DOGE kills its first bill, Zuck vs OpenAI, Google's AI comeback with bestie Aaron Levie
00:00:00.000 | Jekyll, what just happened? Did you just have a moment? An emotional moment?
00:00:02.880 | I don't know. You know, I've been trying to get over Sax and it's hard to get over your ex.
00:00:07.920 | Even when they didn't treat you well?
00:00:09.840 | We were in a codependent relationship and we were working on it.
00:00:13.520 | You have no one to fight with anymore. Who are you going to fight with, Jekyll?
00:00:16.160 | That's the emptiness you feel inside because you're a fighter. You're a street brawler
00:00:20.160 | and you have no one to brawl with anymore. So you feel...
00:00:22.480 | Well, you're the first guy to jump under the table in a bar fight.
00:00:26.080 | I mean, here you go. Here you go. As I was saying, it's always worth a start.
00:00:30.400 | No, I feel like if this goes down, I feel like Sax and I would have like,
00:00:34.560 | we would have thrown down. We were bulldogs. We would have fought like tooth and nail.
00:00:38.640 | I have this image. It's like we're in a bar. Okay. And all of a sudden...
00:00:46.320 | You talking to me?
00:00:47.440 | A fight is about to break out. Four thought bubbles appear. Okay. Jekyll's like, "Let's
00:00:53.600 | go get him." Sax is like, "This guy's a dipshit. Let's roll." I'm like, "Look at that chick. She's
00:00:58.960 | so good looking." And then Freeberg's like, "What will happen to my invite to **** duck?"
00:01:04.160 | Freeberg's like, "Oh, how do I get out of this unscathed?"
00:01:10.320 | All right, everybody. Welcome to the number one podcast in the world.
00:01:32.720 | With me again today, for better or worse, your sultan of science, David Freeberg is back.
00:01:39.840 | How are you doing, Freeberg? I see you're a pawn. Only a pawn in their game.
00:01:44.560 | What's going on with the pawn background?
00:01:46.800 | The imposter syndrome.
00:01:48.960 | Pawn or queen? What is that from? Is that from a movie? What movie is that from?
00:01:54.800 | It's called The Imposter Syndrome.
00:01:56.000 | It's like some French new wave I'm not aware of.
00:02:00.720 | I've learned that Freeberg's background is some like weird, secret communication language. There's
00:02:07.600 | a small but fervent group of people that are really into these backgrounds. They're always
00:02:11.840 | trying to figure something out.
00:02:12.880 | Absolutely. Of course, with us again, Chamath Palihapitiya, your chairman,
00:02:16.560 | dictator. How are you doing, Chamath? Nice sweater.
00:02:18.960 | Thank you, sir. Good to see you.
00:02:20.720 | You ready for-
00:02:21.840 | Excited to see you.
00:02:23.040 | I know. Look at this. Three or four besties will be hitting the slopes.
00:02:26.480 | Three or four besties will be skiing together.
00:02:28.320 | Aaron might be there. Aaron, where are you going for the holidays?
00:02:30.160 | Oh, actually, we'll be in **** next week.
00:02:33.200 | What?
00:02:34.240 | Well, there we go.
00:02:35.200 | New bestie in ****.
00:02:36.400 | Aaron, are you staying in the town?
00:02:37.600 | Just at the hotel.
00:02:39.440 | Okay. This is perfect. We're all going to be there, bro.
00:02:42.320 | It's $2,000 a night. How many rooms you got with the kids and the fam? You have three?
00:02:46.800 | Bro, boxes at all-time highs. Take it easy.
00:02:49.280 | What did you do? Did you start buying Bitcoin with your treasury? What do you mean boxes
00:02:53.280 | at an all-time high? Are you doing a Michael Saylor or something?
00:02:56.320 | It turns out if you didn't get crushed during the post-COVID period, you can just keep cranking.
00:03:05.040 | That's what we've been doing.
