YT YouTube2Text

YouTube2Text → Transcripts → Chamath: AGI Is Here, Anthropic’s $100B Giveaway, Elon’s SpaceX Shares for 72M Kids | EP #304

Chamath: AGI Is Here, Anthropic’s $100B Giveaway, Elon’s SpaceX Shares for 72M Kids | EP #304 — Transcript

by Peter H. Diamandis · 22,693 words · 3,630 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00I do think that we have achieved AGI. It
0:02cannot diffuse. And the reason it cannot
0:04diffuse has nothing to do with the
0:06technology. It has everything to do with
0:08the build-up. Until you actually build
0:10it out so everybody can feel it or a
0:12plurality of human beings can feel it,
0:14it's just something that exists in a
0:16lab.
0:16>> Welcome to the singularity. Anthropic
0:19announced the Anthropic Public [music]
0:20Benefit Fund. 5% is somewhere between 75
0:23to 100 billion dollars.
0:24>> I think that these guys kind of walk the
0:26walk and talk the talk. I don't think
0:29that there's something nefarious [music]
0:30there. I think they genuinely believe
0:32>> Next year, I will give one of my SpaceX
0:34shares to every kid in America. This
0:36will be the first [music] of many. There
0:38are 72 million Americans under age 18.
0:41So, that's roughly 10.5 billion in
0:43total, 146 dollars per kid.
0:45>> This [music] is really the better
0:47version of the Giving Pledge.
0:48>> Teaching this entire generation to own
0:51equities is a really good move, I think.
0:53>> Now, the big question, can Trump
0:55accounts contributed from billionaires
0:57and Frontier Labs [music] actually move
0:59the needle to the average American? And
1:01I think
1:04>> Now, that's a moonshot, ladies and
1:06gentlemen.
1:09This episode is brought to you by the
1:11Abundance Summit and Link Ventures.
1:15Welcome Moonshots, everyone. Your number
1:16one podcast in all things AI and
1:18exponential. Your front row seat to the
1:20accelerating singularity. Today, I'm
1:22here with the original crew. AWG, our
1:25in-house ASI. Alex, good to see you. You
1:28changed your orchid today.
1:29>> No, it's just a flood fill.
1:31>> [laughter]
1:31>> Okay.
1:33Dave Blundin, our impresario of AI
1:34investing. And Salim Ismail, our warlord
1:37against linear thinking. Salim, you're
1:39home.
1:40>> I am actually home, thank god.
1:42Loving every second of it.
1:44>> [laughter]
1:45>> And from the All-In podcast, the man his
1:47co-host call Chairman Dictator, Chamath
1:50Palihapitiya.
1:52Chamath, a pleasure to have you on our
1:54crossover episode.
1:56>> Thank you. It's great to be included.
1:58>> Now, at the end of this, I you to tell
2:00me which of the Greek moderators you
2:02prefer more, me or Jay Kahl, okay?
2:05>> [laughter]
2:06>> Okay.
2:07>> No, Jay Kahl is Greek.
2:08>> Definitely you.
2:10>> [laughter]
2:11>> But for the 1% of people who don't know
2:13who you are, let me do a proper
2:14introduction. So, Chamath was born in
2:16Sri Lanka and grew up in Ottawa with
2:18very little, working at Burger King to
2:20help pay the bills, make ends meet. Uh
2:22he became AOL's youngest vice president,
2:25then built the growth team that helped
2:27Facebook to a billion users. And since
2:30founding Social Capital in 2011,
2:32he's been on a rocket ship, purchased
2:3410% of the Golden State Warriors, put
2:36the first check into Grok, which Nvidia
2:38just licensed in a $20 billion deal.
2:41Today, he's a CEO of 80/90, which raised
2:44$135 million this summer to build the
2:46world's legacy software with AI.
2:49Chamath, a pleasure have you on the fun
2:51side of the podcast universe.
2:53>> God, the way the way the way you say it,
2:55it just sounds like it all it all works.
2:58>> [laughter]
2:59>> And what you overlooked there, Chamath,
3:01is shedding all material possessions.
3:03So, the Golden State Warriors, I guess,
3:04is going to be off the docket, right?
3:08>> Uh I I mean, my
3:09parts of my family still own part of it,
3:11but I I can't not.
3:12>> And everybody on Peter Diamandis, your
3:14host in Abundance Projector. Our mission
3:16is always help you understand what just
3:18happened, what it means, keep you
3:20optimistic and abundance-minded as we go
3:22through the singularity. And this is
3:24another crazy week. But before we start,
3:27I just had the most amazing lunch with
3:30William Shatner and his son-in-law,
3:32Joel. So, I spent 2 hours with with
3:35Captain Kirk, my childhood hero. And he
3:37was grilling me about uh about
3:41geothermal energy and what the best
3:43companies for drilling are.
3:45>> Mhm.
3:45>> And I was just so impressed. He's 95
3:47years old, and the guy has got just an
3:50incredible passion for life.
3:52So, and and guess what, Salim? He's
3:54joining us at Abundance Summit this
3:56year.
3:56>> Oh, that's awesome.
3:58>> Yeah.
3:58>> I am shocked that Spock wasn't your
4:00childhood hero.
4:01>> Nah, nah. He was the guy who got the
4:03girls, for sure.
4:04>> Oh, okay.
4:06>> And the one who was like just pushing
4:07the universe forward.
4:09>> Well, Chamath, it's it's incredibly
4:10awesome to get you on the pod here. Um,
4:13you know, back during COVID,
4:15my oldest son, Daniel, came to me and
4:17said, "I just saw the most amazing
4:19podcast ever. There actually four people
4:21on the planet who know what they're
4:22talking about. You've got to check it
4:24out." So, it was like a lifesaver
4:26through that whole, you know, window
4:28where where everyone was trapped at
4:29home.
4:30So, I've been a huge fan ever since. So,
4:32it's great It's great to meet you.
4:34>> Thank you for saying that.
4:34>> Yeah, I think all of us are all in
4:36podcast fans.
4:37>> It's the one podcast I never miss, ever.
4:41>> Really?
4:42>> Yeah.
4:43Yeah.
4:44>> That's a commitment.
4:44>> Because you So, Saline, you miss this
4:46one all the time.
4:47>> I miss all the time. Wait [laughter] a
4:48minute.
4:49>> How does that work? But, I don't miss
4:51that one.
4:51>> Cuz you're someplace on the planet.
4:53>> Yeah. No, sure.
4:55>> [laughter]
4:55>> The two podcasts I'll listen to is this
4:56one when we record it and all-in.
4:59>> Where Where are we?
5:00>> into the Marvel Cinematic Universe with
5:02all-in crossover here. I I think where's
5:04Iron Man? Where's Doctor Strange?
5:06>> I see. [laughter] And then we're going
5:07to have an internal war like Avengers
5:09against Avengers?
5:10>> I know. Inevitably.
5:12>> [laughter]
5:13>> All right. Let's open with a story
5:15that's either the most generous move in
5:16corporate history or the cleverest IPO
5:19marketing ploy ever. So, on Sunday,
5:21Anthropic announced the Anthropic Public
5:23Benefit Fund. Let me read from their
5:25post. Quote, "Subject to stockholder
5:27approval, 5% of Anthropic shares will be
5:29committed today with an additional 2%
5:32added to the board at the board's
5:33discretion for every 50% increase in the
5:36company's value above a set baseline."
5:39The post continues, "It is common
5:41practice for tech companies to offer
5:43escalating compensation awards to their
5:44executives.
5:46We are instead directing that equity
5:48towards public benefit.
5:50So, uh, what is it worth? Anthropic is
5:52heading for a roughly $2 $2 trillion
5:56IPO, though now $1.5 trillion is being
5:59floated. And at that range, 5% is
6:01somewhere between 75 to $100
6:04Uh, now add the founders pledge. When
6:07Anthropic started, all seven co-founders
6:09pledged to give away 80% of their
6:11equity. About half of the employees are
6:13also pledging shares with a matching
6:16grant. Anthropic says all together
6:18roughly 20% of the company is dedicated
6:21to the public good. Placing it somewhere
6:23between 300 to 400 billion dollars.
6:25Uh, compared to OpenAI's uh, Open A
6:28Foundation's value of 130 to 365
6:31billion.
6:32So, um, it's a lot of money going
6:35towards public benefit. Uh, the
6:37reactions on X were instant and varied.
6:39Elon said, "Quote, I recommend giving
6:42the shares directly to the public. This
6:43is what I'm going to do." And more on
6:45this in our second story. AI analyst
6:47Andrew Curran said, "It sounds to me
6:50like they're planning their own form of
6:52self-directed UBI." Brad Gerstner said,
6:55"It's an extraordinary gift." But warned
6:57about quote, "Foundation wealth
6:58concentration." And said, "I hope a
7:00meaningful portion of AI wealth is
7:03directed uh, is direct dividends."
7:05And then Palmer Lucky came out and said,
7:07"This reeks of USA US Aid 2.0."
7:11So, Chamath, uh,
7:12let me go to you first. You're an
7:14investor. So, is 5% donation right
7:18before an IPO generous or is it a
7:21brilliant way to get goodwill?
7:23>> Uh, I mean I mean I think I think it's
7:25none of those things. I think it's more
7:28this is their philosophical bent and um,
7:31you know, they're staying true to their
7:33word. I mean, the idea that when you
7:35start a company you would give up 80% of
7:37the gains even before you have an idea
7:40of what the terminal outcome is
7:42uh, is pretty incredible. So, I think
7:44that these guys kind of uh, walk the
7:46walk and talk the [clears throat] talk.
7:48I I don't think that there's something
7:50nefarious there. I think they genuinely
7:51believe
7:52that what they're doing is the right
7:53thing. I think there's a different
7:55question which is
7:56if you just look at the history of
7:59these sorts of philanthropies and
8:01philanthropically minded efforts, they
8:03generally don't turn out that well.
8:05>> Yeah.
8:06>> Absent of a few.
8:07And so, I think the real question is
8:09what is their fail-safe to course
8:12correct if
8:14this mean reverts and becomes what
8:17you know, everything else in the world
8:19has overwhelmingly looked like when
8:22philanthropically minded people give
8:23large sums of money away in big bursts,
8:25which is, you know, not a lot of
8:28um quantifiable outcome. I think that's
8:30the really big question.
8:31>> You know, one of the critiques I have of
8:33the Giving Pledge is people are pledging
8:36half of their Are you a member of the
8:37Giving Pledge, by the way?
8:38>> No, the Giving Pledge is I
8:40mean
8:40>> Okay, great. Thank you. That's
8:41>> [laughter]
8:42>> So, let me let me be clear.
8:44Let me be direct, right? So, people
8:46pledge to give away half their wealth
8:47while they're still alive, etc., etc.
8:50But the greatest, you know, the greatest
8:51ability you have is not just your
8:53wealth, it's your connections, it's what
8:54you do. I've always said there should be
8:56an impact pledge where people, you know,
8:58pledge their wealth in a measurable,
9:00impactful way and you monitor it and
9:03people make a difference
9:04>> and they came and pitched me about the
9:06Giving Pledge a decade ago.
9:08And my take was this thing is
9:11a scam and the reason is
9:13>> Yeah.
9:14>> it's reputation washing because they had
9:17no
9:18the the only goal was to put my name on
9:20a list and to add my number to their
9:22number so that they could walk around
9:23with a bigger number.
9:25>> [laughter]
9:25>> And uh I was like, this is the stupidest
9:27thing I've ever seen and I'm like, what
9:29do I get in return for all of this? And
9:32uh you know, it's like you get to go to
9:33a dinner.
9:34Um
9:35>> And feel good.
9:36>> There's a weekend. We're like there's
9:37like a get-together weekend.
9:39And so, look, I think that there's a lot
9:41of people that probably thought it was
9:43on the margins the right thing to do,
9:45but it was pretty
9:46pretty obvious to me that it was pretty
9:47stupid. Um
9:49Yeah.
9:50>> Didn't Warren Buffett bail out of that
9:51recently, too? He he signed up.
9:53>> Well, just recently, but he was he was
9:55right there at the front end kind of
9:56like and you know, it's a big deal when
9:58you say no to these guys a decade ago
10:00because like it it looks like you're
10:01doing something wrong and they're do
10:02some they're doing everything right.
10:04>> [laughter]
10:04>> But it But it
10:05>> That's a sign of the times, actually.
10:07There was a lot of social pressure on
10:09many, many fronts uh that has since
10:11course-corrected. But, yeah, that was
10:13that was a one of the first peer peer
10:15pressure, arm twisted. But, if you said
10:17no, you get this negative reputation.
10:19>> here's what I don't understand. What I
10:20don't understand is like if you're
10:22capable of making
10:24>> [snorts]
10:24>> incredible gains in society
10:26economically,
10:27>> Mhm.
10:28>> what it probably says about you is that
10:29you are
10:30at a minimum clever.
10:32Um because, you know, sometimes you can
10:35get lucky, but if you're unlucky,
10:37um afterwards, you kind of give it all
10:40back, you know? So,
10:42in order to sort of like make and
10:44sustain wealth, you have to be
10:45relatively clever, well-timed,
10:47well-read, well-reasoned.
10:50Um you make all of this progress. What
10:52What it says about you is actually a
10:54fair amount. It says that, you know,
10:56you're you're quite a successful,
10:58evolved person.
10:59Doesn't mean that there aren't evolved
11:01people that don't have money, but it
11:02just but what I am saying is that the
11:04people
11:05uh
11:05you know, that that typically tend to
11:07become hyper-successful
11:10are [snorts] functional people.
11:12>> Yeah.
11:12>> And then the idea that those functional,
11:15high-achieving people would all of a
11:17sudden just let some other person that
11:19they don't know to be smart or capable
11:22all of a sudden put a bunch of money to
11:24work sounds stupid. It's like, you're
11:25the smart one.
11:27You should put it to work, you know? At
11:28a minimum, raise your kids because
11:30they're likely to be as smart as you.
11:33So, there's all these ways in which you
11:35can probably think about philanthropy or
11:37give it away like, you know, Elon's
11:39doing, which you'll talk about, I'm
11:40sure, with the Trump accounts.
11:42There's so many scaled ways to do it.
11:43This was the the the one that had
11:45nothing to do with actually achieving
11:47anything and everything to do with just
11:48peer pressure and reputation washing.
11:50>> Yeah, agreed. Uh your greatest gifts are
11:52not your money, it's your intelligence
11:55and your relationships and your ability
11:57to make difference in the world. Uh
11:58Alex, what's your take on these public
12:00benefit funds?
12:02>> Yeah, I think there's an elephant in the
12:04room. There many elephants in many
12:06rooms, and this one was what I've called
12:09a hyper-tithe. Uh the this notion that
12:12frontier AI labs, which many, not all,
12:15think are well-positioned to suck up
12:17much of the value in the economy,
12:19there's a an increasing anchored
12:22expectation that they will donate some
12:25quantum of their equity to the public
12:28benefit. Maybe it's 5% hypothetically.
12:315% is of course also the magic number
12:34that in the US nonprofit foundations are
12:36required to distribute every year. So, I
12:39think there was a latent expectation out
12:41there that Anthropic, probably seeded by
12:44OpenAI and others, would say take
12:47five-ish percent of their equity in
12:49connection perhaps with their
12:51forthcoming IPO, and that they would
12:53donate it to a US sovereign wealth fund
12:56or the 538 Trump accounts or something,
12:59some way for the general American public
13:02to benefit from all of their newfound
13:04liquidity. And the fact, as it appears,
13:07that Anthropic is not doing that, that
13:09instead they're creating a private fund
13:11and not donating 5% to the general
13:14public, I construe as a snub to this
13:18particular administration, a snub to the
13:20idea of a universal basic equity.
13:23Granted, Congress hasn't acted to create
13:26a sovereign wealth fund yet, although
13:28there was some interest from the
13:29executive last year in doing that.
13:31Congress hasn't yet acted. I I think
13:33what the whole world is waiting for
13:36would be, in in my dream scenario, an
13:38active Congress, I I don't think this
13:40could come purely from executive order,
13:43an active Congress that would create an
13:45sort of a middle tier of company that's
13:47halfway between a nonprofit required to
13:50distribute 5% of their assets per year
13:52and a general for-profit that
13:54distributes none of its assets, that
13:56somehow gives a certain percentage of
13:58its equity to the US, to to the public,
14:02to the government or to a sovereign
14:03wealth fund, which also, I would argue,
14:05needs to be created in return for
14:08something. But Congress hasn't done
14:10that. White House, I think with the best
14:12of motivations, tried to put together a
14:15sovereign wealth fund last year, didn't
14:17quite happen yet. And and I think the
14:19question that I have is
14:21in a better world where Anthropic could
14:24have grown into this it's not quite a
14:26public benefit corporation, but some new
14:29class of company that gives up a certain
14:31percentage of equity, not cash and not
14:34free cash flow and not income
14:36to the general public for public
14:38benefit, what would they get in return?
14:40To Chamath's earlier point of the giving
14:42pledge, what would Chamath have wanted
14:45from giving pledge in a hypothetical
14:47better world in return for
14:49>> alone.
14:49>> To be left alone. So, okay, so maybe
14:51>> given him half the money if they just
14:53left me alone.
14:53>> Well, [laughter] the Anthropic version
14:55of being left alone maybe lightness of
14:58regulation which starts to smell like a
15:00safety cartel or some sort of light
15:02touch regulatory treatment, be left
15:04alone by the government, but something
15:05in return for giving up 5% of their
15:08equity. And I think that's the part of
15:09the equation that's still missing.
15:11>> Nice. Saleem, thoughts?
