6:44say, "Maybe I could have proven this
6:46previous question I couldn't answer."
6:49Right? This is how you learn. It's It's
6:51really the proverbial two steps forward,
6:53one back. And when you recognize that
6:56research is not ask a question, you get
6:58an answer, you realize that the AI tools
7:01are lowering the burden for your ability
7:04to actually perform discovery. When you
7:06were doing all of these homework
7:07problems, when I was doing homework
7:09problems as a college student, as a
7:10graduate student, I was learning
7:12techniques, but was I really
7:14discovering? No. What I was doing though
7:16was important. I was learning how
7:18mathematics fits together to help me
7:21become a mathematician who can ask these
7:23questions and participate in the
7:25discovery. So, that process, I think, is
7:28is changing. For students and faculty
7:32who want to stick to the traditional
7:35ways, well, the reality is in some areas
7:38of mathematics, they will be left
7:40behind. We have computers that can
7:43compute 20 million cases overnight while
7:46you're sleeping and you and it might
7:47take you years to do those 20 million
7:48cases. And you have to decide, would you
7:51like to have that power at your
7:53disposal, freeing you up to participate
7:55in the process of discovery? And that's
7:57what we have to value. So, that's what I
8:00think science is going to become. So,
8:03let me give a concrete example. Typical
8:05person that drove a car today doesn't
8:08have the foggiest idea of chemical
8:11reactions and engineering advances that
8:13had to come to fruition before they
8:15could actually get in the car and drive.
8:17The automobile is an incredible
8:19invention, and it requires mastering
8:21chemical processes and engineering
8:23challenges. All of that is is available
8:26to us now for free. But maybe when Henry
8:29Ford made his first car, he had to solve
8:31all of it. How do I make the wheel? What
8:33do I make tires out of? Today, maybe you
8:35only need to know how to pump gas. Now,
8:39is that bad? No, because think about all
8:41the things that mankind can do now
8:43because they can travel great distances
8:45very quickly. What that opens us up to.
8:48And I think that's going to be our
8:51future. Is that rosy now? No. This
8:55year's horrible. If you ask me, I I
8:56would rather wake up and have it be
8:582017. Given that that is our future, I
9:02think we should do our very best to
9:04encourage people of all professions,
9:07teachers, parents, young students, to do
9:10their best to be prepared to be
9:12flexible, to seek out those
9:13opportunities as they pop up. But I
9:15don't think the loss of of jobs is
9:19anywhere near as significant, and I hope
9:23that remains to be true.
9:25But this is really the time to think
9:27very carefully about education, thinking
9:30about opportunities, and being very
9:32human.
9:41>> So, what makes a good question? There's
9:42several things I want to say.
9:45The first thing is a as a teacher, my
9:48immediate response is there's no such
9:49thing as a bad question. Of course,
9:51that's not quite true. If you genuinely
9:55want to know the answer to a question,
9:58then that's a great question. You should
10:00never ever doubt your interest in a
10:03subject. Okay, but I don't think that's
10:05necessarily what you're asking. Right? I
10:07could ask what is the meaning of life?
10:08That's a great question
10:10on the one hand, but on the other, it's
10:12kind of an impossible question.
10:15Another question is like, I wonder what
10:16I have to do to be rich. I want to be
10:18rich. How do I do it? Well, that is a
10:20question, but is it a great question?
10:22No, I think it's a flawed question in
10:23many ways. First of all, the question
10:25is, well, how do I achieve that? So, you
10:27need to break that down so that a
10:29question becomes maybe a plan, something
10:32that's actionable. But, it's also
10:34somewhat hollow. So, questions that
10:36don't speak to your humanity somehow,
10:39whether it's why do you want to be
10:41richer or what are you going to do to
10:42make the world a better place,
10:44makes that line of reasoning richer.
10:48Now, as a scientist, you might be facing
10:51an open problem that your interest in in
10:54your field cares. Maybe people outside
10:57your field might not care so much. If I
10:59told you about the questions I think
11:01about on a daily basis, I'd be very
11:03surprised that you would care at all.
