YT YouTube2Text

YouTube2Text → Transcripts → Alessandro Aresu: "Jensen Huang è l’incontro tra la pazienza confuciana e la velocità della valley"

Alessandro Aresu: "Jensen Huang è l’incontro tra la pazienza confuciana e la velocità della valley" — Transcript

by Chora Media · 7,791 words · 1,244 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00for this bilateral channel to exist.
0:02From this point of view, I wouldn't
0:04judge Trump so negatively, anyway,
0:06because Trump, even with his, you know,
0:08personal vision of "I need a pat on the
0:10back" from Xi Jinping, is still
0:12committed to having a communication
0:14channel with China. Under the Biden
0:18administration, the communication
0:19channel with China wasn't there for a
0:20certain period, I mean, it wasn't there
0:22from a military standpoint, you
0:23understand? which is always extremely
0:26risky. So I hope there will be greater
0:29bilateral cooperation between the
0:31artificial intelligence powers, which
0:34are, well, whether we like it or not,
0:36those two. So, this episode is very
0:45difficult for me, in the sense that it
0:47will then turn into something very easy
0:50because I'll just grab some popcorn and
0:53listen to these two giants on
0:55literature and artificial intelligence.
0:59>> And I'll just do an introduction.
1:00Alessandro is, to me, one of those
1:03figures that Cora always watches with
1:06great attention and considers firmly
1:09within its galaxy. We are like
1:12Bridgewater, right? I mean, it's a
1:14galaxy, and in this galaxy, there's
1:15Alessandro, who for us is just
1:16fantastic. And Raffi is, and I am, the
1:20same thing. Let's start talking about
1:24bottarga, because, well, this is the
1:26moment for bottarga, right? Now,
1:28artificial intelligence has nothing to
1:30do with bottarga. But I want to ask you
1:32a question. In your opinion, with all
1:36this debasement, galloping virtual
1:41inflation, how much will bottarga cost
1:45in 10 years? Well, bottarga will surely
1:50be a safe-haven asset, much more than
1:52gold, because it's the gold of Cabras,
1:54as we certified at the Bottarga
1:56Festival, where I hope next year you'll
1:59also be a guest of maestro Pino Spanu
2:01and the other great artists of bottarga
2:03,
2:04>> because you should know that this
2:06passion unites us and binds us very
2:09strongly to bottarga, which is the most
2:12real thing that can exist and, let's
2:14say, will be a safe-haven asset in the
2:17future, Raffi. I know you're impatient.
2:20Come on.
2:21>> No, Ale, no, I have a question for Ale
2:23right away, because Sandra wrote this
2:25book, The Geopolitics of Artificial
2:27Intelligence, which everyone knows and
2:29so many have read. I have a question.
2:33If he were to write the second volume
2:35of this beautiful book today, where
2:38would he begin?
2:42>> Well, I would always start with history
2:44, so you know, you can't go wrong. In
2:47books about current events, if you
2:49start from historical processes—the
2:51book, for example, has this
2:53chronological start from 1624 with
2:56Bacon's "New Atlantis"—so I am quite
2:59comfortable in the past, because the
3:01past helps you understand the present,
3:04but I would change a few things, I
3:06would enrich a few things about current
3:08events, right? I'm not that kind of
3:11person who says I would do everything
3:13the same in life compared to what I did
3:15two years ago, a year ago. For example,
3:19there are many figures in the Chinese
3:21ecosystem who have crazy stories. Think
3:25of the founder of Muso Teai who used to
3:27be a drummer; he wasn't very good at
3:30playing the drums, but he was in a band
3:32at Tsinghua University. Amodei is
3:36certainly a figure in the artificial
3:38intelligence ecosystem who isn't very
3:39present in my book, Geopolitics of
3:41Artificial Intelligence; he should be
3:43more present. I mean, perhaps from the
3:47things I've read these days, I would
3:49start with this incredible story of him
3:52and his sister who have this kind of
3:54council of little dolls, little animals
3:57they would roleplay to solve company
3:59problems. So I would go hunting for
4:04some stories from the AI ecosystem of
4:07this time, but I would still maintain
4:10my vision in the book about data
4:12centers as satanic mills. Maybe I would
4:18start the book with some data center
4:20protests, those beautiful reports that
4:23Jasmine Sun did in the Midwest on
4:26Substack, and so I would try to adapt
4:29to these dynamics, always giving great
4:32attention to the semiconductor industry
4:34, the theme of infrastructure, and the
4:37epic saga of video games that
4:40characterizes that book's approach. In
4:45this beautiful article on Substack, at
4:47one point you say, "the semiconductor
4:50industry doesn't believe in the
4:52apocalypse because it makes long-term
4:55investments." Can you explain this
4:58concept well to us? Certainly, for me,
5:00artificial intelligence is an evolution
5:03of the semiconductor industry, so I've
5:05always believed what Jensen—who I'm
5:07paying homage to with this leather
5:09jacket, because I have to do some
5:11cosplay—we said nothing.
5:13>> I want to understand one thing. Since
5:15Berzoni is already wearing the jacket
5:17in
5:17>> But I put it on because she put it on,
5:19eh, so I can then blame her
5:20>> But this is such soft leather. And, and
5:23what, what exactly is this jacket? What
5:25is this? This leather from, you know, "
5:28Le vesti di Messalina," that place near
5:31Cavour in Monti. I go there every now
5:33and then because you spend little,
5:35obviously, with many
5:36>> let's say it. I mean, once you come to
5:38Rome,
5:40>> no, because now I've seen that every
5:42once in a while he buys these leather
5:43jackets, you know, those status symbol
5:45ones, a $ 10,000 leather jacket. I'm
5:48always worried because for me it's a
5:51contrary indicator. As soon as he puts
5:54on a leather jacket that's a bit too
5:55cool, it means Nvidia is doing poorly.
5:57Well, that tacky leather jacket
5:59actually means Nvidia still has some
6:02upside. "
6:03>> Sgrauso" is a Roman word, we say "
6:05scrauso," or it's Sardinian.