15:14>> Uh two or three thoughts. You know, I
15:16just this whole construct of
15:17philanthropy is still a very scarcity
15:20era response, right? A company becomes
15:22successful, creates wealth, decides how
15:25to redistribute it.
15:27Um I think as we move to kind of an
15:29exponential abundance world, you you
15:32need new models of ownership, uh
15:34participation, and direct access to
15:38services. I think this is the bigger uh
15:41challenge. The good news is this is
15:43actually starting is the beginning of
15:45the discussion of new economic
15:48institutions. And I think that's the
15:49good news around this. It's going to
15:51force that uh conversation. Because you
15:54got to distinguish the philanthropy from
15:56the structural economic reform that's
15:58needed because we're still taxing labor.
16:01And that's a kind of an archaic
16:02structure in a world where AI will take
16:04over more and more of that.
16:06>> Yeah. You know, interesting, Dave, the
16:08uh
16:08the reality is this philanthropic money,
16:12if you want to call it that, is actually
16:13controlled by the for-profit company.
16:15>> Mhm.
16:16>> So
16:17>> Yeah, and And
16:18and I've been on many many many boards
16:20and like when you authorize the board to
16:23contribute another 2% for each 50%
16:25valuation increase, man, did you just
16:27make this a stressful board to be on.
16:30>> [laughter]
16:30>> Like how how are you going to make that
16:31decision? Like you've got the
16:33shareholders over here saying, "No,
16:34don't take the dilution." Then you've
16:36got the public over there saying,
16:37"You're authorized to do it. What's the
16:39decision, guys?" That's that's just a
16:42It kind of punts a really tough choice
16:44down the road.
16:45>> I I think there's going to be so much
16:47demand from the public to own part of
16:49these companies. I think this is just
16:51the very beginning of the process by
16:53which these large hyperscalers start
16:56having sort of an expectation that part
17:00of their company, part of their equity
17:02is going to be distributed. I think this
17:03is just the early days.
17:05>> Yeah, well, for anyone who lost the
17:07thread of the drama here, you know,
17:09remember Anthropic was born out of
17:11disgruntled OpenAI people who didn't
17:13trust OpenAI to be safe with ASI or SI
17:17now. And so, that was their founding and
17:20all the MIT people, brilliant brilliant
17:22researchers that I know that went there,
17:24went there specifically because they
17:26trusted Dario to be that voice of sane
17:29and safe and rational AI. So, that's
17:31that's core. I think much later when he
17:35said, "Hey, I only own 1% of this
17:36company. We need to reclassify our
17:38shares so the the seven of us are super
17:40voting control." That could be spun very
17:43negatively. So, I think now and and you
17:45look at the memes of Dario, you know, he
17:47was never memed in a bad light right up
17:49until recently. Now with the IPO coming,
17:52like look all over X. I mean, he's he's
17:54painted in all kinds of different
17:55lights.
17:56>> He's the devil incarnate.
17:57>> Oh, yeah.
17:58>> It's tough.
17:59>> No, I think part part of the issue is it
18:01it really there isn't an established
18:03regime of broadly established regime for
18:06the US government to tax equity. It can
18:08tax income and it can tax cash flows.
18:12There are there are more exotic ways one
18:15could imagine for the government to tax
18:16and distribute equity itself, but
18:19generally in the US, this is an
18:21unfamiliar concept. More familiar would
18:24be the US would the US government would
18:26tax profits, for example, or capital
18:29gains, for example, and then distribute
18:32via whatever means, including stimmy
18:34checks, cash, free cash to the general
18:37public. But the government being able to
18:39tax equity and distribute equity isn't a
18:43very familiar muscle movement right now
18:45for the US government overall. And I
18:48suspect if we're going to get to a
18:50regime, which many would like to see, of
18:53universal basic equity or universal
18:55basic dividends, being able to tax and
18:58distribute equity, not just cash, I
19:00think would need to become a new muscle
19:02movement. And maybe if that existed,
19:04perhaps Anthropic would be taking a
19:06different position and not just keeping
19:08this 5% as a private benefit fund and
19:11would instead be more open perhaps to
19:13public distribution.
19:15>> I think that there's a there's a really
19:16important concept that was brought up
19:18earlier. What would the I think the
19:19question is what is the right
19:21level of taxation for labor versus
19:23capital?
19:23>> Mhm. [clears throat and cough]
19:24>> Um
19:25>> Yes.
19:26>> And
19:27there probably
19:29is an argument to be made that
19:32capital sh- could be taxed much more
19:36punitively than labor. And that maybe
19:39labor should not be taxed at all. And
19:42how would you make that argument? You'd
19:43probably say that in the
19:45wealth inequality distribution, the
19:47percentage of wealth that comes from
19:49labor versus capital is inversely
19:51indexed to the place in that hierarchy
19:54in which you're in. So, the you know,
19:56first percentile of salaries versus the
19:5899th percentile of salaries. In the 99th
20:00percentile of or of wealth,
20:02you know, the 99th percentile of wealth
20:03is mostly capital, very little labor,
20:06I'm I'm guessing. And the opposite would
20:08be true at the first percentile. And so,
20:10maybe there's something to be said where
20:11you just say, you know, as you go up, um
20:15you have to pay much more incrementally
20:18um
20:19for capital.
20:21Uh
20:23Which, by the way, has has like an
20:24interesting effect which it effectively
20:27is your gate on
20:30profligate money printing by central
20:31banks anyways. So, like if they, you
20:34know, create fake liquidity and that
20:35ends up in your pocket as capital that
20:37you could otherwise deploy,
20:39you know,
20:39charging larger
20:41take rates on that is probably not
20:43unreasonable. Meanwhile, if you tax
20:44labor at nothing, the people whose
20:47wealth is 99% labor all of a sudden
20:49benefit quite deeply. And there's
20:51probably something that's like very
20:52simple there that that mathematically
20:54works and you just have to find the
20:56break-even point on capital. Uh that may
20:58be the simplest and most effective thing
21:01that people could do, I think.
21:03>> I I I agree I agree with that premise
21:05and I think maybe there's a concept that
21:06we're circling which is many would
21:08regard as unpopular, but the idea of
21:11taxing capital more effectively, of
21:13course, historically the problem with
21:15taxing capital is you want to tax the
21:17gain, not the capital itself. If you tax
21:19the capital itself, we're talking about
21:20wealth taxes in in some sense. And part
21:23of the problem is wealth taxes, at least
21:26in many quarters, are wildly unpopular.
21:28So, in if however, arguendo, if a large
21:32chunk of the economy's the American
21:34economy's wealth were concentrated in
21:36just a handful of entities, then
21:38suddenly, not proposing this as a
21:41desirable outcome, but suddenly the idea
21:43of a wealth tax becomes a lot more
21:45practical because the US government
21:47could by hook or by crook say to say all
21:49of the frontier labs, I want to tax a
21:51few percent of your wealth in return for
21:53following XYZ positive treatment. I I
21:56suspect that's where this may go, but I
21:58guess we'll find out.
21:59>> Mhm.
22:00>> Well, let's move on to the
22:01>> slippery slope. It's a very slippery
22:03slope to try to find define wealth.
22:05>> This episode is sponsored by Google for
22:06Startups. Think about this for a second.
22:09You now have access to the same
22:11generative AI models that cost hundreds
22:13of millions of dollars to train.
22:15Google's Startup [music] Technical Guide
22:17for Generative Media gives you a
22:19complete blueprint for deploying Google
22:21DeepMind's models in production. Images,
22:24video, audio, all of it. Real
22:26architecture, real results.
22:28Find the link in the show notes below.
22:32Let's move on to the second shoe to
22:34drop. So, Elon just in criticized
22:36Anthropic. Within hours, he answered.
22:39So, early Monday morning on X, he said,
22:41"Next year, I will give one of my X my
22:43SpaceX shares to every kid in America.
22:45This will be the first of many."
22:47He had 520,000 likes and 27 million
22:50views in less than a day. President
22:52Trump responded, quote, "A big thank you
22:54to our legendary friend at Elon Musk for
22:57his incredible generosity in giving
22:59SpaceX stock to millions of American
23:01children through Trump accounts.
23:03This is exactly what was envisioned when
23:05we created Trump accounts, bringing
23:07together the great private sector, Elon
23:09and so many more to invest in our
23:11children, by far our greatest, most
23:13important asset, and give every American
23:16child a piece of American dream. Thank
23:18you, Elon." So, let's do the math for a
23:20second. There are 72 million Americans
23:22under age 18,
23:24and SpaceX current stock price about 148
23:27bucks today. So, that's roughly 10.5
23:29billion in total, 146 dollars per kid,
23:33about 1.1% of Elon's stake being shared.
23:37Uh for context, last December, Michael
23:39and Susan Dell gave 6.25 billion, uh
23:42about 250 bucks exactly, to Trump
23:45accounts for 25 million children. Unlike
23:47the Dell gift of cash, Elon's
23:49transferring equity, and honestly, you
23:52know, ownership beats handouts. I think
23:54we would all agree with that, and a
23:55paycheck gets spent and shares compound.
23:59So, we previously talked about SpaceX
24:01hitting 10 trillion, 100 trillion
24:03dollars, and Elon's recent prediction
24:06that SpaceX could become, quote, "worth
24:08orders of magnitude more than the
24:09current Earth economy."
24:11Um the replies on X were classic. The
24:14top reply with 30,000 likes was, quote,
24:17"I identify as a kid in America."
24:19So, now the big question, can Trump
24:21accounts contributed from billionaires
24:23and Frontier Labs actually move the
24:25needle for the average American? And I
24:28think the answer is probably no. So, the
24:30average US household is pulling in
24:33$87,000 a year and to pay every American
24:36just 10,000 a year takes about 2.6
24:39trillion per year.
24:41So, anthropic and Elon's philanthropy is
24:43really just, you know, kind of a
24:45gesture, not a funding source. Even the
24:48biggest fortunes on Earth are just
24:50one-time paychecks.
24:52So, Chamath, you know, you've spoken in
24:54the past about a birth dividend. I love
24:56that idea in which the government invest
24:58$2,000 for every newborn in the market,
25:02locked in until age 65.
25:05I'm wondering if Elon's going to put a
25:06lock-in requirement on his on his shares
25:09as well. I hope he does, so it's not
25:11just, you know, liquidated and spent
25:14right out the bat. So, what do you think
25:16of Elon's move here, Chamath?
25:19>> I think it's great. I think that this is
25:21really the better version of the giving
25:23pledge.
25:23>> Yeah.
25:25>> I think that this is the this is the
25:27right way for most of us who want to
25:30contribute and don't have the
25:31infrastructure or the capacity to
25:33understand how to discern what is a good
25:35nonprofit from a bad nonprofit from a
25:37good NGO from a bad NGO.
25:40Um
25:41direct contributions into the accounts
25:43of kids who can then be a participant in
25:46the
25:48experiment that is capitalism and
25:49democracy in the United States is a
25:51no-brainer. I think it's I think it's
25:53it's like
25:54unbelievably impressive and unbelievably
25:56generous and I think that it has a
25:59it has a lot of legs.
26:00>> Yeah, it's it's great. I my kids, every
26:03dollar they earn, they split out half of
26:05it into saving it into into stocks.
26:08They've all put it into SpaceX. I wonder
26:09where they got influenced like that.
26:11And then savings and and philanthropy
26:14and then their spending budget. Dave,
26:17I'm curious, 72 million people now own
26:19SpaceX shares at the end of this.
26:22Is that a risk if they dump it? You
26:24know, what's that going to do to the
26:25stock? Uh
26:26>> No, no, no, it's 1%. Like you like you
26:28said, it's not going to move the needle
26:29if they all dump it. But I think the uh
26:33the reason this is brilliant is because
26:35we've all discussed on this pod before
26:36that the future of value comes through
26:38asset ownership and not through
26:39employment, not through paychecks cuz
26:41what is AI not going to do 2 years from
26:43today, 3 years from today? I was over at
26:45MIT actually yesterday talking to a
26:47bunch of under, you know, undergraduates
26:49and like, your window of opportunity to
26:51tap your big MIT brain is about the next
26:542 years.
26:55But do you really think that AI can
26:57outthink you in every conceivable way
26:59more than 2 years from today? I mean, it
27:01like that is your window of opportunity.
27:02You got to move fast. But that doesn't
27:04mean wealth goes away. It all moves to
27:06asset value. So So teaching this entire
27:08generation to own equities
27:10>> Yeah.
27:10>> is a really good move, I think. And And
27:13yeah, if they dump it, well, then their
27:14friends will tell them they're idiots
27:15for dumping it when SpaceX, you know,
27:17Elon, you said a second ago that this
27:19doesn't move the needle on the needs of
27:21Americans, but Elon believes SpaceX is
27:23going to be bigger than the current
27:25total global economy.
27:26>> Yeah, 1 to 10 quadrillion dollars was
27:29the estimate.
27:30>> Yeah.
27:31>> It becomes a
27:31>> maybe he's right, maybe he's wrong, but
27:33but that'll keep these kids engaged with
27:35each other. The ones that dump it will
27:37be talking to the ones who didn't.
27:38Something good is going to come out of
27:39that conversation. I I think it's really
27:41cool.
27:41>> Saleem, your thoughts, pal?
27:43>> Um Chad, very quickly, just to build on
27:45what Dave said, I think the fact that
27:47you get all of these kids looking at at
27:50the Now you have broad participation in
27:52in the appreciation of capital, I think
27:54is a really good thing. I think that's
27:56the conversation that's going to make
27:58the much, much bigger difference, like a
28:00100x compared to the actual dollar
28:01amount, because now all these kids are
28:03thinking in that modality as they grow
28:05up, watching this, comparing it, seeing
28:08who sold, who didn't sell. It's awesome
28:10as a conversation.
28:12>> I also think this might set a trend, you
28:13know, what what if all other public, you
28:16know, newly public companies that want
28:18to be cool with the next generation just
28:20follow the same trend. Then then, you
28:22know, you get 50, 100, 200 copycats,
28:24then it really does move the needle.
28:26>> Yeah.
28:26>> Yeah, I think it moves the needle and
28:27>> I don't know if you guys have worked in
28:28companies that have had um ESPP plans,
28:32but employee share purchase plans where
28:33you contribute on a you know, quarterly
28:35basis and at the end of the quarter they
28:37buy it at they buy you that defined
28:39[clears throat]
28:39amount of stock at the
28:42low point of the quarter.
28:43Um there's all of these mechanisms that
28:47induce you to buy the stock of the place
28:49that you work to your point. So, if if
28:51all work workers and employers
28:54embrace this, I think it's a big deal. I
28:55also think that this starts to start
28:57this begins the very difficult contribu-
29:00conversation about moving from defined
29:01benefits to defined contributions.
29:04And it shines a light on the fact that
29:07these defined benefit retirement
29:09programs are insufficient and they're
29:10poorly funded. And um when you see on
29:13the other side, you know, people will
29:15say, "Well, why can't we do more forms
29:16of defined contributions?"
29:18And that'll be the right conversation.
29:20And I think it you have to have a
29:21starting point. So, I think that Trump
29:23accounts and, you know, Elon's gift, I
29:25think starts that conversation. Which is
29:27very important.
29:28>> Alex, you know, Elon said this will be
29:30the first of many. Uh I'm curious if
29:33this is going to start setting
29:34expectations by all the public of
29:36Frontier Labs.
29:38>> Yeah, maybe. I I think other than the
29:40obvious universal basic equity angle, I
29:42think it's also potentially very clever
29:44marketing. Imagine an entire generation
29:46of Americans who feel economically
29:48aligned with the outcome of presumably a
29:51merged SpaceX Tesla. Suddenly everyone
29:54owns equity not in the general market,
29:57which arguably would be a a a slightly
29:59better outcome for democracy, but
30:01everyone owns a share in one company.
30:04That this is the one company that every
30:06American owns a share of and now feels
30:08economic alignment with, maybe
30:10regulatory alignment. Think of all that
30:13I was out in California in Santa Monica,
30:15Peter, your neck of the woods, and
30:17walking down Santa Monica Place, and
30:20there was one of these protests, call it
30:24protesting Elon right in front of the
30:25Tesla dealership there. Imagine an
30:28alternative America
30:29>> of public of Santa Monica, for sure.
30:31>> Yeah, imagine an alternative America
30:33where every person in a generation owns
30:35an economic stake in Elon's prolific
30:38economic output. Would that be a very
30:40different political America? I think it
30:43probably would be.
30:43>> I think it's a brilliant move. I you
30:45know, I texted with him uh about 4
30:47months ago. I was trying to see if he
30:49would fund a $10 billion
30:52actually
30:53$10 X prizes and he goes, "Interesting,
30:57but I'm going to give my shares away."
30:58And so he's been thinking about this for
31:00some time. Um Alex, you just mentioned
31:03and it's in the news today that Musk
31:05says Tesla and SpaceX have formed
31:06special committees to discuss the
31:08merger. So that game is a foot, not a
31:11surprise.
31:12>> Isn't that the obvious next move after
31:14the Tesla Roadster? I we didn't get a
31:16chance to to talk about it as a full
31:18group, but in my mind the Roadster
31:21announcement that the car hops, it
31:23doesn't hover. What was the point of
31:25that? In my mind, it was 90%
31:28just to to secure or help to secure a
31:31Tesla SpaceX merger.
31:33>> Yeah.
31:34For sure.
31:35Well,
31:36I I I love this and and kudos to Elon
31:40for for making this happen.
31:42Um
31:43you know
31:45it's one other point, you know, Elon's
31:47talked about universal high income and I
31:50think the only way this materializes if
31:52everybody gets a piece of the pie and
31:54grows with it, right? They couple to the
31:57singularity. They don't sort of like go
31:59off the off-ramp here.