11:05But, you know, I wouldn't take that
11:06personally. I would start by saying,
11:07"Here's a math problem that I deeply
11:09care about." And I would expect that you
11:11would respect that. If you're in a
11:13situation where you have to think about
11:15whether the question you're asking has
11:18value, I think you should pause and
11:20think about who you're asking the
11:21question for. If you're not asking a
11:22question for yourself, well, my question
11:26to you would be, well, then who are you
11:27living for? Are you living the life
11:28meant for you or are you living a life
11:30that you think someone should be meant
11:32for you, and then my question for you
11:35would be why?
11:45I'm not honestly comfortable talking
11:46about super intelligence because it puts
11:49me at unease. Something that is super,
11:52it means that it's better than others.
11:54And I think what we're really talking
11:57about here is a future and a present,
12:00honestly, where AI is a co-pilot, gives
12:05us tools that we cohabitate with at our
12:07service. So, to say that a computer
12:11could be super intelligent is a bizarre
12:14thought to me because I would never call
12:16my automobile super fast compared to
12:19people, right? Obviously, it's super
12:21fast compared to people. I would I would
12:22have never even thought about it for a
12:24moment. Reducing the load and physical
12:26work is super. The only reason we're
12:29really worried about
12:30super intelligence is that so much of
12:32our identity is based on thinking
12:35skills. Many of the exams I took in
12:37college that I crammed for, did my best
12:40to get a good grade in, only to
12:42recognize that I'd forgotten the facts
12:44maybe by the middle of summer. Yeah, I
12:46did learn something from that, the
12:47process,
12:49but is what I learned the information
12:51that I'd forgotten? No. So, let's not
12:55talk about what is super intelligence
12:57because I don't know what intelligence
12:58is, but I do know quite well when I see
13:02achievement. We live at a time where the
13:06world places so much emphasis on
13:10benchmarks. In sports, I get it. Runner
13:13A runs faster than runner B, they're a
13:14better runner. Okay, that's academic.
13:16But when it comes to assessing
13:18intelligence, do we honestly believe
13:20that someone who gets a higher IQ score
13:23is somehow smarter? Do you actually
13:25believe a school that might be ranked
13:26fifth in the college rankings is really
13:29better than a school that's ranked
13:30seventh, only to turn around the next
13:32year to see that the rankings have
13:33changed. And now you think about how
13:36these AI firms and the state-of-the-art
13:38large language models are competing for
13:40these crazy scores, and we're all caught
13:43up in that. Is any of that intelligence?
13:46No, of course not. But if somebody
13:48writes a poem that just knocks you off
13:51your feet, if someone solves a math
13:54theorem, even if it's with the help of
13:55AI, that represents knowledge mankind
13:58had never seen before, that is
14:01intelligence. Is that superintelligence?
14:04Absolutely.
14:10The easiest way to make a mistake in the
14:12era of AI is to confuse
14:17what people are saying when they're
14:18talking about AI. It's important to
14:20first understand that AI comes in many
14:22different forms. The forms of AI that
14:24most people encounter these days would
14:27be the chat GPT, but make no mistake,
14:30that's only one form of AI. AI's ability
14:33to use machine learning techniques to
14:36conduct a superhuman search that no
14:38person would ever want to do. And this
14:40is how John Jumper and Demis Hassabis
14:42won the Nobel Prize in chemistry for
14:45solving protein folding. It's just
14:47smarter and it and it is accelerated.
14:49And the third part of AI is is where I
14:51think there is so much hope. The third
14:53part of AI is called formalization. And
14:55the idea in formalization is to take
14:58human natural language, transform it
15:01into computer code, which is an enhanced
15:05or at least an exact interpretation of
15:07the human language.
15:09And then have AI study this code and
15:12look for vulnerabilities. It's called
15:14verifiable computer code. We live at a
15:17time now where an enormous proportion of
15:20the computer code that's written and
15:22deployed in the world is not the stuff
15:24of human programmers. It's called vibe
15:27coding. But make no mistake, that code
15:29is not perfect. And so, the space that
15:32we're in now in terms of formalization
15:34is to cut back on those inefficiencies.