00:03:06.320 | Oh, you've been building a real business, of course, with us now, our new fifth bestie,
00:03:09.840 | Aaron Levy. He is the CEO of the publicly traded company Box, which means
00:03:14.960 | he's got the most to lose by coming here. So welcome back to the program.
00:03:17.920 | Based on what you guys were just talking about before this recording started,
00:03:24.560 | I have no idea what we're in for.
00:03:26.480 | Yes.
00:03:27.600 | Listen, we're living in an age of meme stocks and people selling convertible notes to buy
00:03:33.120 | Bitcoin for their treasury and then becoming a Nasdaq 100. It's that simple. Aaron, when you see
00:03:38.560 | a company buy billions of dollars' worth of Bitcoin and get added to the Nasdaq 100,
00:03:44.320 | what method of suicide do you think of taking your own life with?
00:03:47.520 | There was a brief week in 2021 where the thought kind of crossed my mind.
00:03:55.200 | Yes.
00:03:56.080 | You're just going to use the sword, the short blade?
00:03:59.200 | We have a very sophisticated audit committee that prevented the action.
00:04:03.760 | Will you do this for me?
00:04:06.480 | Yes.
00:04:07.200 | Just do this for me. How much cash does Box have on the books, ballpark?
00:04:12.320 | We just did a convert, so we're probably 600 or 700 million.
00:04:17.520 | Okay, here's what I want to do. A little experiment for next week. Just put 5% of the
00:04:21.440 | treasury, 30 million in Bitcoin, and then we'll invite you back in two weeks and we'll see what
00:04:25.680 | happens. Okay? Just put 5% of the treasury in Bitcoin. Hey, everybody, here's another announcement,
00:04:30.960 | a little housekeeping. As you know, we successfully got the allin.com domain. That was a big victory
00:04:37.120 | for us. We now have an email list. Four years into this, Meshuggah, we now have the ability
00:04:45.280 | to take your email address and spam you with all kinds of great information, like a notification
00:04:50.960 | when the pod drops. Wow, so compelling. Insights from the besties who you've had enough of,
00:04:59.280 | and first dibs on overpriced event tickets to come hang out with us. Wow. This is the compelling
00:05:05.680 | pitch we have for giving us your emails. If you'd like to give us your email and get spammed,
00:05:10.160 | go to allin.com and let the spam begin. All right, that's out of the way. Sax is out again this week.
00:05:17.440 | I don't know what's going on. We've been trying to figure out where he is. He's MIA. If anybody
00:05:22.240 | knows... He's in DC this week, yeah. Oh, is that what's going on? Did you see all the
00:05:26.400 | meetings? I know, I'm being facetious. He's in meetings. Mr. Sax went to Washington,
00:05:34.640 | creating waves, but Mr. Sax is in Washington. Have you guys seen the photos?
00:05:40.400 | Aaron, what do you think of Sax in this role? I think it's a strong pick.
00:05:47.440 | Let the genuflecting begin. Go ahead and say more, Aaron.
00:05:53.920 | So Crypto I'm a little bit indifferent on. We haven't spent much time there,
00:06:03.120 | leaned in much there. But on AI, I think it's a very strong pick. And I think you want somebody
00:06:11.120 | that has a general deregulation, anti-regulation bent at this stage in the evolution of the
00:06:20.640 | technology. I think there's risk of slowing down too much progress right now. And I think he'll
00:06:27.920 | provide some good parameters and principles around how to avoid that. So I think very strong. And
00:06:34.240 | then Crypto again, don't know too much about. And then we'll see the rest of the topics.
00:06:38.240 | As a software CEO of a public company, when the Biden administration was putting forward
00:06:44.720 | their proposals on how to regulate AI and have definitions on the size of a model and
00:06:50.480 | what the models could or shouldn't do and the regulatory authority they would have over the
00:06:54.240 | software that's written, what was your reaction? And you were supporting Harris at the time,
00:06:58.800 | I believe, or Biden at the time, right? But how did you react to that when you saw those proposals?
