Latest / Trojan Talks / Ep. 3: President Beong-Soo Kim with Eric Schmidt
Transcript
- 0:03Eric, welcome to USC.
- 0:07You wrote a sonnet?
- 0:09I like to write.
- 0:10I like to write sonnets.
- 0:12This is about the Marshall School.
- 0:15Pay attention.
- 0:17When Fortune's wheel
- 0:19doth turn in time swift course ...
- 0:21You don't know what I'm about to say.
- 0:23And commerce seeks new methods to prevail.
- 0:26The wise must harness
- 0:28innovation’s force lest ancient ways
- 0:31grow stagnant, cold and stale.
- 0:34Behold, the thinking engine
- 0:37doth arise.
- 0:39A tool to amplify the student's mind.
- 0:41To parse through data vast
- 0:43as are the skies
- 0:45and patterns in the chaos swift to find.
- 0:47Let Marshall's halls embrace this
- 0:49mechanic art,
- 0:51for business leaders must with times align.
- 0:53The future calls.
- 0:54They must not stand apart, but
- 0:56blend the human touch with the divine.
- 1:00So let USC with boldness
- 1:01lead the way and teach
- 1:03tomorrow's mind what rules today.
- 1:07Wow.
- 1:12Very impressive.
- 1:13It took me a really long time to write that.
- 1:15It took me about one sentence.
- 1:19Welcome to the new world of AI.
- 1:23Amazing.
- 1:23Well, let's just dive right in.
- 1:26When people talk about AI,
- 1:28when they think about AI,
- 1:32especially as it pertains and will impact
- 1:34higher education and universities like USC,
- 1:37I think there are two different narratives: that it's
- 1:40going to be an incredible enhancer
- 1:42of learning and research,
- 1:44and there's another narrative
- 1:46that it's going to completely disrupt
- 1:49the entire university model as we know it.
- 1:52Which narrative is correct?
- 1:54Both.
- 1:56I think almost all of these questions,
- 1:59and by the way, I'm really excited to be here.
- 2:01So I'm just ... have fun with you.
- 2:04What you've done since you took over is remarkable.
- 2:06Look at, look at,
- 2:07I mean, I don't need to sing your praises, but they're well known.
- 2:12Universities are going to adapt
- 2:15because universities, the American universities, 18 of the top 20th,
- 2:1820 universities in the world are in the United States.
- 2:20It’s one of the greatest things about America.
- 2:22They're adaptable.
- 2:24Right?
- 2:24An example is that you have this amazing film school here,
- 2:28and you have a digital tools, a digital, a digital,
- 2:31how to use digital tools, right?
- 2:33Which is the leading program.
- 2:34It's like a little subgroup within a larger
- 2:37super group.
- 2:39That's how your university works.
- 2:41And so you're going to see the same thing
- 2:43with respect to labor allocation, education and so forth.
- 2:46The university will adapt,
- 2:48and the university will not adapt because of the faculty.
- 2:50It'll adapt because the students.
- 2:52Because the innovation occurs essentially from graduate students
- 2:54and postdocs.
- 2:56Right?
- 2:56And the faculty are good enough to allow that to occur.
- 2:59It's always been true in my world
- 3:01that two graduate students and an untenured
- 3:04assistant professor change the world.
- 3:07There's something about that pressure cooker,
- 3:09which USC is very much a part of.
- 3:11That's not going to go away.
- 3:12What will go away is an awful
- 3:15lot of the sort of traditional education and common knowledge,
- 3:19and you'll find ways of having much more self-learning and so forth.
- 3:23And I think that the butts-in-seats kind of model
- 3:26is going to get more ...
- 3:27Most universities are experimenting
- 3:29with various forms of collaborative research,
- 3:31working teams and so forth,
- 3:33because it's a better learning model.
- 3:35I don't think ...
- 3:36So, let me exclude a bunch of things.
- 3:39Teachers are not going away.
- 3:41Universities are not going away.
- 3:42Jobs are not going away. Right?
- 3:44All those predictions are all ill informed.
- 3:46But what you do will ...
- 3:48you will use computers a lot more in everything.
- 3:51And the first class that I hope you
- 3:52introduce after this conference
- 3:54is a class on prompt engineering.
- 3:56How could you have students at USC
- 4:00that do not know how to do complicated prompts to generate
- 4:03whatever it is they're care about, which is often
- 4:04very funny?
- 4:06Do you think there's
- 4:07a greater danger for universities in under-utilizing
- 4:11AI or over-utilizing it?
- 4:14Well, I'm always pro-AI,
- 4:17and the reason is that the arrival
- 4:19of a new form of intelligence
- 4:22that’s at the peer level of humans,
- 4:24which is roughly what's happening,
- 4:26is a very big historical day.
- 4:29This is like, like
- 4:30the arrival of electricity,
- 4:32maybe even more so.
- 4:34We, and I've written two books with Kissinger
- 4:37on this question about
- 4:39what happens when your best friend is non-human,
- 4:44what happens when your collaborator is non-human?
- 4:46What happens when your non-human collaborator
- 4:49is smarter than you are in some areas?