32:02Anyway, um
32:03so our next story I want to hit on is a
32:06three-part series of clips coming from
32:09the Walter Isaacson interview with Elon
32:11today
32:12published on Builders. In our first
32:15clip, Isaacson asked Elon
32:17you know, whether we're heading towards
32:19a post-capitalist world. Elon said
32:22money's going to stop being relevant at
32:24some point in the future.
32:26>> future look like
32:27>> money becomes irrelevant. I mean
32:29there will be still currency, but I
32:31think currency will be measured in
32:32fundamental physics metrics like um
32:35you know, power, wattage and tonnage.
32:38mass and energy
32:40You measure power in joules.
32:42So,
32:43um
32:46But eventually currency will not matter
32:47in the future. So, that that's that's
32:50that's my guess. I think it's pretty
32:51obvious.
32:52Um and in in the sense you start getting
32:55too many orders of magnitude bigger than
32:56than Earth's economy,
32:58um satisfying whatever humans can
32:59possibly think of for their needs is is
33:02very easy. You know, if you're several
33:04orders of magnitude bigger than the
33:05Earth's economy, um
33:07then
33:08providing humans with whatever goods or
33:10services they can think of that they
33:11might want is is trivial.
33:14>> And so, what do we do then?
33:16>> As humans.
33:17>> What gives us meaning? How do we
33:19organize ourselves?
33:20>> I mean, you struggle with this question.
33:22People will will will find
33:25ways to be happy
33:26in the in the same way that there were
33:28people who retire seem to be relative
33:29you know, fairly happy even though
33:30they're not working. But I think the
33:32reality is that you humans will not
33:34They won't They won't be anything that
33:35humans can do that that they the AI and
33:37the robots can't do better.
33:38>> So, you know, if currency's being
33:40measured in watts and tons, whoever owns
33:42the power plants, the mines, and the
33:44rockets owns the money supply.
33:47Uh and today it's a pretty short list.
33:48So, Chamath, you've locked up what,
33:50nearly 6 gigawatts of power? Um do you
33:53feel like you're holding, you know,
33:55humanity's future currency?
33:58>> Yeah, I'm very aligned with what he's
33:59saying. I think the the different way of
34:01thinking about this,
34:03um less eloquent, but um may make sense
34:06is that at some point you can only
34:08debase a fiat currency so much until it
34:10becomes fundamentally worthless and you
34:11have to shift to an alternative way of
34:14measuring value. And if you instead move
34:17the measurement to what is the
34:21uh implied energy or the implied um
34:25effort
34:26uh that that currency gives you, I think
34:29that that's a more symbolic
34:31representation of future value. So, you
34:33know, take a cryptocurrency that takes,
34:36I don't know,
34:38n number of watts to computer, you know,
34:41n number of joules or whatever, that's
34:44a more valuable thing in the future than
34:47pick your fiat currency on which, you
34:50know, it's just completely uncoupled to
34:53any form of value. So, if this one asset
34:55actually has, you know, some amount of
34:57effort that was burned to make it, and
34:59the other one was just a decision by a
35:01bureaucrat, I think the latter becomes
35:03far less useful. So,
35:06the the question is probably like what
35:08what How do these assets trade in a more
35:10fungible way and as a mechanism of
35:11exchange? And I I think
35:13typically or I I think if I had to
35:15guess, it would be more like you have
35:17things like data centers, things like
35:19power plants, things like turbines,
35:20things like net gas supplies.
35:22So, you have this effectively like
35:24embedded compute, embedded energy, and
35:26around that you have some sort of like
35:28crypto token that represents some unit
35:30of that access, and that's how you
35:31exchange value.
35:32>> Interesting. Um,
35:34you know, he's talked about a
35:36post-capitalist society for some time.
35:38You know, he's famously said, "Don't
35:39save for the future."
35:41Do you Do you sort of concur with that?
35:44>> Well, I think I think what I would say
35:46is maybe slightly different is like
35:47yeah, I think like saving in the
35:49conventional sense is what we've all
35:51been taught to do. There hasn't been a
35:53bridge that's been created where you can
35:55save in an alternative sense and these
35:57other new forms. So, a different way
36:00where one could save in a in a new form
36:03of value would be where when you have
36:06enough local compute, um, you
36:08participate in the economy in a way
36:11where now you actually do represent
36:12value. Um, and that's different in a in
36:16a traditional sense where the work that
36:18which we get paid for today effectively
36:20goes away because it's done largely via
36:22AI, and so the compensation for that
36:25work then changes materially. But on the
36:27other hand, if you have resources that
36:29you can put into different pools to
36:31accomplish different tasks,
36:33that I think will have value. So, that
36:35is sort of how I think people would will
36:37make that transition.
36:39>> You know, Alex, we've been talking on
36:40the pod for a while that GDP sucks as a
36:42measure. So, we're we're to get a
36:44Kardashev scale as a new measure of
36:46of wealth for a country?
36:48>> Maybe. The premise of the Kardashev
36:50levels though is energy output, and I'm
36:52not convinced I'm not as convinced as
36:54perhaps others are that energy wants to
36:57be the ultimate unit of currency. I
36:59suspect it'll be something more exotic
37:02or to Elon's point entirely irrelevant
37:04and not even the right question. I have
37:06conversations with AIs or SIs all the
37:09time about what the future capitalism
37:12looks like, what the future of economics
37:14looks like. My working hypothesis is
37:16that 30 years from now it will just look
37:19unrecognizably different and we'll look
37:22back on conversations like, well, is is
37:25energy, is land, is powered land, are
37:27these the right units of currency? We'll
37:29look back and we'll laugh at these
37:31discussions.
37:31>> Intermediately they will be.
37:32Intermediate values.
37:33>> Oh, I totally agree with you, but
37:35>> The the terminal the terminal, who
37:36knows? I totally agree with you.
37:38>> Yeah. I I I suspect
37:39>> By the way, the GDP thing I agree with
37:41you as well. GDP
37:41>> is so so dumb. So basic.
37:43>> Elon loves his little parable of people
37:46passing feces back and forth or or
37:48whatever and calling that GDP.
37:49>> let that hang. You got to tell the
37:51actual story. We we have it in the past.
37:53Look at the Elon interview podcast, you
37:55can get the
37:55>> I I I shorthanded Elon but politely. But
37:58I I I I I do think that if I had to
38:01guess what does the successor call it
38:03capitalism 2.0 Charlie Stross style to
38:06money look like, I suspect imagine a
38:09world imagine a world where 30 years
38:12from now
38:13the ability to make intelligent resource
38:17allocation and other allocation
38:18decisions is a lot easier. Right now
38:21it's hard to intelligently allocate
38:24capital, so it all gets pushed to the
38:26edge and we we live in an eye pencil
38:28world where all of the decisions live at
38:30the edge. Imagine alternatively, without
38:33this being misconstrued as some sort of
38:35centralized planning or communist
38:37utopian vision, imagine a world where
38:41human civilization, thanks to
38:42superintelligence, is just so smart that
38:45it's able to model its own needs and
38:47doesn't need to push resource allocation
38:49decisions to the edge. So, the notion of
38:52a currency as a way to do distributed
38:54resource planning doesn't make any sense
38:56and instead intelligent decisions I
38:59don't want to say can be centralized
39:00because I think that's probably also the
39:02wrong metaphor, but they can be done
39:04>> technological
39:06come
39:06>> No, I I don't want to imply centralized
39:08planning. I want to imply that humanity
39:10as a whole becomes so super intelligent
39:12that all of civilization is capable of
39:15making to the extent anything remains
39:17scarce is capable as a whole of making
39:19those decisions and as a result money
39:21changing hands between individual actors
39:23makes a lot less sense.
39:25>> Mhm. Dave, you were saying
39:26>> Well, Kardashev, you know, we're talking
39:2830 years in the future for everything to
39:30be in space, but between here and 30
39:32years
39:33it's mostly terrestrial and it's we're
39:35not going to get another chance to have
39:36a guy who has 6 gigawatts locked up
39:39to pick his brain. I mean, this is such
39:41an incredibly brilliant investment. But
39:44we know, you know, from I don't know
39:46Chamath if you're invested in Positron
39:47or LightQ or we have our company
39:49LightWave, there's a lot of efficiency
39:51coming to data centers
39:53and as the chip costs get undercut by
39:56new designs, the power costs are still
39:59there.
40:00And so, right now when people build
40:01these data centers, they think, "Oh,
40:03well, it's only 10% power cost. The
40:04other 90% all goes to Nvidia." But
40:07that's going to invert and the power
40:09costs will still be there.
40:11So, I think between here and 30 years
40:13from now Elon is exactly right. The
40:15currency that is most dominant in that
40:18next 30-year window is wattage and
40:21tonnage. Tonnage to build the data
40:23centers, wattage to power them. But I
40:26think if you if you can find a power
40:28supply
40:29today, it's the investment theme of the
40:32decade by far and and Chamath, I'd love
40:34to get your thoughts on like you got an
40:36early start on this. I don't know if you
40:37did it secretly or how it accumulated,
40:40but love to get your insights for the
40:41audience.
40:42>> The I think the the the way to think
40:44about it is when you look at load
40:45growth, just take the United States cuz
40:47I think that's a more tractable example.
40:49When you look at the
40:51load growth and you look at the sources
40:52of load growth, they're relatively
40:55pedestrian. It's more electric cars,
40:58more electric fridges, more electric
41:00televisions, PCs, laptops,
41:02>> [snorts]
41:02>> and data centers, but it's just broadly
41:05speaking that we are more of an
41:06electrified society. So, there's just
41:09like this
41:10>> [snorts]
41:10>> you know, this demand of of a need for
41:12electrons, basically.
41:14And then when you look on the supply
41:15side, there's huge issues even if you
41:18assume that every project that's planned
41:21actually gets not only permitted and
41:24greenlit, but also built on time and on
41:25budget.
41:27And when you net those two things out by
41:282050, we're still 1.4 terawatt hours
41:31short, which is effectively six
41:33California's worth of electricity.
41:35So, there is no world in which we catch
41:36up. And then people's reflexive reaction
41:39to that is but nuclear.
41:41And I'm like, well, I I just think like,
41:43you know, the problem with SMRs as an
41:45example is the S in the MR.
41:48>> [snorts]
41:48>> It's like, you know, what people don't
41:49realize is it's the same unit effort to
41:52get an SMR approved as an MR approved.
41:54Um or as an R approved as an LR
41:57approved, a large reactor.
41:59Uh because, you know, you're going
42:00through state local
42:03ordinances and people and conflicts and
42:05pushback and all of this stuff that does
42:07not change whether the form factor is it
42:09goes under your daughter's bed or it
42:11sits, you know, over there and it's it's
42:13gigantic.
42:15So, we're we're just massively short,
42:17and that's where my thesis started,
42:19which was just that observation. Then,
42:22you know, where I spent a little bit
42:23of well, not a little bit, a lot of my
42:25time
42:26was what is the most
42:28tractable all of the above solution set?
42:30Because, you know, once you realize that
42:32we're so short,
42:35you immediately abandon all of the
42:37rhetoric around climate change cuz
42:38you're like, just give give it give any
42:40give me anything.
42:41Oil, nat gas, coal, wind, solar,
42:45nuclear. And if you look at them all on
42:48just a total, you know, cost basis,
42:51what jumps off the page is um
42:54solar is slightly a little bit more
42:56expensive than wind,
42:58of but they're the cheapest. Um but it's
43:00just much faster. It's 2 years, it's do
43:02no harm, it's plunk it on the ground,
43:04and off off you go.
43:06So, I spent a just a bunch of time
43:08accumulating solar assets.
43:09And then that accidentally led me into
43:11the data centers because it was like,
43:13you know, through in Nvidia and Grok,
43:15what I realized was, my gosh,
43:17um the capacity build-out was so upside
43:20down because of all these constraints.
43:22Um because the data center is like this
43:23incredibly temporal moment where
43:26all energy and ener- energy meaning like
43:29people's energy, ire, anger, they all
43:31meet right there. And so, you know, you
43:33have you have the permitting issues, you
43:36have the pushback issues, you have the
43:37political issues, you have pushback
43:39against the oligarchy, you have
43:40capitalism, you have wealth inequality,
43:41all concentrated in that thing.
43:44Um and so, you know, about uh 2 years
43:46ago, I started to kind of parcel
43:48together
43:49um assets that I thought would be
43:51valuable in a world where
43:53that looks like it is it is the scarcest
43:54asset.
43:55>> Not a good That's how I under- That's
43:56how I underwrote it. That's That's
43:58impressive. That pays
43:59>> Yeah, it's not it's not it's not It
44:01wasn't cheap. It wasn't [laughter]
44:02cheap.
44:03But uh
44:04But and by the way, just to be clear,
44:06there's huge issues with that. Meaning,
44:08even when you have these assets,
44:10um when you actually try to get them
44:12turned on and get them built, my gosh,
44:14it's incredibly complicated. The funding
44:16cycle is very complicated. The supply
44:19chain is very complicated. The labor
44:21shortages on critical people because
44:22these are licensed, you know, skilled
44:24workers. You can't just, you know, grab
44:26some bumpkin off the street or go to
44:28like, you know, uh I don't know,
44:30Craigslist. Like I mean, you have to go
44:32and find like a licensed master
44:33electrician. They're very, very hard to
44:35find. They cost a lot of money. When you
44:37start to put all of this stuff together,
44:39um
44:40we are probably going to see a massive,
44:42massive, massive compute shortage in the
44:44next 3 or 4 years. Massive.
44:46>> Absolutely.
44:46>> Um massive.
44:48>> So, the fact that you said that cuz so
44:50many so many people explode
44:52Well, so many people come back and say,
44:54"What What are we going to do? What's my
44:55purpose? Where Where's my mission?"
44:57Everything you just outlined is
44:58mentioned for the
44:59>> 5 years.
45:00>> Yeah, exactly.
45:01>> All this All this whole Look, I do think
45:03that we have achieved AGI. It cannot
45:06diffuse. And the reason it cannot
45:07diffuse has nothing to do with the
45:09technology. It has everything to do with
45:11the build-up. And until you actually
45:13build it out so everybody can feel it or
45:15a plurality of human beings can feel it,
45:18it's just something that exists in a
45:19lab. Now, that has value. Don't get me
45:22wrong, but I would not underestimate
45:26um the difficulty. And if you look at,
45:27for example, the Anthropic S1 that just
45:29dropped today,
45:31it's shocking to me that the claim
45:34that they have to put as a risk factor,
45:35which I would never have thought would
45:37have happened for a company growing this
45:38fast and so powerful with all this
45:40money,
45:41the hyperscalers will not give them
45:44compute.
45:44>> Mhm.
45:45>> I've never seen that before.
45:47>> Wow.
45:47>> Yeah, it's wild. No, it's a really good
45:49indicator. You know, also Elon was
45:51saying there's 50 gigawatts of chips
45:53just waiting for liquid-cooled data
45:54centers to fall into. That's a massive
45:56asset depreciating idly.
45:59>> It's It's worse than that. There's
46:00There's that, then there's also the the
46:02not the not the non-Vera Rubin stock
46:05chips, the older generation that are
46:07also still sitting. There's a bunch of
46:09these guys that have these things
46:10sitting in inventory. Those aren't being
46:12used. Then even when you try to do
46:13liquid cooling, if you saw what happened
46:15with Nscale and Microsoft, you had two
46:18massive leaks of coolant two times in a
46:20row
46:21where Microsoft finally just triggered
46:23the liquidated damages clause and exited
46:24the deal.
46:26>> Yeah.
46:26>> So, everybody thinks this stuff is easy,
46:28and it turns out, well, again, you know,
46:31back to Elon as he's proven, physics is
46:33really hard,
46:34and building things and mechanically
46:37building things is really hard, and it's
46:38probably particularly hard because we've
46:40spent, you know, 20 years teaching the
46:42younger generation of kids who should
46:44have been learning how to do these
46:45trades just to, you know, be on TikTok
46:48and social media and video games.
46:50And so, you know, the the the the
46:52ability to actually build these things
46:54is gone.
46:55>> Yeah, and it's not just trades, too.
46:57There's regulatory components. If you're
46:58a good negotiator, you can get involved.
47:00There's planning. There's It It really
47:02involves every human skill. And it just
47:05drives me nuts that so many people are
47:06saying, "Well, look, math is cooked.
47:08Teaching is soon to be cooked. I guess
47:10I'll just sit around and watch, you
47:11know, late-night TV and maybe play some
47:13video games." I'm like, "No, no, no.
47:14There's a There's a bigger opportunity
47:16that just opened up than anything you've
47:17ever seen before. You just got to get
47:19involved."
47:19>> not that fatalistic. I just think people
47:21grossly misunderstand. They're wildly
47:23wildly wrong in the short term. They'll
47:25be
47:26right in the long term, but they're
47:27wildly wrong in the short term because
47:28of these constraints. I'll give you
47:29another example, just to give you an
47:30example.
47:31A lot of in In America, the the decrepit
47:34infrastructure we have is because has
47:36forced people to organize into some of
47:37these co-ops. What is a power co-op?
47:39It's basically a cross multitude of of
47:41states.
47:42And so, you know, this is the This
47:44undergirds our growth. Um
47:47And so, what happens in a co-op? Well,
47:48you think a co-op is great. Everybody
47:50just like kumbayas and agrees on stuff,
47:52but what really what happens is you
47:53have, you know, small players and big
47:55players in the co-op. And there there's
47:56an example of one that goes right
47:58through the spine of America. And um
48:00there's an incredible amount of
48:03uh data center capacity that one could
48:04build in the northern states of this one
48:07path. Um and it would be great because
48:09it would alleviate tremendous um
48:12constraints that we have in diffusing
48:14AI.