6:09>> I don't know, sometimes in my mind
6:10these Roman and Sardinian words get
6:12mixed with some Taiwanese word that
6:14pops up suddenly, right? Anyway, to get
6:18back to this point, I've always thought
6:20of artificial intelligence as an
6:22evolution of the semiconductor industry
6:24. The semiconductor industry has this
6:28huge reference market in artificial
6:30intelligence, but the semiconductor
6:32industry, in its various segments (like
6:35memory, for example), has always been
6:37accustomed to boom-and-bust cycles. So
6:41it goes up and it goes down; therefore,
6:44there is no concept of infinite demand
6:46in this industry, okay? You have a
6:49customer—the customer could be the
6:51automotive industry, personal computers
6:54, or industrial and military clients—
6:56at a certain point they tell you, "You
6:58need to produce a lot," and so, that
7:00way, you have to increase supply
7:02because there is so much demand. But
7:04there will come a time when you’ve
7:06built too much, you’ve made too many
7:08factories, and therefore, you won't be
7:11able to use them at full capacity; and
7:13you have to manage to survive that
7:15moment. On the other hand, in AI
7:18ideology, there's this eschatological
7:20vision where you arrive at a market
7:22context where in one year...Well, that
7:24is very powerful in Leopold
7:26Aschenbrenner's manifesto, "Situational
7:28Awareness." Regarding "Leopoldian"
7:32discourse, is there anyone who doesn't
7:34know him? I'll just add a quick note on
7:36Leopold, but I would like you to talk
7:38about this manifesto. Leopold is a very
7:42young hedge fund manager. He worked at
7:45OpenAI first and then he worked, I
7:47believe, with the wife of—with the
7:49sister of Amodei. He starts this hedge
7:52fund, starts with very little money,
7:54and the fund grows to 45 billion.
7:56Raffaele and I talk about it very often
7:57. Blackbox is a fund that at one point
7:59in July goes into crisis; he has to
8:01sell everything while Leopold is
8:03getting married. A crazy story, by the
8:05way. And he sets the world record for
8:07losses in a month. He loses like 50
8:11billion dollars in a month, in a single
8:12month—more than Long-Term Capital.
8:16>> Exactly. And yet Leopold is still...
8:19America works like that. Leopold is
8:22still working, it's not like he's out.
8:25And while he was getting married, the
8:27guests would see him stepping away
8:29because he was—well, many of them
8:31were clients, after all. But what do
8:34you have to say about Leopold's
8:36manifesto? Tell us about it for a
8:38second, because many people don't know
8:39it.
8:40>> This manifesto starts with this
8:42sentence: "You can see the future first
8:44in San Francisco." Uh, you can get a
8:47sneak preview of the future in San
8:48Francisco, and in my opinion, the
8:50future he saw—well, he copied me,
8:52because in a book I wrote that San
8:53Francisco is a harbinger of the future,
8:55but I was talking about the homeless,
8:57fentanyl, and other things.
8:59>> Well, you went deeper than he did. You
9:02saw what it was, you also saw the
9:05contradictions. He—
9:07>> Excuse me, there’s another thing now,
9:09which is the autarky of San Francisco.
9:10>> Yes,
9:11>> meaning OpenAI wants to build
9:13everything in San Francisco.
9:14>> Yes. Well, that’s a bit, you know,
9:16it’s also a bit like the ideas of
9:18prepping and tech entrepreneur bunkers,
9:20right? You make
9:21>> the whole supply chain, they want to do
9:23it in San Francisco.
9:24>> And what does he see? He sees this
9:28gigantic data center build-out process,
9:31and so in 2024, he says this process
9:34will be massive. This is the powerful
9:39part of the manifesto, beyond the
9:40political or geopolitical aspects
9:42contained within it. So he says first
9:46investments in the tens of billions,
9:48then he says hundreds of billions, then
9:50he says trillions, so mega data center
9:52clusters. This idea contradicts the
9:56idea of the semiconductor industry that
9:58we were mentioning before regarding
9:59cycles; that is, for that industry,
10:01there is a cycle, so at some point, the
10:03cycle hits. Whereas for this San
10:06Francisco vision, what Eric Schmidt
10:09later called the "San Francisco
10:12consensus," well, it’s not like you
10:15go to trillions and then go back to
10:18hundreds of billions, right? After
10:22trillions, you go to tens of trillions,
10:24and then you go to other planets, and
10:25after other planets, you go to other
10:27galaxies, and after other galaxies, you
10:29go to other dimensions. So there’s
10:33the famous anecdote, I think it was
10:36reported in the Wall Street Journal, or
10:38maybe The Atlantic, about Leopold
10:40Aschenbrenner and the woman who became
10:43his wife, the chief of staff at
10:45Anthropic—the executive who reports
10:47directly to Dario Amodei—and he said
10:50to her, "I have to think about a
10:52wedding gift, I basically have to buy
10:54you a galaxy in the end," because you
10:57have this, this mentality, right? Where
11:01, you know, you compound things and
11:04after a while you arrive at galaxies,
11:06whereas the semiconductor industry has
11:09this other structure that to me is
11:12symbolized by the Taiwanese ecosystem,
11:14that is, by these old timers, these
11:17guys—well, the average age is 73 now
11:20—who have been there for a long time.
11:24These companies, like Foxconn, in
11:27addition to TSMC, Quanta Computer,
11:29Wistron—these assemblers or those
11:32occupying certain niches in electronics
11:35—are now accustomed to the enormous
11:38demands coming from their leader,
11:41Nvidia, whom they have known for a long
11:44time; they go out to dinner, and they
11:47are told: "You must produce much more,"
11:50but they don't think: "I must produce
11:54infinitely." That is, they always think
11:57there will be a moment when we might
12:00take on water. There are also Asian
12:04variations, let's say almost
12:05anthropological from this point of view
12:08, because the Koreans are a bit
12:10different, because the Koreans also
12:12know that the cycle will arrive, but
12:14they are ruthless, more ruthless when
12:17there is an upside cycle. In fact, they
12:21are the ones who in recent years have
12:22increased their margins in a
12:24frightening way.