- 4:52The most important thing for this university is to get
- 4:54your students ready for that outcome,
- 4:58because that outcome is a big deal
- 5:01and an awful lot of people who are much older
- 5:03than your students are going to have all sorts of crises over this
- 5:07because it'll affect their identity, what they do,
- 5:09how they spend their time.
- 5:11You share your view that jobs aren't going away,
- 5:13but jobs are certainly going to change,
- 5:16and perhaps very dramatically in ways
- 5:18that we can't really fully envision
- 5:20over the next five to 10 years.
- 5:23How do you think,
- 5:25how do you think universities need to think about
- 5:28how we need to adapt
- 5:30to the changing
- 5:31nature of work, the changing jobs market, in terms
- 5:34of how we're teaching and what we're teaching students?
- 5:37So, and I'll make an obnoxious comment, the most interesting
- 5:41professor is a professor who has tenure,
- 5:44who studies the dynamic nature
- 5:46of the economics of the labor market
- 5:48because they themselves are not subject to it.
- 5:50But I digress.
- 5:53That's not an attack on tenure.
- 5:54Tenure got me here.
- 5:56My father was a tenured faculty,
- 5:58so forth.
- 6:01Let's consider ...
- 6:02So I've spent 20 years with people telling me that truck drivers
- 6:05were going to no longer going to be needed
- 6:07because we would be driving trucks.
- 6:09The number one, there are many job
- 6:10shortages in America.
- 6:12The number one is truck drivers.
- 6:15It's really hard to drive a truck.
- 6:17It's a very difficult job,
- 6:19and not enough people want to do it.
- 6:21How is it possible
- 6:22that jobs are going to take that
- 6:23jobs are all going to go away because of automation and the job,
- 6:27which is most obviously automatable, right, and is difficult
- 6:31and so forth, is the one in the shortest supply.
- 6:34And the answer is because,
- 6:37two things.
- 6:37First,
- 6:39economic efficiency in a,
- 6:41in a country level
- 6:44fundamentally means that the,
- 6:45the firms that adopt AI and automation
- 6:48more quickly become more efficient,
- 6:50which grows the market.
- 6:53They displace people who are less efficient.
- 6:55But as a societal matter, growing the economy
- 6:58and growing revenue, especially in a democracy, is key.
- 7:02Another view on democracy.
- 7:03Europe is essentially not growing.
- 7:05Thank God America is growing at somewhere
- 7:08between 2 to 3%, which is great.
- 7:11In the democratic systems,
- 7:12you can't take money away from its constituents.
- 7:14So if you have new needs, you have to have new money.
- 7:16So democracies have to grow.
- 7:18So you've got to figure out a way
- 7:20to grow your democracy, which is through productivity.
- 7:23And all of the evidence is that
- 7:25technology invention, especially in America,
- 7:29is the key component of American exceptionalism
- 7:33and the, the
- 7:35characteristic of the American culture,
- 7:37which I'll simplify, which everybody knows
- 7:40is relatively freedom focused,
- 7:43relatively suspicious of the government
- 7:46and relatively law abiding.
- 7:48Right?
- 7:49We can debate the specifics, but that's roughly
- 7:52classic liberalism
- 7:54is a necessary component
- 7:56of the kind of growth that we're having.
- 7:58And other cultures, which are also democracies, don't have any growth.
- 8:01They can't do anything.
- 8:02I've tried for 15 years
- 8:04to try to get Europe to do something.
- 8:06Europe is full of incredibly smart people,
- 8:08but the economic system doesn't form it.
- 8:11I was in China, and I mentioned this to you when we just met.
- 8:14I was in China in August, and I came to ...
- 8:17I was originally thinking that China was going
- 8:19to beat
- 8:21...
- 8:21attempt to beat us.
- 8:23China does not have the depth
- 8:25of the financial markets
- 8:27to fund the crazy stuff that American entrepreneurs are doing.
- 8:30You can't randomly raise
- 8:32$50 billion for your data center in China, right?
- 8:35And maybe that government will ultimately decide that's a good idea,
- 8:38but it's a command and control economy.
- 8:39So the genius of America is this restructuring.
- 8:42So the long ...
- 8:43sorry for the long-winded answer,
- 8:45but the result is, yes, there will be more jobs.
- 8:48The jobs will be different.
- 8:50And furthermore, because demographics are such
- 8:52that we're having fewer and fewer children ...
- 8:54And this is worst, by the way, in Korea
- 8:56and in China and other Asian countries,
- 8:58where the replacement rate is well below 1.0, which is terrible.
- 9:01Those are examples of societies
- 9:03that going to need some serious surgery.
- 9:05I don't exactly know why this is occurring, but it's important.
- 9:08We're not going to have the people to do the jobs
- 9:11we need in the next 30 years,
- 9:12which is why universities are so important.
- 9:14You produce the people.
- 9:16I want to get back to this
- 9:18question about how universities need
- 9:20to be rethinking its curricula and how we teach, but
- 9:23you mentioned democracy a couple of times, and I saw that this morning
- 9:27you published an editorial in The New York Times
- 9:30about how you can't
- 9:31outsource democracy to AI.
- 9:33Can you share why you felt compelled to write that?