00:07:04.320 | And just to be clear, are you talking about the EO that went out?
00:07:07.360 | Yeah, there was the EO, but then they were also drafting. They published a lot of detailed
00:07:12.560 | drafting. And then obviously California had its bill, which you probably saw as well,
00:07:16.240 | which specified the number of parameters in a model and the size of a model and
00:07:20.400 | all these kind of constraints. In reverse order was against SB 1047.
00:07:25.200 | It felt like you had two big issues. One, you probably don't want state-by-state legislation
00:07:31.040 | on this topic. That's going to be, you're in a world of hurt if every state has a different
00:07:35.680 | approach to this. And then secondarily, if you just look at how it evolved from the very first
00:07:41.280 | proposal to the final proposal, and unfortunately, the kind of underlying philosophy that was in the
00:07:46.800 | bill, it was very clearly a sort of like viewing basically AI progress as inherently risky right
00:07:54.320 | now. And so it just ratcheted up the different levels of consequence for the AI model developers.
00:08:02.000 | And the risk is sort of the second or third order effects of that, which is like, does Zuck then
00:08:07.120 | want to release the next version of Llama if you're taking on that much risk? And even the
00:08:12.720 | incremental updates, the liability you have in terms of any of the model advancements.
00:08:19.120 | And so right now we're benefiting from just an incredibly competitive market between five or
00:08:23.680 | six players, and you want them running as fast as possible, not having to have sort of this,
00:08:28.880 | you know, a whole council before every model gets released because they're, you know, in fear of
00:08:33.440 | getting sued by, you know, the government for hundreds of millions of dollars if one person
00:08:37.280 | does something wrong with the model. So that was the problem with SB 1047. That's been the problem
00:08:41.280 | with some of the proposals on national legislation. I felt like the first CEO, it didn't have a lot
00:08:46.080 | of teeth in it. So it kind of was more like, let's watch, you know, this space and continue
00:08:50.480 | to study it. The actual current head of OSTP, Aarti Prabhakar, a lot of folks in Silicon Valley
00:08:56.400 | know her. She's actually very strong, very technical, you know, understands the valley well,
00:09:00.960 | is not a sort of, does not lean into overregulation. So I actually think OSTP has had a pretty good
00:09:07.520 | steward, even under Biden. But I think, you know, the efforts that Sachs would, you know,
00:09:13.600 | clearly be leading, I think would lean even more toward AI progress and sort of not accidentally
00:09:19.680 | overregulating too early in this journey. >> So let me ask you a question then about
00:09:26.400 | crypto. You're not into crypto. Crypto is a little bit harder to regulate. So with Sachs being there,
00:09:32.960 | what do we think the one, two or three things he could do to actually make crypto
00:09:37.920 | not a scam, not have consumers get left holding the bag? Obviously, sandboxing projects,
00:09:47.440 | maybe having people know your customer, you know, some basic regulation there,
00:09:52.240 | the sophisticated investor test comes to mind. Shmoop, what do you think Sachs should do
00:09:57.280 | in terms of crypto regulation in the short term, in the near term?
00:10:01.040 | >> That's a really good question. I think that today there are a lot of
00:10:06.960 | really valuable use cases that can sit on top of crypto rails. I think the most obvious one is
00:10:16.160 | how you can have real-time payments using stable coins. I think the United States government is
00:10:22.320 | already using some of these stable coins for a bunch of purposes. The number of companies that
00:10:28.400 | are beginning to adopt and use stable coins is actually growing very quickly as well.
00:10:32.440 | >> Take a second to define stable coins for the audience just to catch people up.
00:10:36.240 | >> Let me define what a stable coin is, which is that you put a dollar into a wallet somewhere,
00:10:44.000 | and in return, you get a digital token of that dollar. There are two big purveyors of
00:10:50.960 | stable coins. There's Tether and then there's USDC, which is this company called Circle.
00:10:57.280 | I think the easiest way to disambiguate them is Tether is abroad. I think it's run out of
00:11:04.000 | somewhere in the Caribbean. USDC is American. It's run by this company called Circle.