12:26>> And we see that in Korean cinema too.
12:28>> Yes, in a truly game
12:31>> truly incredible way, and so, however,
12:33even they know that this moment is
12:35coming. Instead, the idea, if you want
12:39the idea of recursive self-improvement
12:42or other ideas, the ideas of the
12:44singularity, means that there is a
12:46point of discontinuity and so you are
12:48heading towards the point of
12:50discontinuity; you are not heading
12:52towards a new market normality, a new
12:54market normality where things go down,
12:57or something happens. So what Ashen
13:01Brenner sees is always, yes, there can
13:03be small obstacles along the general
13:06path, right? Like when you are climbing
13:09a mountain, at a certain point you
13:12stumble because there is a pebble on
13:14the road, but you see the goal and
13:17>> he hit the big rock, though.
13:19>> And in fact, the risk is hitting a big
13:21rock, folks. I mean, his real problem
13:24was not having a fund, let's say,
13:27closed for five years without leverage,
13:29and there he probably could have taken
13:32these long-term bets and held them, but
13:35with these bets, there's the time
13:38variable. You might even be right
13:41>> long term. But, but, but
13:44>> but Ale, where do San Francisco
13:46accelerationism and this somewhat
13:48old-school vision meet? Taiwanese,
13:52>> eh, Jensen Huang. Jensen Huang. He is
13:57also the meeting point of these
13:59dynamics because inside him, let's say,
14:01both anthropologies, both cultures, if
14:04you will, are present. Both the culture
14:08of the engineer, the electronic
14:10engineer, what he learned in his
14:13training in the 80s before the founding
14:15of Nvidia, and then the relationship
14:18with these companies in the ecosystem.
14:22This supply chain approach is still,
14:25together with CUDA and other elements,
14:27Nvidia's great advantage within this
14:30dynamic. And then Jensen Huang is a
14:34salesman; he is clearly a person who,
14:37as the Pope also said recently, says
14:39one thing at one time and then says the
14:42opposite of that thing. So, like any
14:45salesman, he doesn't respect the
14:47principle of non-contradiction.
14:49>> Well, the podcast with Ezra Klein was
14:51quite evident, wasn't it? Of
14:53contradictions. He made, he made a
14:55terrible impression, in my opinion, as
14:58>> let's tell it,
15:00>> how Well, Ezra Klein is interviewing
15:02many of the personalities in the
15:04artificial intelligence ecosystem;
15:07basically everyone has been on his show
15:09, even Bill Gates, in his, you know, in
15:12his attempt to, if I may say so, regain
15:14his reputation after the Epstein case
15:17by waving the artificial intelligence
15:19apocalypse around, a billion deaths. I
15:23would say it’s a bit sad to watch,
15:25but anyway, let's move on. And Jensen
15:30Huang also called him, in my opinion,
15:32Ezra Klein, because Ezra Klein actually
15:35went to visit him at the Nvidia offices
15:38. They discussed the themes of security
15:43and cybersecurity that have dominated
15:45the discussion, uh, in recent times and
15:47. And he is always on the defensive, he
15:52is in difficulty like so many of these
15:54artificial intelligence leaders; even
15:56Dario Amodei, when you go on television
15:59and, well, you feel sorry for him, uh,
16:01when he speaks, when he intervenes, or
16:03when in the skits at the White House...
16:06well, Trump also adds his own touch
16:08there. Right, but let's tell this one
16:11because, uh, because it makes you
16:13understand these characters well, whom
16:15I call these "terrible kids." Because
16:17this skit at the White House, even a
16:20few days ago, everyone laughing with
16:23Trump, as if they were...he was there,
16:26Zuckerberg.
16:27>> Yes. And as if they were all, exactly.
16:29As if they were three friends here,
16:32like us talking about bottarga,
16:34>> then when they realize that there are,
16:36let's say, cameras rolling, they put on
16:38a mask, it all becomes very
16:39performative. Totally useless, right?
16:44They managed to write a memorandum that
16:47has no no effect, and there was this
16:50also muscular display by Zuckerberg who
16:53seemed a bit like the little boss, you
16:56know, of the whole group, and there is
16:59the crazy nerd figure of Dario Amodei,
17:01right? Yes, who seems incapable of
17:05being in a social context. Social
17:08context,
17:10>> but it is the plastic demonstration of
17:12nothingness and the lack of desire to
17:14do anything on the part of the
17:16administration, therefore of leaving
17:18these intelligence, or rather let's say
17:20American technology, completely free to
17:22their own devices, at least that was my
17:24impression. Yes, in my opinion, in this
17:28dynamic there are also many internal
17:31fights now between the companies that
17:33then reflect on the administration.
17:37Companies have very different positions
17:39. There are the incumbents, what we
17:43call the hyperscalers. Companies like
17:46Microsoft, which are now betting that
17:49OpenAI, in which they are also
17:52shareholders, won't become quite as
17:54successful because they have this idea
17:57that, through their distribution
18:00network and capacity, they will bring
18:03an artificial intelligence that is,
18:06let's say, less powerful or less
18:08apocalyptic than the one promised by
18:11other companies. Or there's Apple,
18:15which wasn't present at the meeting.
18:20>> In the photo opportunity, in the
18:22meeting they had both with Tim Cook and
18:25now with John Ternus, they are clearly
18:28playing a slightly different game: so
18:31no huge capex for data centers or other
18:33activities, waiting for all of this to
18:36have lower costs and then bringing it
18:39through their ecosystem, their
18:41distribution network.
18:44>> This is something that, let's say, we
18:47are always keeping an eye on, right?
18:50This differentiation between large
18:53artificial intelligence inside a data
18:56center—I often use the example of the
18:59F-35 aircraft—and a lighter
19:01artificial intelligence that can be on
19:04a mobile phone, right? A GPU can be on
19:07a mobile phone, a data center is
19:09energy-intensive, it's a satanic mill;
19:11for a mobile phone you plug it into a
19:13socket, and 99%of the requests we make
19:16to our artificial intelligence can
19:18easily be fulfilled by the mobile phone
19:20. It makes you think that these
19:24gentlemen only want data centers to
19:26better train artificial intelligence
19:29and to know exactly what we think and
19:32what we do. There is also data mining,
19:35as they say, right? Data extraction.