- 9:37We're in a period
- 9:39where people are frustrated with democracies.
- 9:41I believe in freedom, and I believe in democracy.
- 9:43My political views are clear.
- 9:45And, in order to have a
- 9:50...
- 9:50democracy has to sort of organize itself and succeed.
- 9:53When there are troubles,
- 9:54and we're certainly having them now, there is an appeal
- 9:57to authoritarianism,
- 9:59to the quick solution.
- 10:01And I'm very worried
- 10:03that people will say, let's have the AI do it.
- 10:07After
- 10:07all, AI is much smarter than we are.
- 10:09It can figure things out.
- 10:11It can regulate us.
- 10:12That is a loss of human dignity
- 10:14that I'm very, very worried about.
- 10:16Do not give your human dignity
- 10:19to a non-human activity.
- 10:21You want your freedom to think, to write, to associate,
- 10:24and so forth as a human being, and you want to do it collaboratively.
- 10:28So my basic argument is you need to use AI
- 10:31to cause the communications process,
- 10:34which has largely become, at least in America,
- 10:37he said, she said kinds of tribalism.
- 10:40You know, “you're in this” and “you’re left” and “right” and so forth
- 10:43and so on.
- 10:44There's no constructive conversation.
- 10:46There are solutions which involve using
- 10:48AI to try to encourage the dialogue.
- 10:51And somehow we've forgotten that.
- 10:53I wonder if you can extend
- 10:55some of those ideas to higher education, because one of the things
- 10:58that we think a lot about at USC is that you're not just learning
- 11:02knowledge and acquiring skills, but you're also learning
- 11:04how to interact with other people.
- 11:06You're learning how to get things done as a team.
- 11:10And I'm wondering if one of the implications for
- 11:14AI is that universities,
- 11:16of course, need to figure out how to continue
- 11:18teaching the material, but they also perhaps
- 11:22should consider
- 11:23leaning even more into the social aspects
- 11:26of being part of a university community
- 11:28for, for years
- 11:30or however long they're here.
- 11:32To the degree that you start yourself
- 11:34to interfere in students’ behavior.
- 11:37One, you'll be,
- 11:39shall we say, concerned.
- 11:40And second, it's
- 11:42just better to let students do their thing.
- 11:44Right?
- 11:44Because you spend an awful lot of time getting the, really,
- 11:47the quality of the school force here
- 11:49is, as a private university, is exceptional.
- 11:52Let them figure it out.
- 11:53I'm very
- 11:55bottoms up in this view.
- 11:57And I think that ...
- 11:59so, so ...
- 12:03The professors that I've talked to in other universities
- 12:05have a pretty simple rule: phones off in the classroom.
- 12:09You actually have to talk to each other, right?
- 12:12And you actually have to talk to each other without anonymity.
- 12:16And, there's a long discussion
- 12:18about how social media has destroyed society.
- 12:22I think
- 12:22you're familiar with Jonathan Haidt's book because he was just here,
- 12:26and he's a good friend.
- 12:27A guy named Scott Galloway has just published a book,
- 12:30which is quite concerning, about the status of men in their 20s.
- 12:33And all of these are consequences
- 12:35of excessive social media.
- 12:38Right?
- 12:39And the fact of the matter is, most people
- 12:40and an awful lot of teenagers are pretty lonely, right?
- 12:44And so they turn to these things and they take on a huge
- 12:46superhuman role in their development.
- 12:49Thought experiment for you is somebody
- 12:52here has a child next year or the following year,
- 12:55and every year your child gets a better toy.
- 12:58And the toy can talk and is intelligent.
- 13:00So it's a bear or something like that.
- 13:02Well, that toy will become the child's best friend.
- 13:06Now as a, with a three-year-old,
- 13:07you're going to be really excited about it
- 13:08because it'll keep the three-year-old occupied
- 13:10and it'll teach that the three-year-old language
- 13:12and counting and things like that.
- 13:14But what happens when the kid is
- 13:1612 and has no social skills with the other kids,
- 13:19but really loves the bear and you cannot get the bear with him?
- 13:21I was talking to, to one social psychologist
- 13:25who had a mother talk to him
- 13:28that she had a 15-year-old boy
- 13:32who was ...
- 13:33she was unhappy about and she attempted to take his phone
- 13:37and he beat the crap out of her
- 13:39to the point where she had to go to the hospital.
- 13:42And I said, are we at that stage?
- 13:44And I think, at least in some cases, we are.
- 13:47It's really, really important
- 13:48to maintain human dignity at each of these levels.
- 13:51And here in the university, as well.
- 13:54You've written a number of books and,
- 13:56and you've written a number
- 13:58of articles and papers.
- 14:00And if you're like me, Eric,
- 14:03part of the reason that I write is to figure out
- 14:06what I actually believe and think about a matter.
- 14:10Are you concerned
- 14:12about a future where students are no longer writing?
- 14:14Instead, they're just prompting
- 14:17LLMs to give them answers or produce
- 14:20first drafts,
- 14:20and then maybe prompting them again to, to edit those drafts,
- 14:23but not actually staring at that
- 14:26blank screen or piece of paper
- 14:28and the impact on
- 14:30and the degradation of their critical thinking as a result.