15:36And when we start teaching mathematics
15:39or computer science or any field that
15:41has been formalized, we've come to learn
15:43that our original framing of these
15:45subjects was somehow incomplete. So,
15:47I'll give you an example. Our company is
15:50partnering with Scott Coming is a very
15:52distinguished economist at Harvard, a
15:54mathematical economist. And in our work,
15:57we are formalizing as I described for
15:59you before, mathematical theories in
16:02economics. And we've discovered that
16:04some of the foundational theorems in the
16:06subject weren't really accurately
16:09portrayed or implemented or applied. Let
16:13me give you an example. 2026 is the 50th
16:16anniversary of a very famous theorem by
16:18the Nobel laureate Robert Aumann. And
16:20one of his most famous theorems is the
16:23theorem that's called we agree to
16:25disagree. Or can we agree to disagree?
16:29Where the phenomenon is if you have
16:31different parties observing and making
16:34decisions or indicating their preference
16:37preferences based on the same common
16:40prior knowledge, is it possible for
16:42these parties to disagree? And this is
16:45the stuff of modern vernacular. You
16:47might get an argument with a friend, you
16:49listen to each other and you understand
16:50each other's perspective and in the end
16:52it's quite satisfying to say, "Well, I
16:54guess we're just going to have to agree
16:55to disagree."
16:57Aumann's theorem doesn't allow for that.
16:58It can't be that you can agree to
17:01disagree. What really happens is you can
17:03actually end up understanding each
17:05other's perspectives. And that's a very
17:07big theorem. However, there's
17:09subtleties. There're hypotheses. What
17:10does it mean to say you have the same
17:12priors? And that's where the
17:14formalization came in and it's become
17:17kind of a a viral moment in mathematical
17:20economics. Many economists from around
17:22the world are joining our effort,
17:24recognizing that for the sake of getting
17:27economics right, it should be
17:29formalized. And this is happening across
17:32fields. We are even working with
17:33computer scientists rethinking and
17:35formalizing machine learning, which
17:37underlies all of AI to begin with. And
17:39so this is our future. So I said, what
17:42are the opportunities for AI? Maybe
17:44we're worried about the loss of work,
17:46but there are new opportunities. One is
17:49how do we use AI to best guardrail the
17:53other forms of AI? Cybersecurity will
17:56need legions of computer scientists,
17:59also ethicists, make no mistake, and
18:01lawyers who have to rethink or imagine
18:05this new world, right? There going to be
18:07legal issues that come up. And certainly
18:09for the AI experts who are into and
18:13devoted to formalization, that group
18:15will be setting up the guardrails that
18:17will keep us safe. A large language
18:20model is something like the most
18:21incredible librarian, a librarian who's
18:24read everything, but that doesn't mean
18:25you want your librarian to be your
18:26neurosurgeon. In very high-stakes
18:29situations, you need taste. You need
18:31human judgment. And of course, on top of
18:34that, you need someone with the
18:35emotional intelligence to understand how
18:39decisions impact people. Well, all of
18:42those things can be part of
18:43formalization, and I think that's an
18:45opportunity.
18:46And whether you want to help robotic
18:50surgeons be accurate or whether you're
18:53worried about securing the internet or
18:54financial networks,
18:56any system that can be rewritten or is
19:00somehow controlled by a mathematical
19:02language after translation should be
19:06formalized. So, yeah, I think that's a
19:08very big future. And for students
19:11entering college and graduate school, if
19:14you want to be a mathematician, start
19:16formalizing. You may still prove
19:18unsolved conjectures along the way, but
19:21make no mistake. This is 2026, 2027. I
19:25don't believe now
19:27is the race for more compute. It really
19:29should be the race for more truth, and I
19:31think that, and I hope I'm right, will
19:33be by means of formalization.