48:15And uh when you And so, you know, I went
48:17in and, you know, we my partner and I,
48:19we bought some stuff.
48:20And it turns out that they're part of a
48:22co-op. And, you know, when you dig in a
48:24few layers, to your point, it's all
48:26personal relationships, and flows
48:28downhill. And so, you know, the big boys
48:30who are actually in the South actually
48:31dictate how the co-op works. It's like
48:33an organized mafia. And so, you know,
48:36when you're up in the northern states,
48:37you can't do anything. You can't build,
48:39you can't get the right off-take
48:41agreements, and you wonder why, and it's
48:43because the guys down South are like,
48:44you know, Oklahoma Gas and Electric
48:46calls the shots.
48:48>> Mhm.
48:49>> So, when you look at why, you know, muse
48:51won't diffuse or bots, you know, bot
48:53won't dot won't won't diffuse or
48:55brockbot won't diffuse, that's why.
48:58>> That's right.
48:59>> Hey, man, all that complexity just
49:00smells like opportunity knocking. You
49:02know, just to put it in context, too. I
49:04think we're looking for about 100
49:05gigawatts total by 2030 to fuel all of
49:08AI. Uh 6 gigawatts is a big chunk a big
49:12chunk of that 2030 target. So, to get
49:15that all accumulated in 2 years is is
49:17just mind-blowing. But, I tell you like
49:20jump in, you know, anyone who gets
49:22involved in this uh seems to be
49:23thriving.
49:24And it it just is such an amalgamation
49:27of different skill sets.
49:29It's it gives you something to really
49:30really get done before Kardashev, before
49:33you know, you know, Elon is really big
49:35on all this moving to space. But, even
49:36the most wildly aggressive estimates put
49:38a couple percent in space within a
49:40decade.
49:41So, you don't you don't have to worry
49:42about getting spaced.
49:44>> The other thing that I think you have to
49:45think about is I think it makes sense
49:46for Google, Microsoft, Meta, SpaceX
49:50um to essentially deploy 100% of their
49:52compute capacity into space. But, if you
49:54take the Alex Karp AI data sovereignty
49:57thing to the limit.
49:59Let's say you're
50:01uh Humana. I'm just making up a company.
50:03You know, Humana doesn't have the
50:05capability
50:06um to deploy a data center into space.
50:08And even if they did, they'd be, you
50:11know, waiting in line for quite a long
50:12time.
50:13You know,
50:14the the prevailing logic was you kind of
50:17work in a, you know, in a cloud
50:20environment with a bunch of other
50:21tenants. The AI data sovereignty
50:23question would push Humana to basically
50:25say no, at a minimum we need to be in a
50:27EPC and at a maximum we need to be in an
50:28air gap facility that's our own.
50:30>> Mhm.
50:31>> So,
50:32there is going to be a lot of domestic
50:35terrestrial build-out as a result of
50:37these strategic decisions. Every single
50:39company that isn't one of the big boys
50:42will either have to have a multi-year
50:44lag and wait
50:45for capacity for launch, let alone
50:48capacity to deploy, and then capacity to
50:50to actually have it as a managed
50:52service.
50:53Uh and so, you're going to just build
50:55terrestrially while you wait. And if
50:57you're going to build terrestrially
50:58you again you go back to all these
51:00issues. And so,
51:01um
51:02it's a it's a huge problem for
51:04diffusion. I think that that's that's
51:05probably going to dictate the winners
51:07and losers more than anything else over
51:08the next 5 years.
51:09>> Jamal, for the average user who wants to
51:11get in on the energy play and realizes
51:13yes, energy is the innermost loop right
51:15now,
51:16where would you point them?
51:17>> Well, I think that there's a there's a
51:19bunch of different ways. One is that I
51:21think you have to realize
51:23that energy generation is
51:28the utility monopoly on generation is
51:30breaking. And so, you're going to become
51:33you're going to enter
51:34an extremely diversified
51:38deregulated, disaggregated generation
51:43concept of energy. And so, the most
51:45obvious way to do that is is through
51:46residential solar.
51:48Then there's going to be C&I solar.
51:51And the reason again is just like it's
51:52easy to permit, it's super fast, and
51:54you're often generating and you pair it
51:56with a battery, and now you're in the
51:58currency game. You're effectively your
51:59own little money printer. You use what
52:01you need, you sell back into the grid.
52:03So, that's
52:04simple way number one is you just adopt
52:06solar. Simple way number two is you
52:09attach distributed compute and you join
52:11these highly fragmented distributed
52:14pools, whether it's Venice or um
52:18you know, some of these other ones that
52:18their names are slipping me. I'm not
52:20involved in any of these, but I think
52:22they're quite quite interesting.
52:24Um then there are companies that, you
52:26know, give you the ability to buy these
52:28things yourself, like these things this
52:29company called Span which just did a
52:31deal with um
52:34Nvidia where it actually gives you like
52:36the storage, the battery, the GPU all in
52:39like this like unit that you can just
52:40deploy in your house. Um there's there's
52:43things like that. Um that would be the
52:46most simple way.
52:48>> Yeah.
52:48>> Um and I and I think probably the most
52:51obvious way. It's just I I guess maybe
52:52simply put
52:54you people should find the cheapest and
52:57most
52:58technically simple method that they can
53:00manage to get long electrons.
53:01>> Yeah, that's right. That's right. Let me
53:03add two things to that, too. You know, a
53:04lot of people say, "Well, look, you
53:05know, I'm I'm not involved in the
53:07political hierarchy. I don't know Josh
53:08Kushner and Jared Kushner." But, all of
53:11these things are regulated at the local,
53:12state, and federal levels. So, if you're
53:14connected at any level, you have an
53:16opportunity to get involved in these
53:18projects and add value and and just kill
53:20it. So, I think I think a lot of people
53:22underplay
53:23their ability to get involved. And also,
53:25if you're a senior executive of
53:26virtually any company,
53:28almost all of them got frozen out of
53:31compute because it literally got all
53:33scooped up by either quant trading firms
53:35or by Open AI or Anthropic. So, we had a
53:37sellout. Now, they're getting the big
53:39dope slap where, "Oh my god, we need
53:42hundreds of thousands or millions of AI
53:44agents
53:46to be competitive in insurance and in
53:48manufacturing and whatever. Where are we
53:50going to get the agents? The computer is
53:51all sold out." Now, you, especially in
53:53the insurance industry, you have the
53:55opportunity to be the first on your
53:56executive staff to go and say, "We need
53:59to unlock a billion dollars
54:01to do a prepay on Chamath's next power
54:05supply cuz that's where the computer's
54:06going to land. And if we get ahead of it
54:08this time, we'll we won't be frozen
54:10out." And so, that's another career
54:12opportunity where you don't have to know
54:14anything about AI. You don't have to
54:16know anything about how power supplies
54:17work. You just need to know where the
54:19capital is and unleash it for the
54:21buildout. So, there are lots of ways to
54:23get involved that that involve every
54:25skill set.
54:25>> Yeah, I would maybe also add so, I do
54:28buy the premise that at the capability
54:30frontier that supplies, including energy
54:33supplies, are a limiting factor. Not as
54:35confident I buy the premise at the
54:37demand side. Remember, Open AI, we talk
54:39about it constantly on this pod,
54:42infamously lurched toward and then away
54:45from the consumer. Open AI last year
54:47tried to make consumers want to spend
54:50money on reasoning tokens and reasoning
54:52models, and consumers just didn't want
54:55it. They didn't want to pay for it. They
54:56didn't have at the time a good use for
54:58it. So, I I do suspect to Chamath's
55:01point about diffusion of capabilities. I
55:04also think there's another important
55:06part of the equation, which is just
55:07making making these capabilities
55:10economically, rationally useful to the
55:13broader economy. And if they're not
55:15useful and if they can't be metabolized
55:18by the economy, no matter how much
55:20energy is showering down on the frontier
55:23labs or anyone else who would host the
55:24data centers, it's not going to make a
55:26lick of difference if the models
55:28ultimately can't generate enough
55:30economic value for all of their
55:32consumers. That's point one. Point two,
55:34I'll also point out second element in
55:36this room, which is distillation. The
55:38models are getting smaller. The models
55:40are getting less energy consumptive over
55:42time. So, even if we never built a new
55:45data center, never built a new power
55:47plant, SMR, MR, RR, LR, never built a
55:52new watt of capacity, arguendo, this is
55:55a thought experiment, wait a few years
55:57and just through the pure power of math
55:59and recursive self-improvement, I would
56:01expect that we will have models that
56:04take a fraction of the current energy
56:06footprint and generate many orders of
56:08magnitude better capabilities.
56:10>> Welcome to the singularity.
56:11>> Yes.
56:12>> Yeah. Uh so, you want to close us out?
56:15>> Yeah, go for it.
56:15>> I got nothing.
56:16>> All right.
56:17>> Great, I'll take it.
56:18>> I got nothing. [laughter]
56:19>> Wait, that one is every time.
56:20>> Need more energy.
56:21>> Go ahead, Dave.
56:22>> No, what Alex said is just absolutely
56:23fascinating cuz there's a footrace right
56:25now between self-improvement, models
56:27getting lighter, faster, smaller, and
56:29then more ideas for how to use that
56:31intelligence exploding in scale at the
56:34same time. I'm firmly in the camp that
56:36we will not run out of ideas on how to
56:38use every bit
56:39for decades.
56:41Uh
56:41we just have to unleash our
56:42imaginations. But, you know, right now,
56:45um Will Thompson, that brilliant Harvard
56:47junior that I was talking about a couple
56:48podcasts ago, he got access to that
56:511,200 tokens per second Cerebrus
56:53capability. You can now get it on Astra.
56:55You have to go through Codex to get it
56:57cuz, you know, Elon and and Sam are
56:59fighting. But, if you go through Codex
57:01to get access to Astra, you can put it
57:03on like crazy fast hyper mode or
57:06whatever it's called. It's insanely
57:07expensive.
57:08So, try it for like 10 minutes,
57:10But, you will never go back. It's the
57:12total red pill. You will never You will
57:14want it for the rest of your life. It's
57:16a
57:17>> It's your first It's your first dose of
57:18heroin.
57:19>> It is your first dose. So, we have a
57:21long way to go before people run out of
57:23ideas on how to use
57:24how to use more and more and more AI.
57:26So, but it is a really interesting
57:27footrace.
57:28>> hyperinflation universe
57:29>> This episode is brought to you by Blitz
57:31C, autonomous software development with
57:33infinite code context. Blitz C uses
57:37thousands of specialized AI agents that
57:39think for hours to understand
57:42enterprise-scale code bases with
57:44millions of lines of code.
57:46Engineers start every development sprint
57:48with the Blitz C platform, bringing in
57:50their development requirements.
57:52The Blitz C platform provides a plan,
57:55then generates and pre-compiles code for
57:57each task. Blitz C delivers 80% or more
58:00of the development work autonomously,
58:03while providing a guide for the final
58:0420% of human development work required
58:07to complete the sprint. Enterprises are
58:10achieving a 5x engineering velocity
58:13increase when [music] incorporating
58:14Blitz C as their pre-IDE development
58:16tool, pairing it with their coding
58:18co-pilot of choice to bring an AI native
58:21SDLC into their org. Ready to 5x your
58:25engineering velocity? Visit blitzc.com
58:27to schedule a demo and start building
58:29with Blitz C today.
58:30>> [music]
58:33>> All right, our next clip from the
58:34Isaacson conversations. In this one,
58:37Elon says humans are unlikely to remain
58:39in control of digital intelligence when
58:41it becomes a million times greater than
58:43biological intelligence. Maybe except
58:45for for Alex. Um and it's hard to
58:47imagine human beings in control of that
58:50scenario. So, I hope the AI is nice to
58:52us. All right, let's watch this clip and
58:55let's discuss it.
58:56>> I find it unlikely that humans would
58:58remain in control
59:00if the if
59:01I mean
59:04if there's a million times more
59:06digital intelligence than there is
59:08biological intelligence. It's It's hard
59:10to imagine humans being in control in
59:11that scenario.
59:13So, I hope the AI is nice to us.
59:16>> Anything we can do to encourage that?
59:20>> I think programming truth-seeking and
59:21curiosity
59:23um
59:26is is probably good. I think I I guess
59:28we we want to be careful about being you
59:30know, creating too many AI dystopian
59:33scenarios
59:34online because the AIs are trained on
59:37everything online. [clears throat]
59:38>> No, if we say it's going to be useful
59:40and not dystopian, they'll do it.
59:42>> Terminator, you know, because I
59:44whatever they read you know, 80 versions
59:46of the Terminator script.
59:48>> [laughter]
59:48>> So, it's it's like I guess that's what I
59:50guess that's me,
59:51Terminator.
59:53Maybe we we should have a some more uh
59:55let's let's have some more your
59:56utopian narratives for AI.
59:58>> Yes. Yes, exactly.
1:00:00>> You're hearing
1:00:01>> Chamath, I don't know if you know this
1:00:02um about 6 months ago I ran an X Prize
1:00:05called the Future Vision X Prize. It was
1:00:07a global film competition to create uh
1:00:11film trailers and scripts for positive
1:00:13visions of the future. And uh we had
1:00:15incredible over 5,000 entries and we're
1:00:19making two of the winners right now. But
1:00:21we need that. All the All the films out
1:00:23there are dystopian. As Elon said on our
1:00:25on our podcast interview, Dave, um you
1:00:27said we need we need more Roddenberry,
1:00:29less less Cameron.
1:00:31But I think that's true. We're we're
1:00:32training AI with these dystopian
1:00:34narratives and on Reddit and Facebook
1:00:37posts. Uh yeah, be careful what
1:00:39intelligence, you know, gets fed on.
1:00:43>> Uh well, we we we used every word ever
1:00:45written in the training set cuz we're
1:00:47lazy.
1:00:48But we don't need to do that, you know,
1:00:50you can completely control the training
1:00:51set. And I think we're starting to get
1:00:53around to that now actually in some of
1:00:54the uh some of the faster training runs
1:00:56you're cutting down the data.
1:00:59But uh Elon's got a really important
1:01:00point he's been making for a long time.
1:01:01Whoever controls which data get it gets
1:01:03cut down to controls the personality and
1:01:05the decision-making of the AI.
1:01:07Which is a lot more power than being
1:01:09president of the United States.
1:01:11Yet it's just sitting there in the in
1:01:13the labs like at a in sort of you know
1:01:15unregulated hands. It's really weird.
1:01:17>> the the bigger issue by far is human
1:01:19beings using AI against other human
1:01:21beings. We are very fractured global
1:01:23society. That's much more concerning
1:01:26than AI maybe becoming
1:01:29power hungry. And that's a that's a
1:01:32stretch of an assumption.
1:01:34>> In the short term it it definitely is.
1:01:36But Chamath, I'm I'm curious, has there
1:01:38been any conversation that you've heard
1:01:40of where the labs are basically being
1:01:43asked, "Please generate training data
1:01:46sets that are aligned with humanity."
1:01:49That teach the AI through all of the
1:01:51story, you know, that AIs support
1:01:54humanity. And it's trained on synthetic
1:01:57data that aligns us. Anything going on
1:01:59there?
1:02:01>> I think I think that that that is
1:02:02happening. I think the question is how
1:02:05how is it weighed in totality? I'm not
1:02:07sure how much it's weighed because
1:02:10depending on the lab, the founding
1:02:11documents that they use when they seed
1:02:13their models have very different goals.
1:02:16Um and so, you know, I I I think that
1:02:18when when
1:02:19Elon says that Grok is maximally
1:02:21truth-seeking, that's very different
1:02:23from another AI lab that says, you know,
1:02:25the goal is to build something that can
1:02:27alleviate all suffering. And you know,
1:02:29you're going to be this omnipresent
1:02:30judge of all of that.
1:02:33You're going to get two very different
1:02:34outcomes. And so, at that point the
1:02:36training data kind of matters, but you
1:02:38know, these
1:02:39the these AIs are smart to ignore the
1:02:41stuff that they don't want to hear at
1:02:43some point through enough far side
1:02:45because the founding document is a thing
1:02:46that's canonical. So, I think maybe
1:02:48what's maybe what we should be asking is
1:02:51those things should be inspectable
1:02:55um in a much more open-source way. Um
1:02:57and and I think like if governments
1:02:59actually wanted to get involved, that's
1:03:00probably what they should be
1:03:02having an opinion on.
1:03:03>> Yes. Hallelujah. I'm so glad you said
1:03:05that because I I think a lot of the
1:03:07debate on training is academic now that
1:03:10the Chinese have open-sourced
1:03:11self-improvement capable models.
1:03:14And inspection is everything. Like if If
1:03:15going to survive going forward, it's
1:03:17going to be via inspection, and the more
1:03:19voices that are behind that, the better.
1:03:21Cuz even even a Chinese model that's
1:03:22trained to do something evil by you can
1:03:24still inspect it. And that's the only
1:03:27way, you know, global inspection with
1:03:29some kind of authority is the only way
1:03:30we're going to contain it. Cuz cuz prior
1:03:33to this year, there was an argument that
1:03:35you could post-train it to be aligned,
1:03:36and that would be fine.
1:03:37But the new Chinese models are
1:03:39absolutely capable of self-improvement.
1:03:41There's no doubt in my mind. So, that
1:03:42cat is out of the bag. It's the boat has
1:03:45sailed. Forget it. As a practical
1:03:47matter, it's about inspection and
1:03:49containment that way going forward, not
1:03:51about training for alignment.