- 14:32So, so I have every belief
- 14:36that the faculty at USC can determine
- 14:39the people who cheated using LLM
- 14:41and the ones who wrote things originally.
- 14:43Right?
- 14:44I'm sure that, that the faculty here can figure that out
- 14:46from a grading perspective.
- 14:47I'm not worried about that.
- 14:49I have long been worried about a reduction in critical
- 14:52thinking because of reduction in reading
- 14:55and, and the ...
- 14:56Instagram and TikTok and so forth.
- 14:58Reading is not the kind of reading I'm ...
- 15:00You're familiar with
- 15:03stage-one and stage-two
- 15:04memory or thinking?
- 15:06The first is quick, and the second is long.
- 15:09And in academia, we call that being in the zone.
- 15:12And so if your students are not in the zone,
- 15:16sitting there doing it,
- 15:19I've been working with a number
- 15:20of research scientists in their 20s,
- 15:24and when I was a research scientist
- 15:25in my 20s, of course, there were no cellphones
- 15:28and we were busy, you know, doing stupid things in college.
- 15:30And we were otherwise reading books
- 15:32and, in my case, writing a lot of code,
- 15:34which I guess is going away as well.
- 15:36And in watching them,
- 15:38they turn off their phone
- 15:41and they just
- 15:42work and are concentrated for a while,
- 15:44and then they take a break and so forth.
- 15:47So maybe the answer to the addiction cycle,
- 15:50because these are all fundamentally addiction devices,
- 15:52is to train your students to turn them off for a while.
- 15:55My trick, my, my newest invention is
- 15:58I put my phone on airplane mode,
- 16:00and I imagine that I'm in an airplane,
- 16:04OK?
- 16:04And I'm so stupid that this works.
- 16:08OK, I'm in an airplane.
- 16:09I look at my phone, it says I'm in an airplane.
- 16:11You know, it works.
- 16:17Dean Garrett
- 16:18talked about an emphasis of mine
- 16:20on really trying to lean into the private sector
- 16:22more and develop relationships not with just private industry,
- 16:26but also with government
- 16:27agencies and actors.
- 16:29I want to ask you, how has AI raised
- 16:32the stakes for us to get that,
- 16:35that set of relationships
- 16:37as, as right as we possibly can?
- 16:39Because right now it's, it's difficult.
- 16:42It's difficult to collaborate with the government
- 16:45and with the private sector.
- 16:47But I wanted to ask, how has AI increased the urgency that we do that,
- 16:51and then the follow-up question is going to be,
- 16:55how do we have to get this
- 16:57right in order to compete with other countries, including China?
- 17:00Well, you guys have a really good
- 17:02math department, OK?
- 17:04And every one of these firms
- 17:05has no idea what AI really is.
- 17:09Right?
- 17:09So I think you're misstating your own negotiating leverage.
- 17:12They need more ...
- 17:14They need you more than you need them.
- 17:18So the AI that I work on
- 17:20is essentially math and most of it ...
- 17:24And I have a piece in this here.
- 17:25I cannot even understand the math anymore.
- 17:27It's moved so fast.
- 17:29And I think that
- 17:31this generation that's coming out right
- 17:33now, which, who will be talking for the rest of the day,
- 17:36are probably the greatest economic
- 17:38potentials of all.
- 17:40And the reason is they're the ones that are going to take
- 17:42these complicated algorithms
- 17:44and essentially apply them
- 17:46to solve real business problems in health care.
- 17:48And there's a long list.
- 17:50In our industry, we,
- 17:51we talk about technology overhang,
- 17:54where we believe,
- 17:56arrogantly, I might add,
- 17:58that we have produced all this incredible technology.
- 18:00And why do customers, businesses,
- 18:02universities and government, why are they taking it more slowly?
- 18:05Now, one answer is it's not quite ready for use,
- 18:08but the real reason is
- 18:09that technology innovation
- 18:11starts in bespoke industries and requires
- 18:15special handling, and then it becomes standardized.
- 18:18I'm old enough to remember that
- 18:19email was considered an option.
- 18:22Now most of you don't even use email.
- 18:25But email was ...
- 18:26And then one day ...
- 18:27So they used to have this diagram where they said,
- 18:29“This is the vertical.
- 18:30This is the horizontal.” And then somebody announced one day
- 18:32the vertical became horizontal, and email became standard.
- 18:36And then all of a sudden the email had to work.
- 18:38So these are simple ways of understanding
- 18:40how technology adoption occurs.
- 18:43And remember, so I was coming by here
- 18:46and I saw Waymo car here in L.A.
- 18:48Great.
- 18:49Now the first self-driving cars
- 18:51were largely invented in 1994.
- 18:55Stanford won
- 18:56the self-driving competition in 2004.
- 18:59Google hired all those people in 2005,
- 19:04and it is now 2025.
- 19:07So it took 20 years from the time
- 19:09a successful contest among universities to this.
- 19:13So the technology diffusion is much ...
- 19:15takes much longer, and the wealth
- 19:17creation occurs at every stage.