19:41When a scientist says that a fact is
19:43formally verified, the statement is
19:46true, end of story. If there's a
19:47mistake, it's because you didn't frame
19:50the problem correctly. That's not
19:52judgment. That's a yes-no binary
19:54question. Judgment is how do people,
19:58when given this information, choose to
20:01act? We have autonomous drones flying
20:05all over the world doing all sorts of
20:07things, whether it's keeping track of
20:09traffic in Los Angeles or Seoul, or
20:12whether it's looking for dangerous
20:14people in fields of battle. All of those
20:18situations require judgment. In some of
20:20those low-stakes situations, well, you
20:23know, maybe the drone that's measuring
20:25air quality above Los Angeles, maybe the
20:28human judgment there isn't so important.
20:31But if we're talking about whether or
20:32not to target a city, how do you know
20:35that a building that you're targeting
20:37actually has a dangerous person in it
20:40versus being a school or a hospital?
20:43And I don't actually think it's very
20:44difficult to distinguish situations that
20:47really are so high-stakes that most
20:50rational people would not be comfortable
20:52with letting an AI decide. I think in
20:54most cases that we care about the most
20:57that are high-stakes, when you want a
20:59person involved. Maybe it's not that
21:01easy. We have ride-share services that
21:04are driverless, but people like them.
21:06These opinions and these viewpoints can
21:08change over time, but apart from those
21:10strange situations, I think it's very
21:12clear when you want a human in the room.
21:24We live at a time where the world makes
21:28judgments, snap decisions, snap
21:31evaluations on very little data. It's
21:34crazy. You apply for a job, you're
21:37probably going to submit your cover
21:38letter and your CV or resume to an
21:42automated system that has an algorithm
21:44that has a bunch of check boxes that you
21:47have to predict so that you know that
21:48you're not sieved out in the first round
21:50for no good reason. None of us should be
21:53happy with that.
21:54Everywhere you look, we have adopted a
21:58system where we are replaced by numbers.
22:02We are replaced by what an algorithm
22:04seeks. And this is coming from someone
22:05who works in AI. How can any of us be
22:08happy with that? My children, they're 27
22:10and 30. They're beyond the most critical
22:13phases of getting their career started,
22:15but they knew. And I'll be lying to you
22:17if I didn't say when they were applying
22:19to colleges, as a university professor
22:21myself, I knew a university college
22:24admissions committee is going to be
22:25looking for these 10 things. Make sure
22:27you check those boxes, but then still be
22:30absolutely genuine about what you're
22:32passionate about. Yeah, I would be lying
22:34if I didn't say we didn't do that. But
22:36let's pause and think about what all of
22:38that means because if we buy into that
22:42100%,
22:44then you're forgetting that the quality
22:47of someone's character matters. You're
22:49forgetting that the quality of human
22:51judgment and achievement matters. You're
22:53saying that what matters is can you
22:55check every box and imagine what those
22:57boxes are. And I'm sorry, if you want to
22:59find the cure for cancer, it's not going
23:01to be a bunch of check boxes. If it was,
23:03we would have already found the cure for
23:05cancer. So, the question then becomes if
23:08we live in a society and a community
23:11where we are so rigid because the
23:13computer age allows us to. When I was
23:16starting out, you would look for a job,
23:18you might actually go to a company and
23:20drop off your CV and resume, and shake
23:22the hand of a business owner, and try to
23:24make that human contact. Who does that
23:26now? You probably upload your your
23:28resume and cover letter to a website,
23:31and you might even apply to like 500
23:32jobs. I mean, what what's human in any
23:34of that? My dream for the future has
23:37many pieces to it. One, what I would
23:40give to fight against that so that we
23:42could start a movement where we could
23:44slow down and really evaluate people for
23:47who they are, where they've come from,
23:49what their personal experiences are, the
23:51quality of their character, and how they
23:53interact with others. That would be
23:55awesome. Now, how have I been lucky
23:58enough to identify some of my best
24:00students, the ones that maybe other
24:02schools wouldn't have never taken a
24:04chance on? They were the outliers. I had
24:06a graduate student, his name was Robert
24:08Schneider. He was actually, and still
24:11is, a famous independent rock artist. He
24:13was the producer for a band called
24:16Neutral Milk Hotel, and lead singer for
24:18a band called Apples in Stereo. And he
24:21had the most fascinating story. He loved
24:24equipment. He loved to perform with
24:26these old microphones, solid state old
24:30microphones and speakers when they went
24:32on tour. But because they were old, they
24:35were constantly breaking, and they
24:38needed to be repaired. And it became so
24:40expensive repairing them that he decided
24:42that he was going to start learning
24:43electronics. So, he bought a book.