1:03:54>> Alex, I'm super curious about your point
1:03:55of view here.
1:03:56>> I would say let's not kid ourselves that
1:03:59either feeding frontier models just the
1:04:02right information diet of pre-training
1:04:04data or just the right regulatory regime
1:04:07of some
1:04:09government agency overseeing system
1:04:11prompts or as Anthropic seems to like to
1:04:13call them soul documents is going to
1:04:15yield a positive outcome. I I think
1:04:17that's an incredibly brittle approach. I
1:04:19think it's far more likely again, look
1:04:21back to the early creation of OpenAI.
1:04:24One of the reasons I remember chatting
1:04:26with Elon around the time of the
1:04:28founding of OpenAI, the notion was he
1:04:30was concerned that DeepMind would be a
1:04:32singleton that would control the future
1:04:34light cone of humanity. And that the
1:04:36only way to ensure that humanity didn't
1:04:39get lost to some sort of totalitarian
1:04:41superintelligence singleton was to
1:04:43create a counterbalance and competition.
1:04:45So, OpenAI was formed Fairchild style,
1:04:48and there were Fairchilden, and then
1:04:50Anthropic spun out unintentionally of
1:04:52OpenAI, and we got more Fairchilden. I I
1:04:55think there are probably at least in my
1:04:57mind only two robust ways to ensure
1:05:00positive outcomes, what others might
1:05:03call AI alignment or AI safety. I think
1:05:06one of them is competition and defensive
1:05:08co-scaling. We have lots of AIs, lots of
1:05:11SIs that represent a diverse range of
1:05:14viewpoints, just like we have lots of
1:05:15humans that represent a diverse range of
1:05:18viewpoints with diverse values and
1:05:20ideals, in many cases competing with
1:05:22each other to see who can be friendlier.
1:05:25I think that's one way to do it. And
1:05:27then I think the second way, and I was
1:05:28kind of surprised Elon didn't
1:05:30immediately flag this given his
1:05:32Neuralink investment and stake, is
1:05:34humans merging with the machines. One of
1:05:36the easiest ways to, I think, inject the
1:05:39distribution of superintelligent goals
1:05:42and values with human values, human meat
1:05:45body values, is through uploading and
1:05:47side loading and BCIs and humans
1:05:51generally merging with the machines. I
1:05:53think a combination of lots of
1:05:54competition and defensive co-scaling on
1:05:56the one hand and lots of human machine
1:05:58mergers on the other. That seems to me a
1:06:00far less brittle approach rather than
1:06:02talking about controlling AIs. We're not
1:06:04going to control the AIs or the SIs.
1:06:06We're going to merge with them. That's
1:06:07the solution.
1:06:09>> Mhm. Uh Chamath, I'm curious. One of the
1:06:10conversations we have on the pod a lot
1:06:12is whether as we reach superintelligence
1:06:15or exceed superintelligence, whatever
1:06:16that is,
1:06:18uh that does it become more wise? Do we
1:06:22get uh benevolence out of these
1:06:25intelligence at the extreme? And just
1:06:28worry about the, you know,
1:06:30the intermediate period uh where, you
1:06:33know, humans and AIs are are basically
1:06:36in in conflict or in uh in turmoil with
1:06:39each other.
1:06:39>> It's not guaranteed, I don't think. I
1:06:41think that um it depends on who's who's
1:06:43in control of it. Um
1:06:46and how it's aligned and how it's been
1:06:48trained and what it perceives its goals
1:06:50to be.
1:06:51I mean, I think just like there's like a
1:06:52distribution of outcomes in human
1:06:54genetics, I think it's fair to say that
1:06:55in in silicon there's going to be a
1:06:57distribution of outcomes, too. And
1:06:59so there's going to be some silicon
1:07:00distributions that we're not going to
1:07:01like very much. And it's going to be up
1:07:04to other silicon distributions to uh
1:07:06combat them, just like we do we do in
1:07:08humanity. I I think it's probably a lot
1:07:10of the same. So I don't think it's I
1:07:12don't think benevolence is by any
1:07:13stretch of the imagination guaranteed.
1:07:15>> Mhm.
1:07:17>> All right. Uh let's move on to our our
1:07:19next clip. Uh so uh this is Elon
1:07:23reflecting on his time at Doge and
1:07:26reflecting on solving the national debt.
1:07:28So, Elon says, "There's no way we're
1:07:30going to solve deficit from within the
1:07:32government. The only way we're going to
1:07:34solve the national debt is with AI and
1:07:36robots." All right, let's watch this
1:07:38third clip here.
1:07:39>> I came to the conclusion that
1:07:41um
1:07:42that that there was no way we're going
1:07:44to solve the deficit
1:07:45from within the government, impossible.
1:07:47Um
1:07:49you know, maybe if the president made
1:07:50that his primary goal, we we could do
1:07:53that. But but it
1:07:55The problem is if you if you try to
1:07:56solve the deficit and and and stop
1:07:57wasteful spending, um
1:07:59you get a you make a lot of enemies in
1:08:01that in that situation. And if President
1:08:03Trump is not, you know,
1:08:05he he he likes to be liked. He's he's
1:08:06not
1:08:07he doesn't like people,
1:08:09you know,
1:08:10calling him, being upset about what some
1:08:12funding being stopped type of thing. But
1:08:14you know, I can't I came to the
1:08:15conclusion that
1:08:16the only way we're going to solve the
1:08:18national debt is with AI and robots.
1:08:21We're not going to solve it by trying to
1:08:22save our way there. The appetite is not
1:08:24there and it's not just at the
1:08:25presidency. The appetite is not there at
1:08:27the presidency, the House, the Senate,
1:08:29or the the judiciary.
1:08:31Basically, no one in authority in the
1:08:34government whatsoever really wants to
1:08:36stop
1:08:37reduce spending.
1:08:38>> So, let's put some numbers on this the
1:08:41scale of the problem. So, as of October
1:08:435th, the national debt was about $40
1:08:45trillion. It It grew 2.44 trillion in
1:08:48the past year alone. About 7.5 billion a
1:08:51day. We're paying 3 billion a day just
1:08:54in interest. So, how could AI and robots
1:08:56fix this? You know, I count four ways.
1:08:59One,
1:09:00uh, growth. You know, the debt problem
1:09:02is really a debt to GDP ratio. So, grow
1:09:04the GDP fast enough. Again, GDP's a a
1:09:07lousy measure and the debt shrinks
1:09:10relative to the economy. Two, tax
1:09:12revenues. A bigger economy means bigger
1:09:14tax revenues without raising taxes. The
1:09:17third is a cheaper government. So,
1:09:19healthcare is the biggest driver of
1:09:21federal spending. Drive longevity
1:09:23breakthroughs and prioritize AI doctors,
1:09:25AI discovered drugs, robotic care, and
1:09:28people stay healthier longer,
1:09:30contributing to society, cutting costs.
1:09:32And fourth, the deflation of goods and
1:09:34services if robots make things
1:09:35dramatically cheaper, uh what the
1:09:38government buys gets cheaper, too.
1:09:40Uh in quickly in a related story, uh
1:09:43Jack Clark,
1:09:44uh Anthropic's co-founder was at IMF
1:09:46this week in Bangkok, and he said that
1:09:49Anthropic's internal economic models
1:09:51project uh US GDP could reach 44
1:09:54trillion by 2030 compared to 30 trillion
1:09:57today, nearly 50% growth, about 9 to 10%
1:10:01per year. Uh in context, you know, the
1:10:03US has averaged 2 to 3% per year for
1:10:05decades. So, um Chamath, over to you. Uh
1:10:09do you believe these numbers? Do you
1:10:11believe Elon's views that AI robotics,
1:10:14and I'll add longevity medicine, could
1:10:16help get us out of this 40 trillion
1:10:18dollars of debt?
1:10:19>> It doesn't get us out of debt as much as
1:10:21it just makes the debt not that
1:10:23important. Um
1:10:25but I do buy that. Yeah, I think that,
1:10:26you know, if uh let's just take like the
1:10:29this next generation of GLP-3s plus some
1:10:31cancer vaccines, um you know, can you
1:10:34shrink
1:10:36healthcare as a percentage of GDP to
1:10:38under 10%?
1:10:40I you probably could.
1:10:41>> Yeah.
1:10:42>> Um forget the waste, fraud, and abuse. I
1:10:44think he's bringing up an important
1:10:46point, which is
1:10:47the system is
1:10:49riddled with waste.
1:10:52And I think that everybody got too
1:10:53caught up in the fraud.
1:10:55What do I mean by that? Like, I think
1:10:57it's sort of like the situation where
1:10:58everybody was looking for somebody's
1:11:00hand in the cookie jar. Like, "I got
1:11:01you."
1:11:02And it turned out it was it was just
1:11:03like a little bit more leaky than that,
1:11:05you know, where it's like uh a hospital
1:11:08may be upcodes for a couple of extra
1:11:10things. You come in for your knee, you
1:11:12know, it tags you for the sore throat,
1:11:14and then, you know, it doesn't
1:11:16necessarily get repudiated or pushed
1:11:18back on, and so then it gets reimbursed,
1:11:20and blah blah blah. So, it's I think
1:11:22that there's a lot of probably waste in
1:11:24the system.
1:11:25And there there is No, but whose whose
1:11:27incentive is that? again, just to pick
1:11:29on health care, is it the insurers?
1:11:30Well, you know, they necessarily would
1:11:32recognize less revenue as well
1:11:35on top of it all. So,
1:11:37I [snorts] don't know. I think like
1:11:38you're probably
1:11:41if you're going to if you're if we're
1:11:42going to see these um
1:11:45if we're going to see this problem
1:11:46solved, I think it's probably going to
1:11:47start by
1:11:48by things that like are just
1:11:50discontinuous improvements so that the
1:11:52systems aren't uh used. And then that
1:11:54way the fraud and the waste and the
1:11:56abuse can't happen. So, again, uh uh
1:11:59Reda plus some cancer vaccines plus all
1:12:01that stuff, I may I I think like like I
1:12:03said, that probably clips
1:12:06the cost of health care in half.
1:12:08Um
1:12:09and I think that there's probably some
1:12:10stuff on the productivity side. Robots
1:12:13are definitely one.
1:12:14Um
1:12:15and I could I could see where like, you
1:12:17know, maybe our debt kind of slows down,
1:12:21but then GDP is much higher. I could see
1:12:25that. Um yeah.
1:12:26>> Alex, I don't know if you saw the the
1:12:28news that
1:12:30the GLP-1 drugs are slowing aging by 9%,
1:12:33which is pretty extraordinary. We've
1:12:35talked about this in the first episode.
1:12:36>> know if I saw the news. I'm flagging
1:12:38this for you every 5 seconds. Yes, of
1:12:40course I saw the news. I'm I'm the one
1:12:42banging banging the drum that we may
1:12:44already be in spiky LEV. What are you
1:12:46talking about? Of course I saw the news.
1:12:47>> All right, pal.
1:12:49>> Yeah, I I I I
1:12:50>> Ooh, stepped on a landmine there.
1:12:52>> I I I I I do think just regarding the
1:12:55the debt increase, you have to look at
1:12:57where the debt is growing. Debt is
1:12:59growing from entitlements, it's growing
1:13:02from defense costs, and I I think there
1:13:05are sci-fi scenarios where
1:13:08maybe it's a bit of slight of hand
1:13:09almost with AI where existing increases
1:13:14are sort of translated over from the way
1:13:16they would be
1:13:17be conventionally spent over to robotics
1:13:20and AI. And I I'm reminded again, we
1:13:23went from say a world in the 1980s where
1:13:26if you wanted to have knowledge at your
1:13:28fingertips, you had to do something
1:13:30silly like buy the Encyclopedia
1:13:31Britannica and then fast forward and
1:13:34Wikipedia is free and available for
1:13:37everyone. Similarly with defense
1:13:39spending and Medicare and entitlement
1:13:42spending, I do think there it may
1:13:44require a forcing function or a crisis,
1:13:47but I do think there's a way through
1:13:49ephemeralization and superintelligence.
1:13:52There can almost be a slight of hand
1:13:54where, okay, one day you defense
1:13:57spending is $7 trillion
1:14:00that the next or $8 trillion or
1:14:03similarly with entitlements and the next
1:14:05year through a combination of robotics,
1:14:08look, don't look the other way, look
1:14:10this way, suddenly it goes down to near
1:14:13zero in terms of new spending through
1:14:15automation. I I think something like
1:14:17that is possible where if we look back
1:14:20at it, it looks almost like a rebasing
1:14:23of the currency, not a debasing, but a
1:14:25rebasing, but I think it'll require a
1:14:27few more years worth of technological
1:14:29advances before we're able to do that
1:14:31and I I'll predict not financial advice,
1:14:35but I I think it will look
1:14:36with the benefit of hindsight like
1:14:38slight of hand that erases ultimately a
1:14:41good chunk of the national debt
1:14:42increase.
1:14:43>> You know, Dave, when we interviewed Elon
1:14:45back last December,
1:14:48he said he's going to see double-digit
1:14:50growth in the next 12 to 18 months and
1:14:52then triple-digit growth in 5 years.
1:14:54Then in March at the Abundance Summit
1:14:56when he spoke, um you know, he was like
1:14:59we're going to see 10 times the economy
1:15:01in 10 years. Uh he's he's brought that
1:15:05back a little bit and you know,
1:15:08but still, you know, super super you
1:15:11know,
1:15:13uh bullish on economic growth.
1:15:15Your thoughts, pal?
1:15:16>> Well, the timeline, I think, you know,
1:15:18the the story within the story is the
1:15:21timeline. When he got involved in
1:15:22politics,
1:15:24it would have been, you know, before
1:15:252023 basically, before the last
1:15:26election.
1:15:28Uh at that point in time we're in the
1:15:29kind of the GPT-4 era
1:15:31and the consensus around MIT and
1:15:33Stanford was uh where, you know, maybe
1:15:3510 to 15 years away from ASI. And that
1:15:38Dennis Assabis was saying, you know, we
1:15:39need two or three really big fundamental
1:15:42new ideas and we'll be over the finish
1:15:44line.
1:15:45Then as the election unfolded and then
1:15:48the year after that when he was doing
1:15:49Doge, we stepped up to clearly things
1:15:53are happening much faster
1:15:55than anyone expected. And you know, Elon
1:15:57is the ultimate entrepreneur. He like
1:15:59the whole game plan shifts from the
1:16:01impossible chore of trying to fix the
1:16:03government before we bankrupt ourselves
1:16:05to well, if AGI is now eight or 10 years
1:16:08sooner than we were predicting two years
1:16:10ago, that's actually the escape valve.
1:16:13So he's a 100% right, but you know, you
1:16:15have to think that that changed in the
1:16:16middle of that like if he could rewind
1:16:18his life, would he get involved in Doge
1:16:20again?
1:16:21Almost certainly not.
1:16:23But we didn't know that things were
1:16:25going to unfold this quickly. You know,
1:16:26the recursive self-improvement effect is
1:16:28so fast now.
1:16:30So yeah, he did a bunch of things other
1:16:31than just this. You know, he
1:16:33massively doubled down on xAI. He bought
1:16:36every freaking Vera Rubin chip that can
1:16:38possibly come off the assembly line, no
1:16:40matter what it costs. You know, saying
1:16:42all the nicest things you could ever
1:16:43possibly say about Jensen just to try
1:16:44and get some more allocate like these
1:16:46are all new thoughts that come because
1:16:49of that short pass to ASI. So I I I
1:16:52think that's kind of lost in the story.
1:16:54It seems like he changed his mind. No,
1:16:55the world changed very very quickly and
1:16:58he knows how to remap his life plan to
1:17:00fit in the new reality.
1:17:02>> Yeah. Salim, what's your thoughts here?
1:17:04>> Um two things. I think
1:17:07just want to echo two things. One is
1:17:09Elon is making a denominator argument,
1:17:11right? If you have ways of deal if you
1:17:14can make the underlying economy much
1:17:16much bigger and AI and robots attack
1:17:18that denominator by by driving massive
1:17:21productivity, that helps. And I think
1:17:23Chamath pointed Alex's point about we
1:17:25can use technology to demonetize
1:17:29existing goods and services, that
1:17:31combination will go a long way towards
1:17:33solving the the debt issue, which is the
1:17:36biggest god damn issue facing the world
1:17:38today, not just the country, it leaks
1:17:40across the whole globe as we can see in
1:17:42France.
1:17:43>> Well, I I think the challenge is and
1:17:45Chamath, I'm curious about your views.
1:17:47What I'm concerned, you know, when we
1:17:48interviewed Elon last December, he said,
1:17:51you know, UHI and civil unrest are going
1:17:54to be deployed over the next few years.
1:17:55And I think if we hit civil unrest, the
1:17:57only option the government has is to
1:18:00print COVID checks. You know, I think
1:18:02the governments could be handing out,
1:18:04you know, two or three thousand dollars
1:18:05per month to every American just to keep
1:18:08them, you know, from from revolting.
1:18:10I don't know if you if you agree with
1:18:12that or not or what other, you know,
1:18:14avenue there might be.
1:18:16>> I mean, I think that there's a case to
1:18:18be made for giving everybody Reta
1:18:20Rutide.
1:18:21>> [clears throat]
1:18:21>> Yeah.
1:18:22>> And I wonder I wonder how positive the
1:18:24impacts on society would be if we just
1:18:25did that. You know, uh so I think like a
1:18:28lot of these late cycle phenomena that
1:18:30people point at could be nipped in the
1:18:32bud with a good dose of Reta.