19:38Here there is a very significant
19:39concept of differentiation between
19:41decentralized and centralized AI. Apple
19:43, which has its own devices, rightly
19:45seems to me to be following the best
19:47strategy. Now my question is this: why
19:50doesn't Europe, which has always stood
19:52still, do a bit like Apple? Meaning,
19:55instead of this mega data center race,
19:57why not try to build a decentralized AI
20:00in its own way, using public data? Thus
20:03maintaining data sovereignty and trying
20:05to make fewer energy-intensive
20:07investments and such, and more
20:09lightweight investments instead.
20:12>> Well, you can do it in theory, but you
20:14still have to master parts of the stack
20:16. Even to do this, first of all, you
20:19must have semiconductors; you have to
20:22have the chips anyway, otherwise you're
20:24building your data center, right?
20:27Decentralized edge, you're still giving
20:30money to Nvidia or AMD. And those are
20:33the GPUs you put in a mobile phone.
20:34>> Exactly. That, I mean that problem is
20:37insurmountable; you have to invest, and
20:39in any case, there are entities of this
20:41type. For example, there's Axelera,
20:45this now fairly well-funded company in
20:47the Netherlands, founded by an Italian,
20:50Fabrizio del Maffo, who is its CEO. So
20:54this company makes chips, meaning you
20:56need those who design chips and those
20:57who have the chip factories to do all
20:59of this, and then you need your own
21:01distribution network, as he mentions.
21:04>> Which we don't have, so we have neither
21:05the distribution nor the production.
21:08>> We don't have the distribution, nor the
21:10design and production of chips, because
21:13in the chip industry we only have the
21:15part—I mean at these levels—then
21:17other niches exist, but the machinery
21:19part for production like ASML, its
21:21ecosystem, and also the gas part and a
21:24bit of the assembly and testing. But
21:28you still have to close a few supply
21:31chains to do these things, and then you
21:34need the big European private players
21:37to chip in, in terms of significant
21:40resources. So you need a bit of that
21:44investment mentality, even if your goal
21:47isn't to build a gigantic data center.
21:52From that point of view, I can also
21:55agree, but the moment you start down
21:57this path, which is a path that can
22:00have promising aspects, then you must
22:03equip yourself to have the tech stack
22:05to dominate a part of it. Apple is
22:10well-positioned from this point of view
22:12because, for example, in terms of chip
22:14design they are leaders, so they think
22:17immediately, "I'll make the chip to
22:19handle this," and then obviously I have
22:21my distribution network and my
22:23customers.
22:25>> We are entering the usual, shall we say
22:27, war between factions of American
22:28capital, right? Even regarding
22:31artificial intelligence,
22:32>> the ongoing war,
22:33>> the ongoing war because we know very
22:35well that
22:36>> all the venture capital is invested in
22:39open-weight models, and from there
22:41stems this, let's say, media war
22:43against Dario Amodei and everything
22:46Dario Amodei says, right? So, uh, we
22:50are really reaching a reckoning, a
22:53reckoning that, however, has so far
22:56almost proven everyone right, in the
22:59sense that we aren't seeing blood yet.
23:04And then this is Trump's idea, meaning
23:06Trump's idea is that there is never a
23:08reckoning because at a certain point we
23:10all manage to make money, let's say, as
23:12long as we all manage to make money. In
23:16my opinion, that is the argument he
23:18uses; I mean, if I had to imagine what
23:20they said to each other with Danny at
23:22dinner, he said: "Look, don't bust my
23:25chops over this, because you make money
23:27, we make money." I mean, literally,
23:31with this administration's
23:33patrimonialism, and then even those
23:35further from you, like the position of
23:38Palantir or Nvidia, they have to make
23:41money. Even they aren't going to
23:45disrupt all of this. And what we've
23:49seen over these months and years is
23:51that you always need the next big thing
23:54to keep making money for everyone.
23:57Artificial intelligence was just that.
24:01One of the things I've observed is that
24:04when AI arrived—I mean, OpenAI, I
24:06mean Jensen Huang—he proclaimed that
24:09AGI had arrived. This is because, in my
24:14interpretation, Jensen couldn't care
24:16less about the term AGI; there isn't
24:18this background of, "Oh, we're creating
24:21God or similar." For him, it’s the
24:25new target market, because that's how
24:27the semiconductor industry works: you
24:28have a target market, which is video
24:30games, then cars, then other things. Oh
24:33, great. AI makes more money, so we say
24:36it has arrived. Then, when GPT-4o came
24:41out, he said again that AGI had arrived
24:44, so in his mindset, it has already
24:46arrived twice. It’s arrived twice
24:49already, but it could arrive 15 more
24:50times. Listen, I have a question for
24:52you here, because you mentioned it.
24:54What do you think about world models?
24:58That is, this huge—not huge in terms
25:00of money, but very important—
25:03acquisition by AMD of World Labs, from
25:06that Stanford professor, Fei-Fei Li,
25:08which is an alternative model to the
25:11LLM one, right? And let's say the
25:14current market leaders, which don't
25:17really exist yet, are DeepMind, so
25:19Google, and these World Labs folks;
25:21could they be a real competitor,
25:24because in the end, we're talking about
25:26AGI, right? About getting there first
25:29—what is happening there? I mean,
25:32because it's a market niche that nobody
25:34is talking about right now, it almost
25:36seems crazy, doesn't it? Let's remember
25:39—maybe you can explain what it
25:41exactly is—it's basically a way of
25:42reaching artificial intelligence in a
25:44three-dimensional way. Yes. Yes.
25:46>> Yes. You have to simulate a world
25:48instead of
25:49>> spatial intelligence.
25:50>> Exactly. You have to have
25:52>> Let's explain it better.