- 19:19And since we have a lot of business
- 19:21students here and I'm very pro-business, obviously,
- 19:24remember that the majority of the money is not made at the beginning.
- 19:27It's actually made during the mainstreaming of it,
- 19:30because the mainstreaming creates
- 19:32trillion-dollar industries.
- 19:34And that's where the real wealth will be made.
- 19:35How do you think about China, Eric, in this ...
- 19:41in terms of how universities
- 19:43need to balance, on the one hand, the commitment
- 19:46that is part of our academic mission to sharing knowledge,
- 19:50but also the concerns
- 19:53about intellectual property theft and, and,
- 19:57and we see that those kinds of tensions
- 20:00running through all sorts of debates that we're now having with respect
- 20:04to chips to LLMs and in a lot of other parts of AI?
- 20:08Well, I think, I think you and I agree on this.
- 20:10I'm very strong
- 20:12and positive on foreign students
- 20:14in our American universities, especially at the graduate level.
- 20:17When I was running the National Commission on AI for the Congress,
- 20:20we studied very carefully the Chinese contribution.
- 20:23And it turns out all of the
- 20:24really tippy-top papers in America
- 20:27had one or two out
- 20:29of the five or 10 authors
- 20:30who were Chinese, right?
- 20:32Chinese born, Chinese trained.
- 20:34They are an important part of our research infrastructure.
- 20:37We don't want to get rid of them.
- 20:38Furthermore, we don't get we want to get rid of them
- 20:40after they graduate and send them back to China.
- 20:42That's the stupidest thing ever.
- 20:44So staple a green card on their diploma, you know, whatever ...
- 20:47Whatever it takes, right?
- 20:49You know, you issue diplomas,
- 20:51you have graduation.
- 20:52Just take a green card, you know, and just put it on.
- 20:55Just staple it on. Right?
- 20:57They can write their name in it, right?
- 21:00Yeah.
- 21:00It's obvious, and ...
- 21:03So, why in the world
- 21:07would we want the top, top students in the world
- 21:09to be anywhere other than America?
- 21:11Now, if they're criminals, arrest them.
- 21:14It's not that compli-, we have like a police force
- 21:16and they're pretty good, right?
- 21:17And, you know, you have your own police.
- 21:19What is your problem?
- 21:20You have your own police.
- 21:21You can make your own green cards.
- 21:22You do your own diplomas, like a full-stop shopping.
- 21:25So, so my, my point here is,
- 21:27if you take the right attitude,
- 21:29then America can continue to lead.
- 21:32There are national resource differences.
- 21:34So as I mentioned ...
- 21:36Let me just finish on the China thing.
- 21:38China is going to dominate
- 21:39in a set of areas without question.
- 21:42They dominate, and this was a mistake on America's part,
- 21:45the electronic, electric vehicle and the battery systems.
- 21:49There's no scenario where American firms
- 21:51will be able to compete at the scale they operate.
- 21:54I think, and this is controversial, that
- 21:56that will be true at humanoid robots
- 21:59for the same reasons.
- 22:01I went to some of their factories.
- 22:03They're huge, and these people work incredibly hard.
- 22:06So far,
- 22:07if you look at DeepSeek, which is the most impressive
- 22:10of the research groups, they're good.
- 22:12I mean, they've invented a number of profoundly better
- 22:16algorithms to do what is called reasoning,
- 22:19DeepSeek R1 and so forth.
- 22:21But they don't have the scale.
- 22:22So, at least at the moment, we're OK.
- 22:26But I think as a matter of intellectual property,
- 22:28my preference would be to engage with China
- 22:31because I've, I, and I want to be clear here,
- 22:35China and the U.S.
- 22:36will never be best friends,
- 22:37but we can learn how to work together.
- 22:40And even though the systems are different,
- 22:43both have a common interest
- 22:44in the set of things: domestic stability.
- 22:47We don't want terrorism, nor do they.
- 22:49Economic growth.
- 22:50They say they're communist, but they sure look like capitalists
- 22:53to this capitalist, right?
- 22:56And they work very hard and they're very, very smart.
- 22:59If you have any confusion
- 23:00about the quality of Chinese research, stop.
- 23:03You are wrong.
- 23:04All those prejudices from 15 years ago about they're behind,
- 23:08they’re are no longer true.
- 23:10They are really at the doorstep.
- 23:13You mentioned DeepSeek, Eric,
- 23:14and one of the weaknesses,
- 23:17we'll call it, of DeepSeek, is that when you do query it
- 23:20for information about certain topics,
- 23:23right, you're going to see certain biases.
- 23:26And I was wondering, and there's a whole debate
- 23:30that you're well familiar with about
- 23:33bias and building natural language LLMs
- 23:37and whether it's avoidable or
- 23:39even the right way to talk about the issue.
- 23:42But I want to ask you,
- 23:44are you concerned about a future where people start
- 23:48using the same LLMs and, as a result,
- 23:52you start getting a narrowing of opinion?
- 23:56When universities like ours
- 23:59are always trying to cultivate the broadest
- 24:01range of opinions and viewpoints that we can.
- 24:04So what should you teach?
- 24:07Curiosity.
- 24:10Right?