00:11:09.840 | The other difference is that there is some ambiguity around how these assets are
00:11:17.840 | secured. What a stable coin is supposed to do is when you give them a dollar,
00:11:21.840 | they're supposed to go and buy some short-term treasury security so that that dollar is always
00:11:27.920 | guaranteed. Because if you had a billion dollars of stable coins, but only $900 million of assets
00:11:36.080 | to back them up, there's an insolvency risk there. There's a run on the bank risk. Theoretically,
00:11:41.520 | a billion dollars of stable coins should have a billion dollars of cash in some short-term
00:11:46.320 | fungible security. What's incredible is at the scale in which these stable coins operate,
00:11:51.760 | that has turned out to be an enormous business. Why? When you give them a billion dollars and
00:11:57.600 | get a billion dollars of stable coins in return, they just go and put it into the bank. When
00:12:02.000 | interest rates are two, three, four, five percent, they're making billions and billions of dollars.
00:12:08.400 | They get the float.
00:12:09.600 | They get the float. These businesses have turned out to be incredible, but that's beyond the point.
00:12:14.080 | The point is that a lot of companies that you would never think, so for example, SpaceX uses
00:12:19.120 | these stable coins. How? How do you think they collect payments from all the Starlink customers
00:12:24.560 | when they aggregate them in all of these long-tail countries? They don't want to necessarily take the
00:12:29.040 | FX risk, the foreign exchange risk. They don't want to deal with sending wires. What they do is
00:12:34.160 | they'll swap into stable coin, send it back to the United States, and then swap back to US dollars.
00:12:38.800 | It's a very useful utility. Number one is I think we need to make those rails the standard rails in
00:12:47.680 | the United States. What that does is it allows us to chip away at all of this decrepit infrastructure
00:12:54.960 | that the banks use to slow down and tax a process that should never have been taxed.
00:13:01.040 | This is good competition in a way, Chamath.
00:13:03.280 | It's incredible.
00:13:03.840 | Now, the banks have somebody to challenge them for money transfer and money storage,
00:13:10.080 | and it could be regulated and stable, but I guess the question that-
00:13:13.760 | That's the first thing, but then the second thing it allows you to do is you can see a
00:13:18.240 | world where now you can have real competition against the traditional rails, specifically Visa,
00:13:23.360 | MasterCard, American Express, because when you look at great companies like Stripe, I use Stripe,
00:13:29.200 | I pay them 3%. If I use stable coins from Stripe, I don't pay zero, but I don't pay 3%,
00:13:37.440 | it's kind of somewhere in the middle. If I was technically adept at implementing stable coins
00:13:42.400 | through the entire stack of my product, so I use it for this learn with me thing where we publish
00:13:48.640 | research, I would save a lot of money. The idea that you can take that 300 basis points you pay
00:13:55.600 | to these companies and crush it to zero would be a boom to global GDP, because that's 3%
00:14:02.480 | on many tens of trillions of dollars.
00:14:06.000 | Aaron, you're shaking your head. This is something that you're experimenting
00:14:09.600 | with at Box, or you're aware of, or a problem that you recognize.
00:14:13.760 | No, no, just the credit card rails is, I mean, the tax on transactions is obviously insane, so
00:14:19.440 | the stable coin being the intermediary for that in the future makes total sense if you could get
00:14:25.120 | everybody to kind of coordinate against that. So, yeah.
00:14:27.120 | Well, and regulating it would get people off of Tether, hopefully, which has a checkered,
00:14:32.640 | colorful, sorted pass, you can go look it up, but they've had many, many legal
00:14:37.040 | Passions against them, I'll leave it at that, yeah.