25:53>> So, you always think about target
25:55markets, right? You think about a robot
25:58or you think about a car; what
26:01artificial intelligence needs to
26:03support, can support, the operation of
26:07a robot or a car? An artificial
26:09intelligence that processes language
26:12and figures out the next word, or looks
26:15things up on the internet? It’s hard
26:19to imagine that working with the
26:22reaction times a robot or a car needs
26:24in the physical world the moment it
26:27encounters an obstacle. So the
26:31underlying idea is: there must be a
26:34program behind the operation of
26:36robotics or automobiles that functions
26:38differently from these large programs
26:41we have, which are evolutions of Large
26:43Language Models, even though they have
26:46further capabilities with agentic
26:48artificial intelligence. And so from
26:52here you divide from those who might
26:55say, for example, Anthropic can say, an
26:58evolution of Claude without being a
27:01World Model, precisely because we scale
27:04so much from data, from the capacity
27:07and personalization of the agent system
27:10and the program can also impact the
27:13physical world and its simulation.
27:17While there are scientists, including
27:20Fei-Fei, who is the so-called godmother
27:24of visual artificial intelligence, what
27:27was image recognition initially with
27:30the AlexNet archives that then led to
27:33ImageNet by Geoffrey Hinton and his
27:36students, or even Yann LeCun, the great
27:39French researcher and former Chief
27:41Scientist of Meta, who say: "We must
27:47create these three-dimensional
27:49simulations that don't always look like
27:52video games, or rather, quasi-videogame
27:55environments through images and videos,
27:59so that the artificial intelligence we
28:02envision is the navigation in these
28:05worlds and, by navigating within them,
28:08we can navigate in our own world and
28:11thus have, for example, a robot capable
28:15of operating with minimal latency and
28:18an ability to move." The bet is that
28:21Physical AI will move toward
28:24>> The bet is that Physical AI will move
28:25toward World Models, that you have to
28:26have a bit of everything. But think of
28:29a company like AMD, I mean, AMD
28:31invested in Generative Bionics, right?
28:36The robotics startup, a spin-off of the
28:39Italian Institute of Technology, so it
28:41has a foothold in...in addition to
28:44having all its capabilities within the
28:46semiconductor industry, it has a
28:49foothold in a company that then makes
28:51robots through its supply chain, and
28:54then buys a leading player, also taking
28:57on Fei-Fei Li as the company's chief
28:59scientist. So AMD now has this
29:03particular characteristic in the
29:05semiconductor technology industry: it
29:08is the only company that has a CEO and
29:10Chief Scientist from there compared to
29:13the others, and says, "Yes, exactly."
29:16One was born in Tainan and one was born
29:19in Chengdu, the city of pandas, the one
29:21from which these new pandas went to the
29:23Atlanta zoo, which will be the main
29:25theme of relations between the United
29:27States and China in the coming weeks. "
29:31And so the ambition is interesting
29:33because it is trying to put together a
29:36stack for physical AI. Then it will buy
29:41something else, maybe it will buy
29:43someone who makes microcontrollers, it
29:45will buy someone who makes sensors and
29:47actuators, because this can help
29:48development and scalability. And
29:51>> it is the true competitor to Nvidia. Eh
29:54>> So, AMD is still a company, in my
29:57opinion, a competitor to Nvidia. And
30:01issues to consider are also how much AI
30:05companies like OpenAI and Anthropic
30:08will invest in the semiconductor
30:11industry because this OpenAI chip seems
30:15to be well-made, it seems to have
30:18significant performance. I also
30:21listened to various conversations from
30:24the head of chip design at OpenAI, so
30:27these companies also want to become
30:30competitors to Nvidia. Then there’s
30:33Broadcom, then there’s, then
30:34there’s always Broadcom. Broadcom is
30:37always in these things, I mean, even at
30:39Trump's dinner, Broadcom was always
30:41there. One never notices it because
30:44there’s this little Malaysian CEO
30:46that one doesn't consider, but he’s
30:48always there. All these companies,
30:51except Nvidia and AMD, always work with
30:53Broadcom, whether it's Google, whether
30:55it's OpenAI and others. Then there is
30:59MediaTek, which is another chip
31:01designer, a Taiwanese one, that is
31:03doing many things with Google. The
31:06advantage of AMD is that, besides
31:09having GPUs, this stack and these
31:11capabilities, AMD has CPUs. I mean, AMD
31:16is important in this market, which for
31:19inference is essential, and Lisa Su, in
31:22her exceptional management of AMD—
31:25because I even wrote this on Substack
31:27because it’s not highlighted enough
31:30—this CEO took a company that was
31:33worth nothing, it was worth about 5
31:35billion in market cap. Now it has
31:39surpassed 1,000 billion.
31:41>> Exactly. When did she join?
31:43>> 2014. She joined in 2014 after a really
31:46bad period for
31:47>> she joined when it was worth 5 billion.
31:49It
31:49>> joined when it was worth 5 billion. So
31:51>> now it’s a trillion.
31:52>> Now a trillion and
31:54>> very well deserved.
31:55>> And anyway, I mean, she did something
31:57crazy. Of course, inside the cycle,
32:00obviously, with the luck that with
32:03inference, CPUs have become important.
32:07So even there, our friend Jensen, right
32:09, before—since I’ve studied him
32:10enough—he used to say," CPUs are no
32:12longer useful for anything, it will be
32:14only GPUs because before it was all
32:16training. "Then, with inference, Nvidia
32:20started making CPUs, obviously, because
32:22Vera Rubin means CPU and GPU together;
32:25that’s why they are a first and last
32:28name: one is CPU and the other is GPU.
32:32But AMD has the expertise in the CPU
32:35market like Intel does, for that matter
32:38, so no, it is a very respectable
32:41player. In my opinion, also due to
32:44sexism, it’s a company that is little
32:45considered, right? one thing on this
32:49>> there has been this big controversy by
32:51Thiel about the Holy Father, we are in
32:53Rome
32:53>> and
32:55>> uh, Thiel basically says:" The Holy
32:57Father has an attitude that penalizes
33:00the West a lot because by slowing down
33:02AI
33:03>> Nick said, however
33:04>> uh
33:04>> say the exact words he used.