- 24:11And I'm now at the point where I just can't stand it when people ...
- 24:13and I can tell people are paid to convince me of something,
- 24:17I'm intelligent enough.
- 24:18Just give me the facts, and I can form my own opinion please.
- 24:21And I think there's a real danger in society
- 24:23that people will be
- 24:25captured by an LLM.
- 24:27Remember, it's their best friend.
- 24:29And if your ...
- 24:30your best friend has,
- 24:31I don't know about you all, has a huge impact on, on me.
- 24:34You know, I talked ...
- 24:35in my case, it's a guy.
- 24:36I talk to him, he calls me and I do the thing.
- 24:38You know, he has a big impact on me.
- 24:40He's really important to me.
- 24:43And ...
- 24:44or your spouse, you know, whatever.
- 24:46Whatever it, whatever it is.
- 24:50So part of the answer
- 24:52is that there's not going to be a single LLM solution.
- 24:55So let me give you the way to think about it.
- 24:59The ...
- 24:59there's a, in America, there is a race to AGI,
- 25:03which stands for artificial general intelligence.
- 25:06The leading players there are a startup called Anthropic.
- 25:10OpenAI, which everybody here is familiar with.
- 25:12Gemini, of course, which I, is where I came from, Google.
- 25:17Microsoft is now doing its own.
- 25:19Elon is attempting to enter the space in Grok.
- 25:22Maybe there's a couple of others.
- 25:23These things require an enormous
- 25:25amount of money.
- 25:28Those systems ...
- 25:29and Meta is,
- 25:31not at the moment in the lead.
- 25:32It is probably going to have to have a closed system to do so.
- 25:35Closed means they're not going to give away the source,
- 25:38just because the amount of money
- 25:39involved, they're not going to give it away.
- 25:41The Chinese have strangely made
- 25:44the inverse decision.
- 25:46Every Chinese model is open source.
- 25:49So I have a number of startups
- 25:51which have a high quality of very, very smart Chinese people,
- 25:54and they just use the Chinese LLMs.
- 25:56And they're not political startups.
- 25:58They're, they're just useful and they're free.
- 26:00I said, sure, right?
- 26:03So it's likely that
- 26:05the, in the media, in the next five or 10 years,
- 26:07you're going to see this huge amount of tension
- 26:09about, as we get closer and closer to AGI.
- 26:12And when we hit various
- 26:14thresholds, you know, when something ...
- 26:15We already have systems that can answer graduate-level
- 26:20biology, chemistry, math problems,
- 26:22which I look at them and I say it's inconceivable to me
- 26:25that it can do it and it can solve those.
- 26:27Those are like headline events in my world.
- 26:29AlphaFold, headline events in my world.
- 26:31There will be more,
- 26:34but China will offer those in open source form,
- 26:38and there will be millions of smaller models
- 26:40in open source from, from many countries.
- 26:43So I think you won't have as big a lock in,
- 26:46especially if the university teaches you not to become locked into one.
- 26:50I move around.
- 26:51I try different ones.
- 26:53Do you feel like
- 26:54a lot of the American companies that are developing
- 26:57these LLMs have taken the wrong direction
- 27:00with respect to open source?
- 27:02They didn't really have a choice.
- 27:04I'm, my back-, just to let you know,
- 27:06as a computer scientist, my background was open source.
- 27:07Open source means we literally publish our source
- 27:10code.
- 27:11And I was part
- 27:14of something called Unix,
- 27:16which is the underlying,
- 27:18for example, it underlies the Mac operating system.
- 27:21For those of you who don't know what it is.
- 27:24And it was a great time
- 27:26because we were exchanging ideas, we were working collaboratively.
- 27:29And look what we built.
- 27:31The internet is largely
- 27:32built because of open source architectures and so forth.
- 27:35So there's a lot of power in that.
- 27:38I think
- 27:38realistically, the, so, if you look at ...
- 27:42so for example, Sam, one day called me up and said,
- 27:44you're going to be happy.
- 27:45I'm going to take open,
- 27:47o3 and do a mini-version and release it in open source.
- 27:50And they did.
- 27:50And it's pretty good, right?
- 27:52So I think there's,
- 27:53there will be attempts by the American companies to release subset.
- 27:56One of the core questions for the American companies
- 27:59is the raw models are so big
- 28:02and so powerful, will they ever release
- 28:04the biggest models?
- 28:06It sure looks to me
- 28:08like they're going to release
- 28:10a portfolio of models from super
- 28:13huge, huge, medium range,
- 28:15small, open source,
- 28:17with different pricing points,
- 28:18and it looks like that's what they're going to do.
- 28:20And the reasons have to do with the cost of serving and the cost
- 28:23of the architecture.
- 28:25Eric, you mentioned AlphaFold,
- 28:26and maybe you can explain a little bit
- 28:28about what AlphaFold was able to do.
- 28:30And, and your former company, Google,
- 28:35with Demis Hassabis, Demis
- 28:36Hassabis, has just been doing incredible work
- 28:39in bringing AI
- 28:41to generate scientific breakthroughs.
- 28:44And all of this requires
- 28:46a tremendous amount of compute.