00:14:40.800 | Yeah, but in fairness to them, I think, again, both of these two companies look, as of today,
00:14:44.880 | again, you have a jurisdictional issue difference, but as of today, it looks like they're both
00:14:50.400 | peg one for one. But anyways, the point is, if Sachs can really push the adoption of stable
00:14:56.400 | coins, number one, and then number two is to incentivize much, much cheaper transactional
00:15:04.240 | rails, then I think he can go to Bitcoin and these other more long-tail crypto projects
00:15:10.960 | off of a back of momentum. Because these first two things, I think everybody will embrace,
00:15:16.480 | and he won't get caught in a political quagmire. These other things, you have these opponents
00:15:21.600 | always coming into the system, and they have, even Bitcoin, like when I said this thing about
00:15:26.960 | encryption and quantum, and even though I thought I was very clear, the crypto community on the
00:15:35.680 | internet went absolutely crazy all last week. Yes, they piled on, yeah.
00:15:40.000 | But the thing is, some of the maxis piled on, then when they actually took time to understand
00:15:43.840 | the technicalities of what I was saying, other people realized I was right, that they were
00:15:47.280 | tweeting it. My point is, that world is so- Religious.
00:15:52.400 | Animated and energized that I think it's hard to use them as the first place where you find
00:15:59.520 | progress. So I would tell Sachs- Got it.
00:16:01.200 | Go to stable coins, then disrupt the Visa rails, and then go to the other stuff.
00:16:05.760 | I would go stable coins, and I would go even before that, the accreditation test that I've
00:16:10.080 | been talking about, because the SEC has that mandate. People were educated, Aaron. They could
00:16:14.720 | buy crypto and know they're going to lose their money, or know that something's a fugazi or fugazi,
00:16:19.280 | whatever it is. Yeah, I mean, I'm sure we want to move on,
00:16:24.480 | but I guess the- Yeah.
00:16:25.600 | So there's a parallel universe where, so no matter what, like obviously, Gensler did not get his arms
00:16:32.800 | around this whole thing. So that was a big, big mistake. But there's sort of like DeFi/financial
00:16:41.360 | crypto, where almost everything is deflationary, it improves the rails. If you believe in Bitcoin
00:16:48.960 | as this store of value, and digital gold, all of those things can actually kind of make sense and
00:16:55.040 | be a bit rational and improve things. And then, unfortunately or not, depending on your views,
00:17:00.640 | there was this other sort of fractal event that happened, which was, oh, let's also use these
00:17:06.400 | things as a means of kind of creating a virtual currency and equities and tokens for anything,
00:17:15.760 | where that then runs into basically the SEC remit of like, are these things securities or not? And
00:17:22.800 | is there insider trading or not? And can anybody issue them at any time and promote them on Twitter
00:17:26.960 | or not? And so I think to some extent, if you could get back to like crypto 1.0, which was like,
00:17:32.800 | this is a financial infrastructure, I think you would have avoided a lot of the sort of noise and
00:17:38.800 | challenges with crypto. Now, I don't know, you can't put it back in the bag. But there's like,
00:17:44.800 | I don't think you could get 10 crypto people to agree on how you regulate that second category,
00:17:48.880 | because some people believe I should be able to issue an NFT on anything, and I should be
00:17:53.920 | able to trade that. And at the same time, they would obviously, they would sort of claim,
00:17:58.640 | well, that's just the same as an aftermarket, you know, seat to a concert. And yet another group
00:18:04.800 | would be treating this as effectively, you know, a security. And so I don't know how you're ever
00:18:10.240 | going to reel that in, without some people being upset, you know, within the crypto community.
00:18:14.320 | All right, well, more to come and Saks will be back, Saks will be back. And we will be
00:18:20.000 | rotating the fifth seat amongst, you know, friends of the pod, and newsmakers.
00:18:25.200 | Sorry, did I already, did I already get rotated out? Based on?
00:18:28.320 | Well, yeah, basically, the energy was a little low. But, you know, I mean, it's just,
00:18:33.200 | I think, well, what you're already had to warn what you're resting heart rate,
00:18:37.120 | what you're resting heart rate versus we're boys, and then we'll just make a decision.