33:05>> Go ahead a bit, you no I am not, let's
33:07say,
33:08>> you can't say that
33:09>> no, I can't quite provide the exegesis
33:11of Peter Thiel's text yet. Idiot, he
33:13called the Pope, which is something
33:15>> Yes. He effectively said he's the
33:18useful idiot of the Chinese Communist
33:20Party.
33:20>> He added that later
33:22>> before he had with Region. Yes, yes.
33:25Region.
33:27>> Before he had said he slows down
33:28Western artificial intelligence and
33:30therefore favors China. Aside from the
33:35fact that it's interesting how these
33:38new terrible kids, from 2000 to today,
33:40have become—and we've repeated it so
33:43many times with Raffaele—thinkers.
33:47They have their think tanks, phil-os, I
33:49mean they really want to put their
33:52thinking forward personally, while
33:54before they would create foundations
33:56and think tanks, they would hire
33:58scholars, not like reaching Olivetti
34:01with Volponi, everywhere, right? Today,
34:07personally, you go up to the Pope and
34:09tell him he's wrong, and you say it in
34:12such a direct way, beyond the attitude
34:14which seems obvious, but Thiel could be
34:17right, I mean the question is, the form
34:20is terrible. No, I want, I want us to
34:26ask for my opinion.
34:28>> Eh, but then you have to tell us. Well,
34:30it doesn't surprise me that Thiel uses
34:33this language because he is always very
34:36, I mean if you imagine a provocation,
34:39Thiel goes even further than one
34:41imagines and then he is consistent with
34:44his thinking. He has effectively always
34:49said something like that, certainly
34:51also knowing the character's profile.
34:56This fact that, when he came to Rome,
34:58well, actually some religious figures
35:00intervened almost to excommunicate him,
35:02in quotes, for what he was saying. And
35:06then the Pope, in my opinion, I also—
35:08I had the occasion to write it—I mean
35:11the Pope, in my view (naturally I am
35:13not in the mind of the Holy Father,
35:15though) But, when he cited Tolkien and
35:18that madman Gandalf on*Magnifica
35:20Humanitas*, it's like saying to, I mean
35:23, when he said: Tolkien "comma,
35:25Catholic writer", that landscape is
35:27very important. So he said no, Tolkien
35:31is this, is this thing, right, is this
35:34thing here, that line of Gandalf's and
35:37so I think it stuck with Thiel. Then,
35:41by the way, today Giuliano da Empoli's
35:43book came out, the new one, which from
35:45what I understand will have the figure
35:47of Thiel at the center. So Giuliano,
35:50with his abilities, will also always
35:53give us elements on this. Now the point
35:56of what Thiel says In these figures, in
36:01Thiel and others, there's a particular
36:04hypocrisy regarding this word "the West
36:06" that sooner or later we have to
36:08address. I mean, what is the West? The
36:13West is supposedly them, U.S. companies
36:16of which other people, like those
36:19living in Western Europe, are customers
36:22or subordinates. So there is no "West,"
36:26is there? It’s a bit like David Sachs
36:29, who posted that picture of the dinner
36:31at the White House and called it "The
36:33new Bretton Woods." But excuse me, at
36:37Bretton Woods there were 44 countries,
36:39if I recall correctly. Yet here you use
36:43"the West" to mean the United States
36:45and the fact that you have to be under
36:48the United States. So this must be said
36:51: there is no West in their worldview.
36:56There is this very intense competition
36:58between the U.S. and China, and so this
37:01idea that if one pulls the brake on one
37:04side, the Chinese arrive immediately.
37:09The Chinese threat, from these people,
37:11especially in the military sphere, is
37:12also exaggerated on purpose, in my
37:14opinion, because you basically need to
37:16make investments in defense. And this
37:20is written in my book coming out in
37:222024, but I'll give you a preview. I
37:26mean, if you say that your...you have
37:28to say your enemy is Sauron—he must
37:31be Sauron, your enemy—because there
37:33have to be the Nazgûl, the strongest
37:36of Sauron's troops, because otherwise
37:38you can't assemble your army, which is
37:40literally what these people want to do.
37:44So his worldview is that, it’s very
37:47clear, but we must view it with
37:49critical distance, though Alessandro
37:52responded well, he responded on the
37:55merits. No, we
37:56>> on the merits. On the merits. But I,
37:58but I can tell you: at this moment, in
38:00my opinion, I respect Peter Thiel much
38:02more than many others, because there is
38:05an underlying consistency. And in this
38:08important interview he gave, he said
38:10openly...First of all, he even
38:12criticized the Trump administration
38:14heavily, something that the current
38:16puppets don't do; but then he said: "In
38:182028, the Democrats will win in America
38:20, the socialists will win." He believes
38:24it will be the end of the world, but
38:28>> the end of America.
38:29>> The end of America, sorry, but he has
38:31his own consistency, which clearly we
38:33don't like, also because if we're
38:35talking about the apocalypse and who,
38:38let's say, exaggerates with apocalyptic
38:40tones, I think he's number one, right?
38:44In fact, I wanted to lead to a question
38:46for you: that is, the relationship
38:48between Dario Amodei and Peter Thiel,
38:51who both talk about the apocalypse, but
38:53talk about it in different ways; and my
38:56impression is that, for Peter Thiel,
38:58the Antichrist is actually Amodei,
39:00which is to say someone who wants to
39:02regulate the world, right? He used to
39:06say that, I don't know, about Greta
39:08Thunberg, I believe a few years ago.
39:11Now his target, his main enemy has
39:12become Amodei because of this, let's
39:14say, accusation they make against him
39:16of wanting to perform a sort of
39:18regulatory capture, right? That is, for
39:21him to be the one to regulate the
39:24entire world of artificial intelligence
39:26, and so we have two Americans talking
39:29about the apocalypse in totally
39:31different ways, but then, is the
39:34apocalypse a concept all their own,
39:36because we in Europe don't talk about
39:39the apocalypse.
39:41>> Yes. And they don't talk about it in
39:42China either.