- 28:50What's your advice to universities
- 28:52that want to be at that table?
- 28:57I told you already.
- 28:58Find a Republican state,
- 29:01build a large data center there, and run a big pipe
- 29:03to wherever you are.
- 29:06Sorry.
- 29:07No, no, no.
- 29:08I just want you to say it in front of everyone.
- 29:10Oh, I see.
- 29:11You want me to support you, and you have a
- 29:13Republican state in, in mind?
- 29:15Well, you gave me an idea.
- 29:17Well, I think Nevada is pretty close.
- 29:18Yeah.
- 29:20But, But, but, but you think it’s ...
- 29:22I'm not being ...
- 29:23I'm not being facetious.
- 29:25So, so,
- 29:30So universities
- 29:31need to keep up with this.
- 29:33There is no way
- 29:35the future can be invented in the, in universities
- 29:38without enough computing resource.
- 29:41No university I know has anywhere close.
- 29:44These are the richest universities in the world.
- 29:46I mean, think about the British universities have no money at all.
- 29:49At least, I mean, you guys, you have a lot a lot of money.
- 29:51It's private school. Very well run.
- 29:54And you don't have it.
- 29:55So first, you need to build a big data center.
- 29:58And, and many universities are power constrained.
- 30:01And then what I would do is I would get donors
- 30:04to donate the necessary chips.
- 30:07As you know,
- 30:08Nvidia is 90, 95% of the training market right now.
- 30:11They have amazing chips.
- 30:13And of course, they're now the most valuable company in the world.
- 30:16I'm pleased to report that Google now is working on
- 30:19competitive offerings, and the TPU V7 chip
- 30:22appears to be better than the B-300 chip for most training.
- 30:25Of course, that's widely disputed in the industry.
- 30:28That competition is really important.
- 30:31So the answer to the crazy stuff
- 30:33that's going on in tech is hardcore competition,
- 30:36which will benefit the university.
- 30:38Well, I have your number.
- 30:39So we'll be talking about philanthropic
- 30:41opportunities here at USC.
- 30:42So ...
- 30:45And I ...
- 30:46and I'm happy to help.
- 30:48Just a couple more questions.
- 30:50Wanted to ask you, what do you think
- 30:53one of the most underappreciated opportunities
- 30:56for universities are in the AI space?
- 31:00I think I said it before.
- 31:03The world I live in, I'll just ...
- 31:05I'll just be very direct, is
- 31:07largely dominated by the most brilliant
- 31:1020-somethings, men and women,
- 31:12because the stuff is so new
- 31:15that if you're 40,
- 31:16you didn't go through all those painful classes to learn it.
- 31:20Right?
- 31:20Now, you're smart at 40 and you know you can study it and so forth.
- 31:24But the people who have forged through
- 31:25the crucible of their undergraduate work.
- 31:29At Princeton, I'm from Princeton,
- 31:31I met with a bunch of sophomores
- 31:32who were doing alternative reinforcement learning algorithms,
- 31:35and I go like, what, as sophomores?
- 31:39I'm sure that's true here.
- 31:40So I think people do not appreciate the,
- 31:43the quality of the ideas
- 31:46and the capability of moving quickly with the right tools here.
- 31:49So let's go through what that could mean.
- 31:52In humanities,
- 31:53why do you not have a database
- 31:55that has taken all the documents that matter and made them
- 31:58accessible and classifiable in interesting new ways?
- 32:02In physics, people are
- 32:04using models called diffusion models,
- 32:06which are hard to explain, but basically you add noise
- 32:09and then you reduce the noise and you discover an ...
- 32:12you discover the underlying pattern.
- 32:14Physics is a lot about understanding the underlying pattern.
- 32:18In biology with AlphaFold ...
- 32:20So proteins,
- 32:23so, the primary things that LLMs,
- 32:27what you know, is open ChatGPT
- 32:29do is they do sequence prediction.
- 32:32So what they're really doing is
- 32:33they're doing next word prediction.
- 32:35They just do it really well.
- 32:37And I often tell the story
- 32:38that it's like on a Thursday night in San Francisco
- 32:41and they were sitting, at OpenAI,
- 32:44and these are a couple of my friends.
- 32:46And they decided to turn on what became GPT-3o.
- 32:49So this is, you know, four years ago.
- 32:52And they said,
- 32:53“Oh my God, this thing can write.”
- 32:57They didn't know.
- 32:58It's a real historical discovery.
- 33:01And there's a, there's a moment
- 33:03where that occurred which will ultimately be recorded in history.
- 33:07And what's interesting
- 33:08is we think, “Oh, that's so stupid.
- 33:10Next word prediction?” By the way, how does your brain work?
- 33:13It's called next word prediction.
- 33:15It doesn't use exactly neural nets the same way.
- 33:18And it has other agencies and so forth.
- 33:20But fundamentally it's closer to the way people think
- 33:23than you're willing to admit, speaking as humans.
- 33:26So what I like about, about
- 33:29all of this
- 33:30is that the expansion and the speed
- 33:33that is possible when you take next word
- 33:35prediction and you add planning,
- 33:37which is essentially a series of steps,
- 33:39you can kind of do anything.