24:46And the first formula he saw in this
24:48book was Ohm's law. And he said to me
24:51the first time I met him,
24:53and it was the craziest thing. He had
24:55decided to go back to school. He was a
24:58college dropout. He stopped touring. He
25:00went to college, got his math degree,
25:03and found his way into my office. And it
25:05begins with what I what I just described
25:07to you. And when he said when I saw
25:09Ohm's law, it made me stop and think
25:11about what is it that I am producing
25:15when I'm writing and singing music.
25:18Electrical circuits populate my brain.
25:20That's the creative part. I somehow
25:22write down the music on paper, and then
25:24I perform on my guitar to be picked up
25:26by the microphone to go back into my
25:28brain. And all of this was modulated by
25:30an equation called Ohm's law. And I
25:32wanted to figure out how does the
25:34biology work? How does that equation
25:36work? How does the world work? 3 hours
25:39later, said, "You know, you have to be
25:41my student because you made me rethink
25:44everything I thought about mathematical
25:46equations thinking that I knew how you
25:49could find inspiration in math. I never
25:52thought I would have found that story."
25:54So, from Robert to some some of the
25:57other students that I could tell you
25:59about, I'm proud of all of my students.
26:00I've had 35 PhD students. But if if we
26:03were to go through them one by one, I
26:05could tell you a story.
26:07His is just particularly colorful.
26:09What I like about the process is when
26:12they finish their graduate degrees or
26:14when they finish their undergraduate
26:15theses, there's a huge moment,
26:19undeniable. You know it when it happens.
26:21And this is particularly for graduate
26:22students. When you can look at the
26:24student and say, "You know, you're like
26:27a professor now."
26:28And they look back at you, and they know
26:30exactly what you mean. And it's not
26:32because they checked some box. They
26:35fulfilled their thesis. That has somehow
26:37become irrelevant. It's the other part.
26:39So, to answer your question, how do I
26:41recognize that? It circles back to what
26:43I was saying earlier. We have no
26:45shortage of students who mistakenly
26:48think, and it's not their fault, who
26:51mistakenly think that the path to
26:53success is you go to the right schools,
26:55you get the right grades, you fight you
26:57know, you get the right degree, and all
26:59good things will happen to you. That's a
27:02mindless way
27:04of going about one's life. It's not
27:07actually giving yourself permission to
27:09live the life that was meant for you.
27:11It's just saying I'm following a recipe
27:13that we think will be very successful,
27:15and the odds of success are very high.
27:17That's on us. That's on the
27:19universities. That's on us the parents.
27:21That's because we have decided that
27:25there are benchmarks that will evaluate
27:27whether you're successful. Go to the
27:28number five school instead of the number
27:3010. Get the best test scores. Do all of
27:32that. We haven't given enough credit
27:36where credit should be due, and we place
27:38so much emphasis on all of this other
27:40stuff that we're now paying for it. And
27:42we have to fix that right away. I don't
27:44know if this is controversial, but I
27:46think it's all true.
27:52I know what you're talking about. I work
27:55at an AI company, and for the last year,
27:58I work with AI models. I study them. My
28:02wife will say, "Ken, you must have a
28:03relationship with these models." And I
28:05don't think she was wrong. It's
28:07sometimes quite satisfying when the
28:10models start thinking like you do,
28:12because they learn. But I also believe
28:15that if it's not cared for, and those in
28:18charge aren't mindful of its use, it
28:20could be a train wreck. Do you want to
28:22take a trip with your AI? Hey, chat GPT,
28:24here we are. I'm in Rome.
28:27What kind of wine would you like with
28:28dinner? Now, that's not living. I was in
28:31a taxi cab from Incheon Airport to the
28:34my hotel in Gangnam yesterday, and the
28:36traffic was horrible.