39:43>> No, well, in China it's natural they
39:45don't talk about it because this
39:47concept is, in fact, non-existent. So
39:50you can have, uh, obviously, the
39:52century of humiliation, the Warring
39:54States period, many aspects of Chinese
39:57history, but there is no eschatological
40:00sense, in fact. So, no, the
40:03secularization of theological concepts
40:06—you had also talked about this with
40:08the legendary Walter Siti—is always
40:10very present in our culture and is not
40:13present, however, in the current
40:15European experience. Instead, the
40:20apocalyptic sense in figures like Dario
40:23Amodei derives substantially from the
40:25interpretation of the concept of
40:28singularity. So in that worldview, the
40:33apocalypse is more like extreme
40:36discontinuity. Uh, in fact, it's not
40:41that the world ends, but the properly
40:44human world ends, because a history
40:47hybridized with machines begins, but
40:50not hybridized with a fork or a club,
40:53but hybridized with another
40:55intelligence. And so there is this
40:59enormous transformation, and all this,
41:01however, for Amodei also places itself
41:03in a specific political context. For
41:06example, if you read his various essays
41:08, he insists on this theme in all of
41:10them. Artificial intelligence is a
41:12weapon, first and foremost, it is a
41:14very powerful weapon. This weapon must
41:16be at the disposal of democracies. He
41:20says, he never just says the United
41:22States, he says democracies. Um, so
41:26there must be a world government of
41:28this weapon, but the government must
41:30exclude authoritarian regimes. So Thiel
41:34and Amodei are aligned on this idea
41:36that the Chinese Communist Party is the
41:39enemy. Amodei, who among other things
41:43even knew them, seeing as in San
41:45Francisco he worked for Baidu, the
41:47Chinese Google, but in a previous
41:49period, seeing as for these people,
41:51practically anything goes. However,
41:54within his essays, there are these
41:57other parts that have been read very
41:59carefully and interpreted in China,
42:02where he says: "Artificial intelligence
42:05can provide important tools to the
42:07intelligence apparatuses of democracies
42:10to weaken authoritarian regimes." So he
42:13says,
42:14>> I then ask what his concept of
42:17democracy is, because in the end,
42:19although America is one, I mean, okay,
42:22outside the equation of risks, in my
42:25opinion, there are the biological
42:28weapons that can be created by
42:30artificial intelligence. I remember
42:34very well this book by Mustafa Suleyman
42:36that talked only about that, and it was
42:382022. It is true that this has somewhat
42:41faded out. Amodei talks about it a
42:43little, but Altman never talks about it
42:44.
42:44>> So, exactly. Why isn't there talk of
42:47the physical danger of artificial
42:50intelligence, namely the possibility,
42:52which is becoming uncontainable, of
42:55building biological weapons thanks
42:58precisely to the advancements in
43:00artificial intelligence? I find this
43:04truly dangerous, here too: are we all
43:07just waiting for the big incident
43:09before we truly realize the danger? I
43:14mean, it is clear that now it almost
43:16seems to have become a meme, that of
43:19extinction, but the risks—which are
43:22clearly not today, and probably not
43:25even in 2027—that at some point
43:27groups, even terrorist ones, will be
43:30able to build lethal weapons, they
43:32exist. And how do we contain this risk?
43:39>> Well, the risk is contained by
43:41interrupting capabilities for sure, and
43:44through international collaboration
43:46among the main actors involved in this
43:48ecosystem.
43:52>> The risk is contained in part. The
43:55point is that I believe the
43:56collaboration must happen first and
43:57foremost between the United States and
43:59the People's Republic of China, right?
44:01There had to be this bilateral channel.
44:04From this point of view, I wouldn't
44:06judge Trump so negatively, anyway,
44:08because Trump, even due to his personal
44:10vision of needing to pat Xi Jinping on
44:12the back, is committed to having a
44:14channel of communication with China.
44:19With the Biden administration, the
44:20channel of communication with China was
44:22not there for a certain period; that is
44:23, it wasn't there from a military
44:25standpoint, you understand? Which is
44:27always extremely risky. So, I hope that
44:31there will be greater bilateral
44:33collaboration between the artificial
44:36intelligence powers, which are, well,
44:38whether we like it or not, those two,
44:41first and foremost; then there are also
44:44other actors, South Korea, as we said,
44:47and the others. So, and that is what I
44:51have always thought, it is difficult
44:54for Europeans to impose a cooperation
44:57system on others, because why should
45:00others join a system of cooperation
45:02developed by those who are not relevant
45:05within an industrial supply chain?
45:10Instead, countries that are, and that
45:12see common threats—and I think Modei
45:14is right to bring some of those threats
45:16to light. Anyway, showing for example
45:20that, uh, since all human beings are
45:22intelligent—something we sometimes
45:24forget, right?—maybe Boko Haram takes
45:28, uh, a program and re-engineers it to,
45:31uh, do some very worrying things. The
45:35Houthis, since they know how to do so
45:38many things, uh, are also using, using
45:41these programs. North Korea, which is
45:45certainly a country that pays a lot of
45:48attention to all aspects of a war
45:50economy, including cyber warfare, must
45:53be working very hard to increase its
45:56offensive capabilities from this point
45:58of view, and, therefore, cooperation is
46:01essential. Uh,
46:05>> so it is true that artificial
46:07intelligence is potentially arming a
46:10crowd of bad actors who could truly be
46:13dangerous to humanity, not in terms of
46:15extinction, but in terms of real, major
46:18damage. In my, in my opinion, it is
46:23true that, uh, it is not necessary for
46:25artificial intelligence—meaning these
46:28programs—to be, you know,
46:30civilizations as Ted Chiang called them
46:32in his famous story, right? It is
46:36enough that they are computer programs,
46:39in my view, because we must not forget
46:41how much of our life and how much of
46:43the critical infrastructure of the
46:45world we live in is digital. Even the
46:48power grid is digital, the water
46:51network is digital, arms control
46:54systems have aspects and profiles, and
46:57aspects and digital profiles. So, for
47:00me, what he says might even be correct,
47:03what Jensen says about the fact that:
47:06they are simply programs, they are
47:09codes. But keep in mind how many other
47:13things are codes, almost everything is
47:15a code; therefore, if you manage to
47:18break through many other codes that are
47:20then very poorly defended because the
47:23culture and investments in
47:25cybersecurity in many countries,
47:27including ours, right, are too low. So,
47:31furthermore, this is also an important
47:33investment thesis. It is no coincidence
47:35that we have cybersecurity unicorns,
47:37because it is important that these,
47:39that these capabilities exist. So, in
47:42my opinion, in the medium term, we are
47:45living in a dynamic where these
47:47vulnerabilities, which are real, must
47:49be considered with great attention and,
47:52and concern.