00:18:40.560 | All right, listen, Doge is fully operational. Elon and Vivek have helped kill the last minute
00:18:48.960 | omnibus spending bill, Wednesday night, the bill had been killed. And we were looking at the
00:18:55.760 | government shutting down starting Friday, December 20. Today, when you're listening to this, for some
00:19:02.000 | quick background in September, Congress approved a bill that would keep the government funded
00:19:05.440 | through December 20, the day this episode published. Keep that December date in mind
00:19:10.880 | for a second. This Tuesday, December 17. Three days before the deadline leaders in Congress
00:19:17.280 | unveiled what was presented as a bipartisan stopgap bill that would keep the government
00:19:22.160 | funded through March 14. That kind of bill is called a continuing resolution, basically give
00:19:27.600 | you more time for the incoming GOP majority to reach an agreement on the funding for the
00:19:33.280 | government. But there are two major problems with the bill. It's a rush job, it had to pass the
00:19:37.840 | House and the Senate by Friday night after being presented on Tuesday. It's absurdly long 1500
00:19:43.760 | pages with $340 billion in spending, including pay raises and better healthcare benefits for
00:19:50.480 | the members of Congress. My Lord, read the room gentlemen and ladies, funding for a global
00:19:55.440 | engagement center for another year. That's the disinformation watchdog group that was wrapped up
00:19:59.760 | in the Twitter files, 130 billion to renew the farm bill for another year, $110 billion in
00:20:05.920 | hurricane disaster aid, just money being thrown everywhere, a billion three to replace the
00:20:11.440 | Francis Scott Key Bridge in Baltimore, they had some spicy comments on it. Congress has known
00:20:16.480 | about the deadline since they created it in late September. He said the urgency is 100%
00:20:22.160 | manufactured designed to avoid serious public debate. But serious public debate is exactly
00:20:26.880 | what's happening on Twitter. People are screenshotting and using chat GPT and Claude and
00:20:33.360 | Gemini to work their way through these documents. And it looks like this is not going to happen.
00:20:40.240 | Freeberg, your thoughts. So the proposed bill made no real change to the current spending level
00:20:46.000 | of the federal government at roughly $6.2 trillion on an annualized basis, which by the way, is
00:20:52.480 | roughly call it 23% of GDP. Just to give you some context, in 1860, nearly 100 years after the
00:21:01.440 | founding of the United States government, federal spending to GDP was less than 1%.
00:21:08.800 | And it took off during the Civil War for a couple of years. But, you know, we're at these kind of
00:21:15.120 | like unprecedented levels year after year now, really speaking to how the federal government
00:21:20.960 | has grown, as we talked about many times so much in our life. And, you know, our roads were really
00:21:25.760 | bad back then, though. Yeah, yeah, our roads were really bad back then. But remember, the objective
00:21:33.040 | of the republic was to have the states make local decisions about how to take care of their
00:21:37.760 | infrastructure. The national highway effort obviously changed that in the mid 20th century.