28:38Monday 4:00, you can sit for 10 minutes
28:40at a at a block. So, I did a little
28:42experiment. I started counting the
28:44people walk by with their phones in
28:46their hands like this. And it was
28:48something like 70% of the folks here in
28:51Gangnam walking on the street, probably
28:53going home from work, were looking at
28:55their phones like this. That's messed
28:57up. Think about all the opportunities
28:59that you are missing because you think
29:01your world evolves from that little
29:04screen. You might be missing the
29:06opportunity to make a new best friend.
29:08If you find yourself engaging with a
29:12chatbot as if it was really a person,
29:15stop. Put it down. Go for a long walk.
29:19Put yourself in a position where you see
29:21something beautiful or provocative. Do
29:24something that reminds you that the
29:27world before AI has a much longer
29:29history than the world with AI.
29:39As a 58-year-old mathematician, I want
29:42to see some questions answered in my
29:45lifetime. We're already beginning to see
29:48that happen. There are famous examples.
29:51OpenAI a few weeks ago announced a proof
29:54of a theorem called the Erdős unit
29:56distance conjecture, which is a problem
29:58that I thought was never going to be
30:00solved in my lifetime. And post hoc,
30:02meaning when you go back and look at how
30:04this was achieved, the truth is it was
30:06achieved by a little bit of human
30:08collaboration, the mathematicians at
30:09OpenAI with their system, but I don't
30:11think it could have been solved by
30:14people alone unless you had a remarkable
30:18collection of experts from different
30:20fields who somehow came together. I
30:23don't think this would have been the
30:24stuff of one person. And that I think
30:26represents some of the strength and
30:29possibility in AI where think about all
30:33the things in science that you would
30:35like to have solved, and maybe the
30:37accumulated wisdom of mankind can solve
30:39it, but when would you ever be in a
30:41position to put the right people
30:43together in a room to discuss it? So
30:45what AI offers in promise is the access
30:48to the accumulation of human knowledge
30:50tirelessly, and it lowers the bar for
30:53solving these problems. Is it the case
30:56that some of the ideas and solutions are
30:59beyond what humans have ever come up
31:02with? And this is probably the most
31:03provocative point. There are many who
31:06will argue that yeah, AI is going to
31:09come up with ideas, genuinely new ideas
31:12that people have never thought of
31:13before. I don't know that I believe
31:15that. I do believe that AI, computer
31:18systems, can compute more than people
31:21have ever done before, can find patterns
31:24in different areas of science that
31:26humans are unable to do, but the ideas
31:28are somehow already there. Do people
31:31come up with new ideas all the time? The
31:34artwork that you find in Picasso, good
31:36luck finding evidence of that before
31:38Picasso. Do I think AI has that ability
31:42to come up with those new ideas? I don't
31:44know.
31:46Do I hope it does? God, I hope never.
31:55What worries me about what you just said
31:59is this need to compare your personal
32:02situation now with others.
32:05That sounds horrible. If you have to
32:08live up to the standards set by someone
32:10else, then you're not living for
32:12yourself. You're not giving yourself
32:14credit. Whatever pressures [snorts]
32:17someone may feel that inspires them to
32:20constantly be comparing themselves to
32:22others, I consider that toxic. Because
32:26you know what the brutal truth is? The
32:28brutal truth is if you're not LeBron
32:30James or a Nobel Prize winning
32:33scientist, the reality is you will
32:35always be able to find someone that
32:38looks better, achieve something that you
32:40cannot do, and you're not then giving
32:42yourself permission to live the life
32:44that was meant for you.
32:45In my life, I was not a good student in
32:47college. In fifth grade, we had a a math
32:50contest, and I got third. Now, third is
32:53a pretty good out of a fifth grade,
32:55but you probably would have thought Kono
32:57is a famous mathematician, he probably
33:00won easily. No, I got third. In fact,
33:02when I was in fifth grade and I got
33:04third, I thought I let my parents down.
33:06My father was a famous mathematician.
33:08They came to the competition, and son of
33:10famous mathematician gets third in
33:12Hampton Elementary School. And this is
33:15one of those defining moments. On the
33:16drive home, it was just silence.
33:20Mom didn't talk about it. My dad didn't
33:22talk about it.