47:53>> Listen, in the great debate going on
47:56right now, let's say, at the NBER, with
47:58this book coming out called "The
48:01Economics of Transformative AI," there
48:03are two schools of thought: the
48:05continuists and the discontinuists. The
48:08continuists say, "Yes, we have here a
48:11normal technology, right?" Like
48:14electricity, so we have the toolbox to
48:16handle it; and the discontinuists say,
48:19"No, this technology is not normal, it
48:21will be so transformative, even in the
48:24relationship between labor, capital,
48:26and work, that we don't have the
48:28toolbox to address it." Where do you
48:31stand?
48:32>> Continuist.
48:33>> Continuist?
48:34>> Yes, I do, yes. I continue to side with
48:39this because I always think there are
48:41levels of complexity that technology
48:44has to face. In any case, these can be
48:49political complexities, they can be
48:51social complexities, so I think the
48:54discontinuists always make the error of
48:56technological reductionism, that is,
48:59reducing other areas of existence that
49:02are important in everything. Think
49:05about the military aspect, right? One
49:07of the things that has made me reflect
49:09the most in recent months is when you
49:11say, "I have overwhelming military
49:13technology.""I, the United States or
49:16Israel, identify the supreme leader of
49:19Iran and take him out.""Ah, okay." And
49:22then, this is the effect of the
49:25technology. Then there are other points
49:28of cultural, anthropological,
49:31organizational complexity, and this
49:34happens in everything, in all fields,
49:37whereas if you think this is the
49:40philosopher's stone and it solves
49:43everything, you miss things, in my
49:46opinion,
49:48>> in fact, right? And you say this is the
49:51thing that solves it for you, but then
49:53another problem begins that you hadn't
49:56foreseen.
49:57>> May I? I'd like to end with Musil, Rob.
50:00So,
50:01>> Ale wrote this beautiful piece on
50:03Substack and for the epigraph he used
50:05this sentence: "If humanity were
50:07capable of dreaming collectively, it
50:09would dream of Musbrugger." Okay, this
50:12is a sentence from "The Man Without
50:14Qualities" by Robert Musil. Musbrugger
50:18was a murderer, and almost as a
50:20comparison, let's say, there was at the
50:22time—in 1914—all of Austria
50:24obsessed with this murder, much like we
50:27might all be watching what's happening
50:29with Chiara Poggi now, right?
50:32>> But what did you mean, why did you put
50:34this epigraph? Because
50:36>> it's not entirely clear to me, I mean,
50:39Musbrugger is human and you put him in
50:41a piece about artificial intelligence.
50:45Okay. Yes.
50:48>> Why would it dream of Musbrugger at
50:50this moment?
50:52>> Because whenever you have a dream, you
50:54also always have a nightmare. Yeah, a
50:56dream doesn't exist without one. When
50:58you dream of paradise, you are actually
51:01dreaming of hell as well. And having a
51:04dream means abstracting yourself from
51:06this earth. In this triad, we have our
51:10conflicts and our contradictions.
51:13Moosbrugger, essentially, in The Man
51:15Without Qualities, is the dialectic of
51:17the parallel action, isn't he? Of the
51:21great ceremony and festival organized
51:23to celebrate the Emperor of
51:26Austria-Hungary, which must be much
51:28more important than the celebrations of
51:31what is called Kakania. So, the whole
51:36game of Musil's is that when you build
51:39buildings, perfect architectures, the
51:41unexpected arrives. And if you are
51:45focused on building your great
51:47architecture and you don't look at the
51:50unexpected, the unexpected becomes
51:52something gigantic and becomes
51:54Moosbrugger, therefore it becomes the
51:57shadow of perfection. In fact, the
52:00shadow of the ceremony is the monster.
52:03In fact,
52:04>> we, Europe, is Kakania. Uh,
52:08>> because he wrote this book in 1930, but
52:11he speaks of 1914, before
52:14>> the First World War, and he describes
52:17this imaginary country, Kakania, which
52:19will then fall and end up in war; yet
52:22there are many problems very similar to
52:24those of Europe, an enormous
52:26bureaucracy, a sense of stagnation. I
52:29mean, do you see any similarities? Why
52:32is this book so important now? Well,
52:36Europe was Kakania, it was Kakania,
52:39then it committed suicide in the First
52:42World War. So we are no longer those
52:46from a century ago, who were the ones
52:49who committed suicide; we are those who
52:52have fallen asleep, and so we are
52:54asleep regarding this matter and we
52:57must wake up, paying attention to the
52:59dreams and nightmares we have so as not
53:02to dream of Moosbrugger. So, to close,
53:06let's close as we used to at Noi
53:08Diavoli, our collective, where we
53:09always say that at the beginning of
53:11every nightmare there is always a dream
53:13.
53:16>> But I really want to thank you because,
53:18right now, uniting artificial
53:20intelligence and 20 th-century
53:21literature with a giant like Musil,
53:24only with Alessandro Areso, is
53:25fundamental.
53:28>> Thanks, Ale. Bye.

About this transcript

This page contains the full transcript of Alessandro Aresu: "Jensen Huang è l’incontro tra la pazienza confuciana e la velocità della valley" by Chora Media, generated from the public captions YouTube serves with the video. The transcript has 7,791 words across 1,244 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 choose a caption track available on the video.

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, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.