- 33:41Here's an example.
- 33:42So I like to buy houses.
- 33:44I like to build houses.
- 33:45I want to buy a house in California, which is always a mess.
- 33:47So the first agent, an agent is input and output.
- 33:50The input is language.
- 33:52It has memory.
- 33:52It has output, which is language.
- 33:54The first agent tells me where to buy.
- 33:56The second agent consults the, the
- 33:59published rules.
- 34:00The third agent designs the home
- 34:02with a known architect, which it selects.
- 34:04An actual human.
- 34:05The fourth hires a contractor,
- 34:07an actual human, to build the house.
- 34:08And the fifth agent sues the contractor when it doesn't work.
- 34:12Right?
- 34:13That's a workflow.
- 34:16This is ...
- 34:16Welcome to my world.
- 34:17Now, you sit there and you go,
- 34:19why is this so ...
- 34:21I mean, why is he giving us such a stupid example?
- 34:23Because you'll remember it.
- 34:25I just described the workflow in a university.
- 34:28I just described a workflow in a hospital.
- 34:30I just described the workflow in government
- 34:33and in businesses.
- 34:35So the agentic revolution, which is what,
- 34:37how we in the, in the computer industry call it,
- 34:40is the next big thing.
- 34:42So another piece of advice
- 34:44for universities is just start building agents.
- 34:47So if the graduate
- 34:48students in computer science, who in this area will be the smartest,
- 34:51just build an agent orchestration framework to run all the agents
- 34:55and tell everyone else in the university
- 34:57to build agents and have fun.
- 34:59Most of the agents will be kind of useless, but that's OK.
- 35:02It's a university.
- 35:03But people will learn what agents are like.
- 35:05So combination of right agents, which the computer can do for you,
- 35:09and then prompt engineering,
- 35:10which allows you to program anything, you're kind of done,
- 35:13right?
- 35:14Think about the power of your labor
- 35:16force, which is here in the room.
- 35:18Fifty thousand total?
- 35:21Boy, I'd take that.
- 35:23Last question,
- 35:24and maybe you started to answer it just now.
- 35:26So I was wondering, Eric, if you could just speak directly
- 35:30to all the students who are listening today?
- 35:33We see all the headlines, you know, jobs are going away,
- 35:38reorganizations.
- 35:39And I think there's,
- 35:43there's reason
- 35:44to be really, really excited.
- 35:45And there's also reasons for our students who, again, as we discussed,
- 35:49span hard sciences to the performing arts
- 35:53to be really worried about the future
- 35:55and how AI is going to impact it.
- 35:57What's your, what's your message to them?
- 35:59There's this problem with smart people
- 36:01that smart people are always depressed.
- 36:03Stop.
- 36:04Stop thinking so hard.
- 36:07I am super jealous
- 36:09of your undergraduates.
- 36:10I want to be very clear.
- 36:12One, you're in California, stop complaining.
- 36:14OK?
- 36:15Like I moved here
- 36:16as a graduate student for the weather.
- 36:19So we're clear.
- 36:20Second,
- 36:22in your lifetimes,
- 36:23you're going to see life extensions to 100 or more.
- 36:27Your children will be infinitely healthier
- 36:29than my generation ever was.
- 36:32Hopefully, the world will get
- 36:35more safe and less dangerous.
- 36:37So far, that's generally been true.
- 36:39There are plenty of studies on this,
- 36:41but most important,
- 36:42you're going to work on the arrival
- 36:44of a new form of intelligence
- 36:47that you can work with.
- 36:48And yes, yes, yes,
- 36:49there's all the movies about how the intelligence gets out of control.
- 36:53And then, you know, the the lady scientist goes
- 36:56and kills the computer and the man is, you know, whatever.
- 36:58Those are fun movies, and you're good at movies.
- 37:01But the important thing is the arrival of intelligence
- 37:03is such a productivity increase, and productivity,
- 37:06I mean, as a personal level, for entertainment,
- 37:10right, for your social life.
- 37:12Right?
- 37:12So start thinking as a 20-something
- 37:16about how you can participate in the agentic revolution
- 37:19and think like a computer scientist.
- 37:21Don't do something for you.
- 37:23Do something for everyone around
- 37:25you and everyone you know and all of their friends.
- 37:29A good rule is,
- 37:31so why are computer scientists doing so well
- 37:33besides, as I mentioned, our own arrogance.
- 37:35Because we think in terms of scale,
- 37:38and because I'm, I'm
- 37:39well known now, a lot of people come up to me
- 37:41with interesting business ideas, but none of them will scale.
- 37:44Come up with a scalable business model.
- 37:46And why is a scalable business model interesting?
- 37:49Because you can make a fortune,
- 37:51which is why you're in business school.
- 37:53Sorry.
- 37:54You see my point?
- 37:56I would be ...
- 37:57You should be, in fact, this is probably why you work here.
- 37:59You want to hang out with them because you're jealous of them, too?
- 38:02Absolutely.
- 38:03But thank you for having me.
- 38:05This is a great honor to kick this off.
- 38:07Eric Schmidt, thank you for your contributions to our society.
- 38:09Thanks for being here.