33:23I thought
33:25for 50 years of my life that I would had
33:28utterly failed. Now,
33:31it's not true that this experience
33:34weighed on me so much that I thought
33:35about it for for decades and decades and
33:37decades, but it was instances like that
33:40where, like you, I was worried about how
33:43I would stack up with others. But, the
33:45reason I bring this up is that when my
33:47dad passed away in January, we were
33:50cleaning up his belongings. There was
33:52very little left because they'd already
33:54downsized to very small apartment in
33:57Florida,
33:58where my parents we had just moved them.
34:01And of all the things that he could have
34:03kept, and it was just in a closet now,
34:05was that plaque from my fifth grade
34:08contest.
34:10And I I thought, "Wow, I had
34:13misinterpreted
34:15that event my whole life." It actually
34:18meant something him to keep a plaque
34:20where I didn't win, but I got third
34:22place. And although he had passed away,
34:24I could never ask him about it, it's
34:26obvious he saw something else. What he
34:28saw, I'm sure, was that I wanted to do
34:31well. So, I hope that's a lesson for
34:33anyone who thinks this way because I
34:36thought that way.
34:37And if you put yourself in a position
34:40where you're always comparing with
34:42others, you might not actually be right.
34:45And you might actually be completely
34:47wrong. So, you have to give yourself
34:49permission to the life that was meant
34:52for you. And this might be morbid, but
34:55one day you will be on your deathbed.
34:57You may only have a few days left. And
34:59someone might ask you some questions.
35:01What are your five deepest regrets? And
35:04this comes up all the time. I'm not
35:06making this up. People, certainly when
35:08you get to my age, you start being
35:09around these kinds of conversations.
35:12And I think the number one regret is I
35:14wish I
35:15had the chance to live the life that was
35:17meant for me. I wish I was able to keep
35:21in close contact with the friends that I
35:23lost touch with. So, try to imagine what
35:26those four or five wishes are.
35:30And at your age, do your very best to
35:32recognize
35:34that you don't want those to be your
35:35regrets. You need some inspiration
35:38often. You need a creative idea often.
35:41And so, encouraging students,
35:44encouraging all people to wonder about
35:46the world that they live in is one
35:48giving permission to think that way. And
35:51wouldn't the world be a much better
35:52place if everyone thought about what
35:55their talents are, gave themselves
35:57permission to be creative? Wouldn't the
36:00world look a lot more interesting
36:02instead of, yeah, I'm supposed to do
36:04this or I'm supposed to do that, so I I
36:07do it. So, I hope that is food for
36:09thought.
36:10I think I've said several times today
36:12that it's important to give yourself
36:14permission to live your life. Now, that
36:17doesn't mean ignore all the signals of
36:20what might help you be successful.
36:22That's not We don't want to be ignorant.
36:24But giving yourself permission to lead a
36:26life that was meant for yourself, also
36:28is giving yourself permission to find
36:30your passion. And that passion might be
36:32something that isn't popular. But if you
36:35find it, you can draw strength from it.
36:37I'm a Japanese kid that grew up in a
36:40very white suburb of Baltimore, Maryland
36:43at a time when it wasn't good to be
36:45Japanese. I wore glasses. I was Mr. Four
36:48Eyes. But that gave me strength. As
36:50difficult as that was, being one of the
36:52only Oriental kids in an all-white
36:54school, being different, I ultimately
36:57drew strength from that. Wasn't easy,
36:59and it probably took 10 years to
37:00overcome that. But whatever demons
37:03AI or culture or family and friends
37:07impose, they don't all have to be there.
37:09And quite frankly, the moral of this
37:11conversation is
37:13there's very little you can do about the
37:14world that's around you. So, how can you
37:17choose a life that's meant for you? Be
37:19flexible and embrace and chase
37:22opportunities that were that seem to be
37:24destined for you.

About this transcript

This page contains the full transcript of A Top Mathematician's 9 Lessons for Anyone Who Feels Behind | Ken Ono, Axiom Math by EO, generated from the public captions YouTube serves with the video. The transcript has 6,205 words across 926 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.