1:18:34>> Maybe [laughter] maybe throw in Chamath,
1:18:36maybe throw in a little bit of oxytocin
1:18:37just to to keep everyone feeling
1:18:39friendly.
1:18:40>> Well, you could, you know, I mean, as
1:18:41you know, like, you know, you can just
1:18:43get everybody to hug more and you'd get
1:18:45the same oxytocin release. So, maybe
1:18:46forced hugging and Reta [laughter]
1:18:48Rutide. But the Reta Rutide, if you
1:18:50think about it, you know, social media
1:18:52consumption will go down, alcohol
1:18:53consumption will go down, cannabis
1:18:55consumption will go down, pornography
1:18:56consumption will go down. Um general uh
1:18:59um
1:19:00>> behavioral addictions would basically go
1:19:01down. Um your cardiac health would go
1:19:04up. Um
1:19:05>> Cognitive health goes up. Yeah.
1:19:07>> Cognitive health goes up.
1:19:08>> GDP would drop, though.
1:19:10>> And it would be a brave new world with
1:19:12Reta Soma for everyone.
1:19:14>> Reta Soma, yeah.
1:19:16>> Yeah, that oxytocin is, uh you know, all
1:19:18kidding aside, the digital or the AI
1:19:20friend version of that is imminent. I
1:19:22don't know how that's going to play out,
1:19:23but but
1:19:25AI can make you very, very happy if if
1:19:28it's trained right and tuned right and
1:19:30it knows so much about you and it's it's
1:19:31a good coach, it's a good it's a digital
1:19:34hug, you know, it has that same effect
1:19:35of of just uh making you feel like I
1:19:38have meaning and I'm loved and I and I
1:19:40have a plan.
1:19:41>> I mean, I don't know about you guys, but
1:19:42my experience with these models is that
1:19:44they're inherently sycophantic except
1:19:45for Grok. And I think that's a design
1:19:47feature and I appreciate the design
1:19:49feature actually as like a very basic
1:19:51kind of user of AI in most of my
1:19:54kind of like non-working life,
1:19:56uh I like the sycophancy. It makes me
1:19:58feel good.
1:19:59>> Yeah. Well, the emotional IQ is going up
1:20:01just like the regular IQ at a shocking
1:20:03rate, too, I think. You know, the
1:20:04original training data, it's all text.
1:20:06There's no images. It's not
1:20:07particularly, you know, emotive. The new
1:20:09training data involves every video ever
1:20:12taken and it's much, much more in tune.
1:20:14Like the the sycophancy feel is when it
1:20:17crosses a line and it feels like, "Ooh,
1:20:18now it's icky." It doesn't really do
1:20:20that as much anymore. It's it's it's
1:20:22super friendly and super kind without
1:20:23crossing the line of being just an ass
1:20:25kiss.
1:20:26It it and it's only going to get better
1:20:28from here. I mean, very quickly.
1:20:29>> definitely masked better. It's not
1:20:31better, but it's definitely [laughter]
1:20:32masked better.
1:20:33>> Well, like Sam says, like if you can't
1:20:34tell the difference, does it really
1:20:36matter?
1:20:37>> You you know what, Chamath, you're
1:20:38absolutely right.
1:20:42Go on.
1:20:42>> Thank you. Very, very funny. [laughter]
1:20:44>> No, that's it.
1:20:45>> That's it.
1:20:46>> Zach crossed the line.
1:20:47>> Be nicer.
1:20:47>> That's it.
1:20:48>> Hey, just just one other note,
1:20:50everything in this story
1:20:52everything in this story is a US
1:20:54perspective, the US debt, the US way out
1:20:56of the debt.
1:20:57>> Yeah, good point.
1:20:58>> becoming clear that if you're a European
1:20:59list a European listening to this, I
1:21:01mean, what is your your debt is worse,
1:21:05but you don't have the AI lever coming
1:21:08up.
1:21:09And the US is clearly like if if
1:21:11Anthropic's wet labs start discovering
1:21:12new medications at an alarming rate,
1:21:14which seems likely,
1:21:16we're going to export those out of the
1:21:17US.
1:21:18People all over the world are going to
1:21:20want them. That money's going to come
1:21:21pouring into the US.
1:21:23And I would be worried sick if I'm in
1:21:25continental Europe and we have no AI
1:21:28models and we are not on this curve.
1:21:30Like what what are they talking about in
1:21:32terms of their debt and how they're
1:21:33going to get out of it? And it it can't
1:21:34be pretty. I'm not in those
1:21:35conversations, but it must feel really
1:21:38bad. broad
1:21:39>> the broad generalized conversation
1:21:42outside of Europe that I would make is
1:21:43that every country essentially becomes a
1:21:45vassal state if they don't have a
1:21:47resource that's worthwhile to trade with
1:21:49the two countries with ASI. So, that's
1:21:52us and China. And so, you'll have a you
1:21:54know, a handful of remora fish, you
1:21:56know, around each of the big sharks.
1:21:59And then everybody else is in a very
1:22:00difficult spot. So, you know, if you're
1:22:02Italy, which I have
1:22:04tremendous emotional allegiance to
1:22:05because my wife is Italian.
1:22:07You have critical resources that are
1:22:09actually quite valuable that
1:22:11I'll just be honest, Liechtenstein and
1:22:13Belgium do not. And so, then the
1:22:15question of like, why are you aligned to
1:22:16Liechtenstein and Belgium Belgium? It it
1:22:19the the
1:22:21I I get why you are today. It's a
1:22:22historical artifact of a bunch of
1:22:23decisions that made sense back in the
1:22:25moment. This is the complicated
1:22:27calculus, I think, for Europe over time
1:22:28is where are the real allegiances and
1:22:31and
1:22:34who do you offer assets to
1:22:36that actually give you all the things
1:22:37that you thought you were going to get
1:22:38when you were aligned as a geographical
1:22:41consortium?
1:22:42>> You can only get I I think so far with a
1:22:44degrowth policy, Tremaine. I I recall
1:22:48Portofino being one of your favorite
1:22:50spots in Italy. I'm I'm sure you tracked
1:22:52the news that the mayor of Portofino was
1:22:54sending drones around with thermal
1:22:56imaging looking for anyone with the
1:22:58gumption to put air conditioners in in
1:23:00their homes because again, under the
1:23:03Yeah. Yeah, this is quite a scandal in
1:23:05Portofino.
1:23:07Like
1:23:07Italy large portions of Italy last I
1:23:10tracked were dependent on France and
1:23:12French nuclear for importing their
1:23:14electricity. And I I think you can only
1:23:16get so far by sending drones around with
1:23:18thermal imaging cams to to punish
1:23:20citizens for air conditioners before
1:23:22eventually you just give up and you
1:23:24become a living history museum.
1:23:27>> Yeah.
1:23:28>> Well,
1:23:28>> Well, I think the move the move that
1:23:29matters is all about power and data
1:23:31centers. And if if you don't jump on
1:23:33this next window of opportunity, then
1:23:35once it hits Dyson swarm, you're toast.
1:23:37Like you you're just irrelevant forever
1:23:39thereafter.
1:23:40Um you know, except as a tourist
1:23:41destination or whatever. But but I think
1:23:44there is a window of opportunity now um
1:23:46especially because all of the data
1:23:47centers in the Saudi and and Middle East
1:23:50got bombed.
1:23:52And so, there's not a lot of new chips
1:23:54going into that area right now. It's
1:23:56just too turbulent. So, that opens the
1:23:58door to rethink
1:23:59a bunch of other regions around the
1:24:01world where like, "Hey, can we get on it
1:24:02like right now?" And try and be a player
1:24:05in this and it all it all comes down to
1:24:06that decision to unleash power and data
1:24:08center opportunities in your home
1:24:10country before the Dyson swarm.
1:24:12>> You know, some of this is probably
1:24:14priced into the market, but some of it
1:24:15probably isn't. I do think there is an
1:24:17opportunity for those who want to build
1:24:20anti-aircraft guns into data centers.
1:24:22We've already seen the consolidation of
1:24:24energy, co-located energy as Chamath
1:24:27perhaps would attest with compute. What
1:24:30we haven't seen as perhaps fulsomely
1:24:32developed is defense consolidated with
1:24:35energy consolidated with compute. So,
1:24:37what we see right now with the data
1:24:39centers in certain parts of the Middle
1:24:40East, I I think is perhaps a harbinger
1:24:42for what could happen elsewhere in the
1:24:44world where the data centers want to be
1:24:46able to defend themselves in addition to
1:24:48powering themselves and it what what
1:24:50happens right now perhaps in Middle East
1:24:53won't stay there. We see it perhaps in
1:24:56the Ukrainian Russian theater. We might
1:24:58end up seeing it elsewhere in East Asia.
1:25:02Maybe there's just like this
1:25:03consolidated bit of computronium that's
1:25:06self-powering, self-defending,
1:25:08self-computing. That may be an
1:25:10opportunity.
1:25:11>> All right, I'm going to move us on to
1:25:12our next story from a close friend and a
1:25:14friend of the pod, Bill Gross. Bill
1:25:16founded Idealab in 1996, the original
1:25:18startup incubator and more than 175
1:25:21companies came out of it. Yesterday, he
1:25:23launched idealab.ai,
1:25:25a free platform with more than 1.2
1:25:27billion business ideas generated by AI.
1:25:30Let's take a look at
1:25:32a video from Bill and talk about it.
1:25:34>> Recently, AI did something that stopped
1:25:37me cold. I found that it could
1:25:38outperform my own superpower, coming up
1:25:41with business ideas. What was my role if
1:25:44AI had gotten so good at this? Then I
1:25:46realized if I could scale that ability,
1:25:49I could give it to the world for free.
1:25:51That got me very excited. The more than
1:25:531 million business ideas we have
1:25:54developed are completely free. Come
1:25:56explore idealab.ai. Give us feedback and
1:26:00find an idea worth building and make an
1:26:02important positive impact on the world.
1:26:04>> Nice Bill. So here's how it works. The
1:26:06system analyzes more than 17 million
1:26:08sources and applied Bill's own 31
1:26:11methods for spotting business
1:26:13opportunities such as unmet needs, new
1:26:15technologies, regulatory shifts,
1:26:17cultural changes. On the site you can
1:26:19filter ideas by your interest,
1:26:21geography, how much money you'd need to
1:26:23start them, and market size. His goal, a
1:26:25million new businesses and a hundred
1:26:27million new jobs. Bill told Fortune
1:26:30quote, "With Idealab I could do 10
1:26:32companies at once. Now I can do 10,000
1:26:35at once."
1:26:37Cloudflare's Matthew Prince has
1:26:38committed $100,000 in credits to each
1:26:40startup built on this platform. So for
1:26:43me this is the abundance thesis at
1:26:45large. Business ideas are no longer
1:26:47scarce, they're free. Now scarcity moves
1:26:50to execution, taste, and courage. And of
1:26:52course AI can help you build your
1:26:54business ideas. You know, there was a
1:26:56great
1:26:57article by Mark Suster at Upfront
1:26:59Ventures looking at the cost of
1:27:02launching a tech startup. He said back
1:27:04in 2000 it was about $5 million.
1:27:07You know, 10 years ago it was about
1:27:08$50,000. Today, you know, it's a couple
1:27:11of K to get a company going. So to your
1:27:14point I mean you funded hundreds of
1:27:16founders. Does a free list of 1.2
1:27:19million ideas help? Or is the idea the
1:27:23least valuable part of the startup? How
1:27:24excited about about this are you?
1:27:26>> I don't think ideas are all that
1:27:28valuable.
1:27:29I think they're mostly a dime a dozen. I
1:27:31mean I
1:27:32>> Agree.
1:27:33>> don't think I've had an original idea in
1:27:34my entire life.
1:27:36>> [laughter]
1:27:38>> Somebody Somebody's Somebody's had it
1:27:40many times before me.
1:27:42Um
1:27:43So
1:27:44>> Uh
1:27:45>> There's just
1:27:46>> It's all about execution.
1:27:48>> of
1:27:48Yeah, just the difficulty of starting
1:27:50and putting one foot in front of the
1:27:51other and
1:27:52and um
1:27:54it's the it's the low points and
1:27:57working through all of that that that
1:27:59just has nothing to do with an idea. It
1:28:00has everything to do with your makeup.
1:28:03Um
1:28:04And I I've just been I've just been
1:28:06through so many failures.
1:28:08Um
1:28:09I just think that
1:28:11So the idea is fine, um
1:28:13but if the person it doesn't define what
1:28:15the person is in terms of his or her
1:28:17capability to
1:28:19um to manage through that. Uh that's
1:28:21that's the key.
1:28:23>> Uh Saleem, we've been talking about one
1:28:25person unicorns.
1:28:27>> Yeah.
1:28:29>> What are your thoughts on that?
1:28:30>> I'm very very excited by this and I'll
1:28:32tell you why. You know, people like us,
1:28:34Chamath certainly, you are swimming in
1:28:36ideas. People are coming to you with you
1:28:38with ideas a lot of the time. There's a
1:28:40lot of people that are kind of stuck
1:28:41going, "I don't know what I want to do."
1:28:43Uh and I think what this gives them is a
1:28:44tick list to go, "Oh, well, does that
1:28:46excite me? Does that excite me? Does
1:28:48that excite me or not?" I actually went
1:28:50through this and I found four that I
1:28:51thought were absolutely amazing. I want
1:28:53to just throw them out. One was a global
1:28:56global abundance grid where you replace
1:28:59resource scarcity with algorithmic
1:29:01orchestration. Right? So you find a a
1:29:03decentralized planetary scale
1:29:05optimization layer that looks at
1:29:07physical resources as fluid assets and
1:29:10reroutes as needed. I thought that was
1:29:12really amazing. Uh second one was
1:29:14post-bureaucratic liquid government
1:29:16protocols. That means you get rid of
1:29:18centralized parliaments, which is
1:29:21representative government doesn't work
1:29:22in a world where
1:29:26that was a for a world of information
1:29:28scarcity. We don't need a representative
1:29:30in Missouri come across the country to
1:29:32say, "Here's what my people are
1:29:33thinking." You know with social media
1:29:35what they're thinking and you can pull
1:29:36it on a real-time basis and you can deal
1:29:38with that. A third was a open-source
1:29:41planetary R&D network, which I found
1:29:43absolutely fascinating, kind of an open
1:29:45science
1:29:46kind of concept taken to full extreme.
1:29:50And the final one I found super exciting
1:29:51was algorithmic of time with in legal
1:29:54structures conflict resolution
1:29:57with with where you have to pay to play
1:29:59and you could do an open access where
1:30:02layer of judicial infrastructure almost
1:30:04becomes part of the internet. The
1:30:07Frederico asked this has created a
1:30:09wholly separate judicial system for
1:30:12small claims in South America and he's
1:30:14tilting towards this direction. But
1:30:15there was a few here that I found
1:30:17thought was absolutely amazing. What I
1:30:19love about this is this you could
1:30:21essentially take this give this to an AI
1:30:24and say go execute on all of these and
1:30:26start seeing what works and what
1:30:27doesn't, right? We should be able to
1:30:29really make a massive difference. You
1:30:31could incorporate Argentinian
1:30:35AI native organizations, point them at
1:30:37this thing and go go run all of these,
1:30:40right? Let's start on 1.2 million
1:30:42companies and see which ones work or not
1:30:43at a very low cost. This is I found this
1:30:46incredibly exciting.
1:30:47>> Alex,
1:30:49how were how were you involved in this
1:30:50with Bill? You mentioned something to
1:30:51me.
1:30:51>> Yeah, so I should disclose as a
1:30:54preliminary matter that I have a
1:30:55financial interest in the venture behind
1:30:57this. So more than a year ago, I said to
1:31:00Bill
1:31:01>> Was this at the Abundance Summit or was
1:31:02this after before?
1:31:03>> it was it was before.
1:31:05I went to him and said, "Look, you have
1:31:08this storied history of starting lots of
1:31:10companies. You may not believe me and by
1:31:12the way, I hope Bill you're listening."
1:31:14He didn't believe me at first. I said to
1:31:16him, "AI in a year or two is going to
1:31:19have the ability not just to incept good
1:31:22businesses but also execute on many of
1:31:25them." He didn't believe me. Fast
1:31:27forward a year,
1:31:28he started to believe because the
1:31:30capabilities continued to improve
1:31:32exponentially. So this is the result
1:31:35idealab.ai. I think there's
1:31:37I'll I'll be limited in what I say.
1:31:39There's much more to come, but I I think
1:31:42the idea that we achieve universal high
1:31:44income will require to some extent that
1:31:47AIs again not just or SIs not just
1:31:51making ideas too cheap to meter, but
1:31:54also as Saleem you gestured, idea
1:31:57execution too cheap to meter as well. I
1:32:00think we are going to see a
1:32:01proliferation of so many companies or
1:32:05business units, either way, that are
1:32:07basically end-to-end AI ideation,
1:32:11management, execution engines, and
1:32:14that's running the economy. It makes
1:32:15almost no sense to the earlier
1:32:17discussion about Elon in a world,
1:32:20although I don't actually think this is
1:32:21going to be the case, where Agent
1:32:23Smith-style most of the thinking is done
1:32:25by machines and not by humans for most
1:32:28of the businesses not to be run by the
1:32:31machines as well. So, I I think call
1:32:34this a baby step in the direction of
1:32:36having many, many businesses, not just
1:32:38incepted by the machines, but also
1:32:41operated by the machines.
1:32:43>> I mean, Chamath, one of the
1:32:44conversations we have on the pod a lot
1:32:45is, you know, the old social contract of
1:32:49do well in high school, get a good
1:32:50college, go get a go get a degree is
1:32:52cooked or going to be cooked and it's
1:32:55far more important to find something
1:32:56that you find purposeful in life and
1:32:58then go build something. You know,
1:33:00become an entrepreneur, have agency in
1:33:02your life and and and build. And I I
1:33:04think to Saleem's your point, you know,
1:33:06for people who don't know what they want
1:33:08to do, this is at least a a shopping,
1:33:11you know, mall to go and and find ideas
1:33:14and then and then work on it.
1:33:16>> And consider the the idea that you could
1:33:18you this takes lean startup to the next
1:33:20level, right? There's a 1.2 million
1:33:22experiments that can now be starting to
1:33:24be run and modeled out. They the
1:33:27implications are fantastic for this.
1:33:30Uh it's uh I really love the the the the
1:33:33leap that you made to say, I've
1:33:35generated a million ideas, but AI should
1:33:37be able to help me do that. Now, let's
1:33:38just go document them.
1:33:39>> But I still think if you, you know, if
1:33:42if you're
1:33:43aware in the world and you find a
1:33:45problem that you are passionate about or
1:33:48something you hate that exists, you can
1:33:50take that and just put it into your
1:33:52favorite large language model and say,
1:33:54"Hey, help me ideate an idea and help
1:33:56me, you know, design a business, and let
1:33:58me, you know, run a a prototype for me.
1:34:01I mean, you can do this just by being
1:34:03aware of what you love or what you hate.
1:34:06>> You know, I
1:34:08did not grow up in an entrepreneurial
1:34:10family, right? My mother was a German
1:34:12professor, my dad was an engineer. It
1:34:14was at Waterloo when I took a course on
1:34:16technical entrepreneurship that I went,
1:34:18"Oh, you can actually build a business?
1:34:20Like, there's a thing That's a thing
1:34:22that you can do that?" And if I'd had
1:34:24then it took me like another 20 30 years
1:34:26to actually get the full entrepreneurial
1:34:28bug. If I had this list when I was
1:34:30younger, I would have gone, "Wow, look
1:34:31at all these things. Let me pick one and
1:34:33go with it. Pick which picture or
1:34:34whatever one strikes my fancy at that
1:34:36time." So, I think this is incredibly
1:34:38exciting for younger generation that are
1:34:40a lot of them are lost not knowing what
1:34:42they want to do. This gives them a
1:34:43starting point.
1:34:46>> Dave, how about you, pal?
1:34:48>> Actually, before I lose the thought, I
1:34:49want to riff on Waterloo for a minute
1:34:51here cuz I was just out there last week
1:34:52with Mike Lazaridis, and I cannot
1:34:54believe how that place is. I'd never
1:34:56been there before.
1:34:58You know, I I I've always heard it's the
1:34:59MIT of Canada, you know, blah blah blah,
1:35:01but it's actually now the crossroads of
1:35:02all the talent in the world that wants
1:35:04to get involved in AI and knows their
1:35:07home country is doing jack squat,
1:35:09but you can't as easily get into the US
1:35:12or into Palo Alto necessarily. Uh
1:35:14Waterloo is just as good, and boy are
1:35:16they attracting I mean, you'd think
1:35:19you're in the crossroads of the entire
1:35:21world just culturally
1:35:23and the brain power and the size of the
1:35:25school. It's like 20 What? 25,000
1:35:28students? It's like five times bigger
1:35:30than MIT and Harvard 50,000 students.
1:35:32It's like this mecca of technologists
1:35:35and optimists that I had never seen
1:35:36before.
1:35:37I mean, highly I know it's not really on
1:35:39the way anywhere, but I highly encourage
1:35:41people go check it out. It's amazing,
1:35:44and there's tons of power there.
1:35:46Which means they're going to be data
1:35:47centers there, which means it's going to
1:35:49be something big is going to happen. I'm
1:35:51telling you.
1:35:52Um so, on Bill Gross, I you know, I I'm
1:35:54on the board of Parity I see Bill quite
1:35:56a bit. Um
1:35:58yeah, I I think the ideas are a wide
1:36:01range, you know, some of which you can
1:36:02act on with virtually no budget. Those
1:36:04are kind of cool, you just go. But, a
1:36:06lot more of them require some kind of
1:36:08funding and some kind of compute. So, I
1:36:10think Bill, by unleashing the entire
1:36:12list, people will pick up a really good
1:36:14idea off the list, but then they have to
1:36:15come back to him and say, "Hey,
1:36:17fund help me get funded or help me get
1:36:19the computer I need to do this." So,
1:36:21it's a really good business plan for
1:36:22him. A lot of this is going to come back
1:36:24to him. So, now I know why he
1:36:26open-sourced the whole thing and, you
1:36:27know, I think the ideas are only going
1:36:29to get better.
1:36:30He's clearly going where the puck is
1:36:32going. So, it's really pretty brilliant.
1:36:34But, the the thing that makes it make
1:36:35the most sense is that a lot of them
1:36:37have to call them up and say, "Hey,
1:36:39I want to execute on idea number
1:36:41800,462.
1:36:44>> [laughter]
1:36:44>> Help."
1:36:45>> Yeah. I mean, it's a great magnet for
1:36:47entrepreneurs.
1:36:49If you want to collect all their emails
1:36:50and their contact details, for sure.
1:36:52>> Everybody, [snorts] welcome to the
1:36:53health section of Moonshots brought you
1:36:55by Fountain Life. You know, we talk
1:36:56about AI on this Moonshots podcast all
1:36:58the time. One of the most important
1:37:00things AI is going to be able to do for
1:37:02you, besides educating your kids and
1:37:03helping you with your taxes, is making
1:37:06sure that you're living a healthy
1:37:07lifestyle, that you get a chance to get
1:37:10to 100-plus. I'm here today with Dr. Don
1:37:13Mussalem, the chief medical officer of
1:37:15Fountain Life and a part of my medical
1:37:17team. Don, a pleasure.
1:37:19>> Great to be here.
1:37:19>> You know, the thing that people are
1:37:21concerned about most about living to 100
1:37:23or 120 is their cognitive abilities,
1:37:26making sure they don't have dementia.
1:37:29And, uh, the numbers about dementia are
1:37:32problematic. Uh, can you share what
1:37:34you've learned?
1:37:35>> Such an important point and you're
1:37:36right, at Fountain Life, our members,
1:37:38the number one thing people are most
1:37:40concerned about is losing their brain
1:37:42health. Forgetting the name of their
1:37:43child, forgetting the face of their
1:37:45loved one. We know that when it comes to
1:37:46dementia, the conservative estimates are
1:37:48that 45% are entirely preventable. What
1:37:52was amazing is with the advanced testing
1:37:55we're doing at Fountain Life, one
1:37:57quarter of our members had advanced
1:37:59brain age.
1:37:59>> Wow.
1:38:00>> But, what was really awesome is, again,
1:38:02back to that prevention. When he
1:38:03partnered it with healthy living, this
1:38:05gives me chills, eating healthier,
1:38:07moving our bodies, sleep. Optimizing
1:38:10sleep is so important. You know what we
1:38:11saw? We saw that we improved that brain
1:38:13age by 26%.
1:38:16That is a big, big number to show that
1:38:18the majority of those individuals were
1:38:20able actually to improve the brain age.
1:38:22>> And one of the things I love about
1:38:23Fountain is we're searching the world
1:38:24for the best therapeutics, the best
1:38:26approaches, and making sure we bring it
1:38:28to our members.
1:38:29So, if having healthy brain function
1:38:33till 100, 120 is important to you, check
1:38:36out Fountain Life. Go to
1:38:37fountainlife.com/peter.
1:38:39Make sure you become the CEO of your own
1:38:41health. All right. Now, back to the
1:38:42episode. The next story here reads like
1:38:44a spy novel. On October 9th, the FBI
1:38:47arrested Abdul Rahman Alfonsi, a
1:38:5035-year-old Saudi national and OpenAI
1:38:53engineer.
1:38:54Prosecutors say on September 28th, he
1:38:56hid a recording device in a USB port on
1:38:59the back of a TV in an executive
1:39:01conference room at OpenAI. It stayed
1:39:03there for days. That room hosts OpenAI's
1:39:06most sensitive technical and strategic
1:39:07discussions. Here's the twist.
1:39:09Prosecutors say he used an AI chatbot on
1:39:12his wife's account to learn how to build
1:39:14the encrypted recorder.
1:39:16And this weekend, new court filings say
1:39:18he was frequent WhatsApp contact he was
1:39:21in frequent WhatsApp contact with a
1:39:23member of the Saudi royal family who
1:39:25encouraged him to come home and join a
1:39:27Saudi AI initiative. On Monday, a
1:39:29federal court ordered him to stay in
1:39:31jail after requesting a million-dollar
1:39:33bond from the Saudi government had
1:39:35offered. It's worth pointing out that
1:39:38these are still unproven allegations.
1:39:40OpenAI says no evidence of unauthorized
1:39:43access to weight models or source code
1:39:45or customer data
1:39:47was acquired.
1:39:49So, I mean, my question, how in the
1:39:50world did they find this USB recording
1:39:53device in the first place? And how many
1:39:55frontier models are out there now
1:39:56searching for bugs?
1:39:58I'm curious, Alex, what's more valuable
1:40:01to steal these days, model weights
1:40:03uh what's said in the conference room?
1:40:05>> Allegedly, the conference room, which is
1:40:07a little bit strange. I I I think that
1:40:09the way one would siphon model weights
1:40:12is the still favored way is distillation
1:40:14attacks from China via reasoning traces.
1:40:18But I I I do think putting the
1:40:20allegations aside, bigger picture,
1:40:22there's a great game afoot between great
1:40:24powers. And in the 1950s, it was who
1:40:27held nuclear secrets. It was over nukes.
1:40:29And then in the 1970s, it was oil. And
1:40:33now in the 2020s, it's super
1:40:35intelligence. And I I do think Chamath,
1:40:38to your earlier point, I I think you
1:40:40were gesturing at spheres of influence
1:40:42or the Pax Silicon Western Allied sphere
1:40:46and maybe the Pax Sinica uh for the
1:40:49Eastern sphere. I I think we are
1:40:51probably going to end up in a world in a
1:40:54few years, to the extent we're not there
1:40:56already, where there is a de facto
1:40:58second Cold War between the West and the
1:41:01East over super intelligence. And then
1:41:03you have the in-between states, Middle
1:41:05Eastern countries, European countries
1:41:07who feel tugs either way. Some of them
1:41:10try to remain unaligned or non-aligned.
1:41:13But ultimately, the tidal and
1:41:16gravitational pull of super
1:41:18intelligence, I think to the extent the
1:41:20world is either bipolar, probably
1:41:23bipolar, maybe tripolar if some third
1:41:26pole comes out of nowhere, is I think
1:41:28going to be very difficult to resist
1:41:30over the next few years. And I I think I
1:41:32would expect to see many stories like
1:41:33this.
1:41:34>> You mean Chamath, do do frontier labs
1:41:36now become national security targets?
1:41:38Should they be secured like defense
1:41:40contractors?
1:41:42>> Well, I think this is the same that's
1:41:43been happening in Silicon Valley for a
1:41:45while. I remember
1:41:47when I was at Facebook a long time ago,
1:41:48I had this conversation. I said, "Guys,
1:41:50we just passed 3,000 employees. We have
1:41:53spies from at least four countries." And
1:41:55people were like, "What are you talking
1:41:56about?" And I said, "Well, just look at
1:41:58the number of H-1Bs we have. And look at
1:42:00the distribution of them. And do you do
1:42:02not think one of,
1:42:04you know,
1:42:05from country A and one from country B
1:42:07and one from country C, and one from
1:42:08country D, couldn't be a spy. Of course,
1:42:10they they
1:42:11The odds are you should assume that they
1:42:13are.
1:42:14And, [snorts]
1:42:14you know, my posture at that time was
1:42:16you should just assume that most of
1:42:17everything we do is being exfiltrated.
1:42:19Um
1:42:21So, that's one example. Then, another
1:42:23example, I was a, you know, an investor
1:42:25in a company that had
1:42:27their stuff was supervised by ITAR.
1:42:30And, uh
1:42:32had a hack, and everything was
1:42:33exfiltrated. Like, everything.
1:42:36And, uh I remember like, you know, we
1:42:38brought in um third-party consultants,
1:42:40and the FBI, and the NSA, and um
1:42:44And, I thought, wow, like, you know, how
1:42:46common place is this? And, they were
1:42:47basically like, it's
1:42:49it happens everywhere. Um So, I guess
1:42:52what I'm saying is that, you know, this
1:42:53is a little bit
1:42:55um
1:42:57it's looks it's I I think I think I
1:42:59guess it's written or reported to sound
1:43:01more like
1:43:02uh cloak and dagger, 007 type stuff, but
1:43:05I think this is happening literally
1:43:06everywhere in every tech company that
1:43:08matters.
1:43:09The labs obviously matter. Um
1:43:12And, so the best way to deal with it is
1:43:14probably just to understand that
1:43:18some very minor distribution of your
1:43:21foreign folks are spies.
1:43:24And, could proceed accordingly.
1:43:27>> Well, and also recognize that how big
1:43:28the stakes are, you know, I think if you
1:43:30had spies like that at Meta or Facebook
1:43:32at the time,
1:43:33now the stakes are what, a thousand,
1:43:3510,000, a million times higher. So, the
1:43:38pressure from foreign governments to
1:43:40install spies is going to be
1:43:42proportional to the the amount that's at
1:43:44stake. I don't think California was ever
1:43:46set up for this, you know, like this is
1:43:48this is the way nuclear secrets are
1:43:50protected, but California has always run
1:43:52on, you know,
1:43:54people leave a big company to start a
1:43:56small company, and that keeps the chain
1:43:58reaction of the economy going. So, now
1:44:00you're like, well, we want to keep this
1:44:01all locked down like it's a nuclear
1:44:03secret because it can create terrible
1:44:05weapons. Well, we are totally not set up
1:44:08for that. You'd have to rethink the
1:44:09entire structure.
1:44:11>> Yeah, I just hope this doesn't become a
1:44:12sort of a national meme that starts to,
1:44:16you know, have us turn away foreign
1:44:18nationals. You know, I've always felt
1:44:19like when you get a PhD or master's
1:44:21degree in the United States, we should
1:44:22be stamping a green card onto it and
1:44:24encouraging those people to stay here
1:44:26and work. If this turns the other way
1:44:28and we start, you know, sending people
1:44:30back,
1:44:32I think that's a problem for the US.
1:44:34>> I should point out though this time is
1:44:35different. So, historically when there
1:44:37were secrets, for example, to Chamath's
1:44:39episode at Meta, the secrets themselves
1:44:42couldn't defend themselves. I I don't
1:44:44know quite what was targeted Chamath at
1:44:47Meta at the time, but to the extent that
1:44:49superintelligence is being targeted now
1:44:51allegedly at Frontier Labs, I I thought
1:44:54it was an interesting twist to the
1:44:55story, perhaps an inevitable twist, that
1:44:58OpenAI used their own AI in order to
1:45:01understand how the device allegedly that
1:45:04was planted worked. And that to me looks
1:45:06like breadcrumbs, baby steps towards the
1:45:09superintelligence itself preventing
1:45:12itself from being exfiltrated,
1:45:14preventing itself from winding up
1:45:16perhaps in undesirable hands, which if
1:45:18if you look at nuclear secrets in the
1:45:201950s, nuclear secrets couldn't defend
1:45:23themselves. You look at games over oil,
1:45:26who owned oil, oil couldn't defend
1:45:28itself, but now superintelligence in
1:45:30some sense is the first critical and
1:45:31emerging technology that's potentially
1:45:34capable of defending itself
1:45:35territorially, and that's a new factor.
1:45:37>> goes gallium, yeah.
1:45:38>> Yeah.
1:45:39>> Yeah.
1:45:40>> Yeah. Now, there's the flip side of that
1:45:41is if you look at, you know, the Llama
1:45:433, Llama 4 models and the changes they
1:45:46made. Now, those were crappy models, but
1:45:48you know, stick with me.
1:45:49Um
1:45:50if you read the research papers cuz
1:45:52those were all open source,
1:45:53the changes that worked and didn't work,
1:45:55you could summarize in like 10
1:45:57sentences. So, if you've bugged the
1:45:59conference room and you pick out the
1:46:01hey, after a hundred million dollars of
1:46:02R&D and all these tests,
1:46:05the Swiglu worked really well.
1:46:08That one sentence is so valuable to
1:46:12China or to somebody else coming up the
1:46:14curve. So, it's really really hard to
1:46:15contain something so small yet so
1:46:18impactful on their research, their
1:46:20budget, their timeline.
1:46:22So, it's tricky. It's really
1:46:24it's hard to say, "Hey, hey, AI, don't
1:46:25let that happen." Well, how's it going
1:46:27to
1:46:28how's it going to defensively co-scale
1:46:30against something so slippery?
1:46:32>> The the way I mean, if if I had to paint
1:46:34the the sci-fi thought experiment, it is
1:46:37to the extent that the frontier labs are
1:46:39locked in a recursive self-improvement
1:46:41loop. It's the AI's designing the
1:46:42replacements, not humans sitting in a
1:46:44conference room. So, if it's just AI's
1:46:47designing further AIs, then the AIs can
1:46:50monitor their own digital conference
1:46:51rooms and make sure there are no
1:46:52interlopers.
1:46:54>> Yeah, it kind of implies though then
1:46:56there's nobody at OpenAI that knows what
1:46:58it's doing to talk in a conference room.
1:46:59And that's kind of a scary scary end to
1:47:02that sentence. Like, yeah, yeah, we
1:47:03don't have any idea what it's doing
1:47:04anyway, so bug us all you want.
1:47:06>> [laughter]
1:47:07>> That's not good.
1:47:08>> Uh Chamath, we want to close out with a
1:47:11uh Chamath hot seat if you're open for
1:47:12it.
1:47:13>> Great.
1:47:14>> All right. So, uh I'll kick it off. So,
1:47:16OpenAI solves 722 math problems.
1:47:20Uh none of them reported encryption. I'm
1:47:23assuming that encryption was to a
1:47:25significant degree cracked if it was
1:47:27pointed in that direction.
1:47:29Uh what's your view on crypto uh given,
1:47:32you know, the meteoric rise of
1:47:35superintelligence here? Are we safe or
1:47:37we not?
1:47:38>> it's capped. I think that traditional
1:47:40crypto um lacks the fungibility to have
1:47:44all forms of buyers. The buying pool is
1:47:46really retail,
1:47:47then institutionals, and then central
1:47:49banks. And retail lacks the fungibility
1:47:51to be owned by central banks. So, it's
1:47:53capped from that perspective. Then it's
1:47:54capped because um
1:47:56the
1:47:58there are enough leaks
1:48:00around who owns what that now creates
1:48:03security risks. I think France being the
1:48:04best example where I think the French
1:48:06government forced publication of who
1:48:08owned what crypto assets, and you had
1:48:10just a complete flurry of kidnappings
1:48:12and, you you all kinds of really bad
1:48:14things. And then the third is around AI
1:48:16where if you believe that, you know,
1:48:18recursive self-improvement is pointed
1:48:20towards, you know, cryptographic hacks,
1:48:24um the most obvious thing to point it to
1:48:26would be to Satoshi's wallet and, you
1:48:28know, to break SHA-256 and just kind of
1:48:31drain the wallet and then go short and
1:48:33then tell people that you that you did
1:48:35it. So, then you'd you'd make you'd make
1:48:37the money twice.
1:48:39All those things I think are realistic
1:48:40reasons why it's capped and there's an
1:48:42there's an upper bound and so it's
1:48:44incumbent on these um
1:48:47these efforts to migrate to um
1:48:50um quantum-secure,
1:48:53quantum-resistant cryptographic
1:48:55mechanisms, which as you can see in the
1:48:57Bitcoin community is basically
1:48:58impossible because it's a bunch of
1:49:01people that can't even agree, you know,
1:49:03where to meet, let alone
1:49:04>> [laughter]
1:49:05>> let alone,
1:49:06you know, at what time the meeting
1:49:08starts at. So,
1:49:09it's I think crypto's a bit lost in the
1:49:11woods here.
1:49:11>> So, do you think Do you think Bitcoin is
1:49:14cooked?
1:49:15>> Uh I just think it's capped and then I
1:49:17think eventually it gets hacked and then
1:49:19it goes to zero.
1:49:19>> Do you own Bitcoin? Are you willing to
1:49:21say?
1:49:21>> Uh through a company, through a company
1:49:23called Digital Currency Group. So, I
1:49:24have this very long-term investment in
1:49:27Bitcoin. I seeded it to DCG a long time
1:49:30ago. They started up uh a bunch of ETFs.
1:49:33They gave me stock in return and that
1:49:35was much better for me because then I
1:49:36alleviated myself of the storage risk
1:49:38and the security risk and I had equity.
1:49:40Um so, that's worked out quite well. Um
1:49:43I think DCG's great. That's my mechanism
1:49:45to have exposure to crypto. I don't I
1:49:46don't own crypto directly anymore.
1:49:48>> Got it. Alex, over to you.
1:49:50I'll just note not very bullish on
1:49:53crypto, obviously. Crypto cooked, not
1:49:55investment advice. Anthropic's ECDSA
1:49:58announcement demonstrates proof tour de
1:50:01force that frontier models can probably
1:50:05achieve astonishing inversion attacks
1:50:07against crypto systems. So, not bullish.
1:50:09Question for Chamath, um SPACs. So, uh
1:50:14American research universities, I would
1:50:16argue, are sitting on an enormous amount
1:50:19of intellectual capital that they're
1:50:21scared, arguably, of translating out
1:50:24into commercial practice because they
1:50:25don't want to be treated like for-profit
1:50:27venture capital funds. They want to
1:50:29receive the the tax-light treatment of
1:50:31non-profit academic institutions.
1:50:34What would you think? I'll float this as
1:50:36an idea first because this is a favorite
1:50:38hobbyhorse of mine, of reviving, you
1:50:41know, you as sort of the the godfather
1:50:43of the SPAC movement,
1:50:45reviving SPACs for acquiring academic
1:50:49research universities in America to take
1:50:52them for-profit with government buy-in
1:50:55and IPO-ing them. And I I ask because
1:50:58the
1:50:58>> Well, I love that idea. That's a
1:51:00fabulous idea.
1:51:01>> So, maybe maybe
1:51:02it's on Bill Ackman's list.
1:51:03>> Heterodoxical, that's a great idea.
1:51:05>> Maybe maybe you and I should do that
1:51:07together. I would call them sparks
1:51:09rather
1:51:09>> university would you target first?
1:51:11>> I want to IPO Harvard University. I want
1:51:14to IPO Harvard on the New York Stock
1:51:16Exchange. So, I I did this
1:51:18back-of-the-envelope analysis that if
1:51:19you look at the official Harvard
1:51:20endowment stated versus how valuable it
1:51:24would be if you could dissect the
1:51:26venture capital arm versus the real
1:51:27estate holdings versus the education
1:51:30versus the endowment versus the
1:51:32merchandise. It would be like four times
1:51:34more valuable if it could be an
1:51:36honest-to-goodness for-profit, maybe
1:51:38public benefit corporation conglomerate
1:51:40rather than stuck trying to hide from
1:51:42for-profit taxation the way it is right
1:51:45now.
1:51:46>> I think it's a great idea.
1:51:47Uh it did I mean, I think I think it's
1:51:49it's all it's almost too disruptive to
1:51:51be viable, but it's a great idea.
1:51:53>> OpenAI has demo- I figure OpenAI has
1:51:54demonstrated that it's possible now with
1:51:56appropriate government appeasement to
1:51:58take a not-for-profit that's quite
1:52:00valuable and turn it for-profit, so why
1:52:02not turn that loose SPAC-style on the
1:52:04economy?
1:52:04>> Yeah. I like the idea, but I think you'd
1:52:06have to be on the inside to do it, which
1:52:08is hard.
1:52:08>> Well, so maybe Chamath, you and I should
1:52:11explore this option.
1:52:13>> You need the leaders of Harvard
1:52:14Corporation to be in tune with it.
1:52:16>> Sure, requires maybe you start with
1:52:17something smaller than Harvard
1:52:18Corporation, but uh I glad to hear your
1:52:21endorsement of this idea. We should talk
1:52:23about it.
1:52:23>> yeah. I think it's a I think it's a
1:52:24really interesting idea.
1:52:25>> Amazing. Saleem, over to you, pal.
1:52:27>> Um okay, Chamath, I've got a tension I
1:52:30want to set up between you and Alex. Um
1:52:32you've said that AI could compress S&P
1:52:34valuations from 20 to X free cash flow
1:52:37to like 2 to 7 X, right? If you believe
1:52:40that, then why should anybody own the
1:52:42S&P 500? And Alex is famous for saying
1:52:46own the index as in own the market
1:52:48because it's you can't it's hard to beat
1:52:49the market. So, how do you resolve that?
1:52:52>> I think the index is not what it used to
1:52:54be. The index used to be the 500 best
1:52:56companies and it was owned and
1:52:57structured in a
1:52:59um
1:53:01I think it gave you
1:53:02implied diversification that it doesn't
1:53:04today. I think there's an enormous
1:53:05amount of revenue concentration in a
1:53:07handful of names and you essentially
1:53:09have this distribution of halves which
1:53:12are like less than 10 and have-nots
1:53:13which is then the 490 others.
1:53:16So, the index isn't what it what it used
1:53:18to be. So, inasmuch I think you have to
1:53:21be much more precise in owning other
1:53:22things
1:53:23um that are thematically relevant and
1:53:25important in these cycles. Um
1:53:29but more generally, yeah, I think that
1:53:31IP is mostly not valuable anymore. And
1:53:34inasmuch as IP is not valuable, I think
1:53:37that people's
1:53:38typical
1:53:39underwriting of like defensive moats is
1:53:41basically gone.
1:53:43And so, what's left over is cash flow. I
1:53:45think the only thing that matters right
1:53:47now is cash flow only as a bridge to
1:53:49whatever unit of value matters in the
1:53:50future. So, the more cash you have
1:53:52today, the more that you can translate
1:53:53it into whatever it is that people will
1:53:55use in the future.
1:53:56And [snorts] so, a cash flow businesses,
1:53:58you don't pay for future cash flow
1:53:59because you fundamentally can't
1:54:00underwrite past a certain number of
1:54:01years. So, I'm a much bigger fan of like
1:54:03take the S&P 500, take the global 2000,
1:54:06rank them by free cash flow yield, and
1:54:08just buy the top end of those, and I
1:54:10think you're probably better off over
1:54:12the next number of years.
1:54:13>> All right. Dave, over to you, pal.
1:54:15>> Uh selfish question cuz I'm a really,
1:54:17really curious. You were the first
1:54:18investor in Grok and then Nvidia bought
1:54:21it for 20 billion, another brilliant
1:54:23Chamath investment. Uh but now you say
1:54:26you won't invest in chips. So so just
1:54:27for the audience, Groq was an AI
1:54:29transformer accelerator dedicated chip.
1:54:32And now it's part of Nvidia. So now
1:54:33you're not doing chips, but you are
1:54:34doing power. Are you doing data center
1:54:37components, boxes? Like why not chips
1:54:39and why you know how how did you make
1:54:41that choice?
1:54:42>> You have to be able to go to markets
1:54:44where you look dumb. And um
1:54:48when I
1:54:50basically found Groq
1:54:52uh or found Jonathan and then you know
1:54:55we started to work together. The company
1:54:56didn't exist when I wrote the first
1:54:58check. It was literally like a ground-up
1:54:59startup, him, myself, and uh
1:55:02the the third co-founder.
1:55:04Um
1:55:06It sounded really stupid. Um
1:55:09and it sounded stupid to my partners, it
1:55:11sounded stupid to the industry, and they
1:55:13were all wrong, and I was right. Um
1:55:16now it's like conventional wisdom that
1:55:18you should have a a silicon startup.
1:55:20That's why it's completely not going to
1:55:21work. And it's not going to work for a
1:55:23couple of reasons. They They are right
1:55:25about the fact that you're going to have
1:55:26a bunch of domain-specific silicon.
1:55:28They are right that there's going to
1:55:29you're going to have um
1:55:32highly customized solutions that work in
1:55:35very specific operating environments.
1:55:36All of that is right.
1:55:38Um but you will not create enterprise
1:55:40value. And the reason you won't be able
1:55:41to create enterprise value is when you
1:55:43go and actually try to tape out a chip
1:55:45and then try to find production
1:55:46capacity, what you're going to find
1:55:48is
1:55:49all the memory's blocked. Um all the fab
1:55:52space is taken up. All the packaging
1:55:53capacity is absorbed. And now you're the
1:55:56nth person in line. And the nth person
1:55:58[clears throat] in line is effectively
1:55:59DOA.
1:56:01So I think what you will have is some
1:56:03really valuable IP, but again, that's
1:56:05going to go to zero because some other
1:56:07group will be able to direct a an AI
1:56:09model against the model against an AI
1:56:12model and say design me the the
1:56:14domain-specific architecture that I
1:56:16would use to optimize this for inference
1:56:18or for you know some other metric. So
1:56:21you're going to have all of these
1:56:22permutations. So silicon, the the the
1:56:24the surface area of the number of
1:56:25silicon options will explode.
1:56:27But, the ability to make equity returns
1:56:29will go to zero because that's now too
1:56:31obvious.
1:56:32Um
1:56:34if you look down in the stack where
1:56:36people don't go
1:56:37um because people think it's either too
1:56:39stupid or not worth it is in data
1:56:41centers because they think it's too blue
1:56:43collar and a lot of intellectuals will
1:56:44talk themselves out of lots of money.
1:56:47Uh but, that is where you should play.
1:56:49And then the much more speculative area
1:56:51is in applications, which is above. And
1:56:53um
1:56:53you know, TBD around how to make money
1:56:56in the application stack. But, those are
1:56:57the three areas.
1:56:59And um you kind of have to be able to
1:57:01you know, shuck and jive like ebb and
1:57:03flow. And it was obvious that the way to
1:57:05penetrate AI in 2015 was in the silicon
1:57:08layer.
1:57:09And to me um but, to me it's now totally
1:57:13saturated and no money will be made. And
1:57:15now I have to go down or go up. And so,
1:57:17I chose to do both. But,
1:57:18that I could be wrong. But, that that's
1:57:20that's sort of how I think about it.
1:57:21>> Yeah, that's my great advice.
1:57:23>> I'm going to sneak in one final question
1:57:25on behalf of our our viewers. Uh
1:57:28and if you're willing to, what's your
1:57:30investment advice during this hypersonic
1:57:32tsunami right now over the next 3 years?
1:57:34Where do people place capital?
1:57:36>> My gosh, I think that's a it's a very
1:57:38complicated question. I don't know. My
1:57:39honest answer is I don't know.
1:57:42Uh I'm the least invested I've ever been
1:57:45uh
1:57:45over like the last 15 years.
1:57:47>> Are you heavily in cash?
1:57:48>> the least invested.
1:57:50Um
1:57:52and so, yeah, I just don't I just don't
1:57:55know. I'm I'm in a So, I'm I'm you know,
1:57:57to Saleem's point, I'm in this
1:57:59schizophrenia about pivoting towards
1:58:02cash flow.
1:58:03Um
1:58:04you know, to Dave's question, I have
1:58:06gone up and down the stack.
1:58:08Um to Alex's question, I'm still trying
1:58:10to figure out whether there's a better
1:58:12way to create liquidity in the public
1:58:14market. So, different kinds of assets
1:58:16that have different payoff profiles
1:58:18could be available in the public markets
1:58:20when ASI actually lands.
1:58:24I'm I'm honestly the most um confused
1:58:27I've ever been.
1:58:29Um, and then I I I compare that to the
1:58:31surety that so many other people have in
1:58:33the market right now.
1:58:36And it just seems like late cycle, late
1:58:39stage behavior. And so I'm just kind of
1:58:41like, I don't know. I'm not smart enough
1:58:42to know and I just rather
1:58:45bide [snorts] my time and think through
1:58:46it all. Um, but I if I had to pick, what
1:58:48I would say is
1:58:51I like owning cash flow cuz I think
1:58:52it'll give me the oxygen to pivot to
1:58:55what will matter.
1:58:56>> Mhm. [clears throat]
1:58:57>> I like owning power because I think it
1:58:59is the constraint, the fundamental
1:59:01constraint of everything for the next 10
1:59:03or 15 years.
1:59:05And I like alternative mechanisms of
1:59:08deploying capital because I think that
1:59:10the traditional financial infrastructure
1:59:13has become so sclerotic and they will
1:59:16become even more closed if there's a
1:59:17retrenchment in that AI trade.
1:59:19>> Yeah, it's it's it's tough.
1:59:22All right, we always close out with a
1:59:24a fan video outro song. This one is
1:59:28called everything is cooked by Douglas
1:59:31Lance. And just for a little bit of
1:59:32history, uh,
1:59:34Alex has an incredible voice. Alex, you
1:59:36were performing in what uh, when you
1:59:38were younger?
1:59:39>> Yeah, my first career, this is not
1:59:41widely known, was as a boy soprano
1:59:43performing with the New York City Opera
1:59:45Company. I performed on stage in Carmen
1:59:48and La Bohème, uh, did a tour with
1:59:51Turandot, and then my voice changed. And
1:59:53then they kicked me out of the New York
1:59:54City Opera Company. You don't get to be
1:59:56a boy soprano for that many years.
1:59:58>> All right, well, here's another playback
2:00:00of your of your younger self.
2:00:06>> This is AI generated, by the way.
2:00:07>> right now? As I like to say,
2:00:11math and science, engineering,
2:00:14physics [music] and biology are cooked.
2:00:17>> Oh, this is awesome.
2:00:18>> Medicine, computer science, all the
2:00:21social sciences are cooked.
2:00:25What's on the menu right now? [music]
2:00:28The whole economy is cooked.
2:00:33>> [laughter]
2:00:34>> Oh, they didn't extend it? Wait, we got
2:00:36to 100 the 100 day.
2:00:37>> They're going to I I think Doug said
2:00:39that he's going to produce a full
2:00:40Broadway musical based on Didn't Ask for
2:00:42My Permission, by the way, but have at
2:00:45it. I'll be curious to see the result.
2:00:46>> Like a whole musical, like Quadrophenia.
2:00:48>> I I think a whole song, maybe a whole
2:00:50musical.
2:00:51>> And Jamali, thank you so much for
2:00:52joining us.
2:00:54>> Hey, I really appreciate being included.
2:00:56Thanks. I'm sorry it took so long to get
2:00:57scheduled, but thanks, Peter. I really
2:00:59appreciate it.
2:01:00>> I did. It was great.
2:01:05>> [music]

About this transcript

This page contains the full transcript of Chamath: AGI Is Here, Anthropic’s $100B Giveaway, Elon’s SpaceX Shares for 72M Kids | EP #304 by Peter H. Diamandis, generated from the public captions YouTube serves with the video. The transcript has 22,693 words across 3,630 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.