00:21:44.000 | But this was kind of the original intention of the republic. It wasn't to have the federal
00:21:48.480 | government come in and employ people, provide insurance to people, provide energy markets to
00:21:55.680 | people, own football stadiums, etc, etc, etc. If you go through the list of things in this bill,
00:22:01.600 | I think the $100 billion of natural disaster relief, everyone says that seems very reasonable,
00:22:07.520 | that's something the federal government should do when we have a natural disaster, we need help,
00:22:10.960 | we need support. That's a great thing for the federal government to do. But think about the
00:22:15.040 | incentive it creates, it distorts markets. So we've talked about this in the past, in areas
00:22:20.160 | where you have a higher probability of natural disasters, and people have paid a lot of money
00:22:24.240 | for their homes, pay a lot of money for infrastructure, should the federal government
00:22:28.080 | come in and rebuild those homes and provide capital to those individuals and those businesses
00:22:32.880 | to help them rebuild those homes, if there's a high probability of natural disaster events
00:22:36.960 | happening again in the future, it means that the cost of insurance doesn't matter. And the cost of
00:22:41.600 | real estate doesn't matter because the federal government effectively can come in and support
00:22:45.600 | those prices. In the same way, the federal government comes in and supports the prices
00:22:49.520 | in agricultural products, through the work in the farm bill, and through the biofuels mandates,
00:22:54.320 | which were also proposed to be extended in this thing. So the federal government's playing both
00:22:58.160 | a market role, and you know, also kind of this role that I think fills the gap where people want
00:23:04.320 | to have a customer where there isn't a customer and an employer where there isn't one. So like,
00:23:09.280 | how did we get to this point? So first principles perspective, we've kind of I think, lost the
00:23:13.520 | narrative on what the federal government was meant to do. If you think about the simple rubric in a
00:23:18.000 | bill, like just go back some period of time and someone says, Hey, I want something I want to have
00:23:22.640 | this in this bill. And you're the representative that's supposed to vote on that bill and say yes
00:23:26.960 | or no. It's very hard to just say, No, we are not going to spend that money. What's the incentive
00:23:33.520 | to say no? The alternative is you say yes, but give me x as well. There is an incentive in that
00:23:41.360 | response. And the incentive there is to get something for your electorate, the people that
00:23:45.760 | voted you in as a representative of the house, which is how we ended up at this point, where
00:23:50.720 | everyone says I want something if you're going to get something, and eventually the government,
00:23:54.800 | the federal government swells to spending roughly 24% of our GDP. Now, the biggest mistake I think
00:24:01.360 | the founding fathers made was that they didn't create constitutional limits on spending and
00:24:05.760 | enrichment. And this was because they had these deeply held philosophical beliefs that relied on
00:24:10.480 | the House of Representatives to provide a check by the people. If you read the Federalist Papers,
00:24:15.920 | and I went through a couple of these recently, and I use chat GPT to help me kind of,
00:24:19.520 | you know, bring out some of I think, the key points, but in the Federalist Papers 10 and 51,
00:24:26.160 | James Madison emphasized that the structure of government was meant to ensure that both state
00:24:30.720 | and federal governments would limit each other's excesses, including their financial ones. And then
00:24:35.360 | in the Federalist Paper number 58, he said, the House of Representatives has control over the
00:24:40.160 | quote, power of the purse, which gives the people's representatives authority over taxation
00:24:45.440 | and spending. But they also warned along with Alexander Hamilton, of the dangers of unchecked
00:24:51.600 | government power through burdensome taxation, and excess spending, which would ultimately erode
00:24:57.520 | individual freedoms. So they recognized that there were going to be limits, but their expectation was
00:25:02.160 | that the house and the individuals that were electing people to the house that were members
00:25:05.840 | of this republic would come in and say, we're going to keep that from happening. And clearly,
00:25:10.560 | something went wrong along the way that we got to this point, where again, spending is 24% of GDP.
00:25:16.720 | And I think that the biggest thing they got wrong was that they didn't create these constitutional
00:25:21.200 | limits on federal spending or taxation through either a balanced budget structure, spending as
00:25:26.880 | a percentage GDP, no federal debt or term limits or all of these other mechanisms that could have
00:25:31.600 | been introduced at the beginning, that could have created some structural limits. Instead,
00:25:36.640 | they assume that there was this natural limit that would emerge as a function of the democratic
00:25:40.560 | process because of how they formed the government. But unfortunately, I think they failed to realize
00:25:45.680 | that the electorate would eventually not want the freedoms of the people of the time.
00:25:53.680 | Back then in 1776, this was a pioneering country where everyone wanted to come here to have freedom
00:26:00.400 | to do anything they wanted, everywhere they wanted to build a business, to homestead,
00:26:04.560 | to be rugged individuals. It was entrepreneurial. Yeah, it was. And over the last 250 years,
00:26:10.960 | we've gotten used to an increment in lifestyle every year and discovered that we have a mechanism
00:26:16.560 | to force the increment in lifestyle through the actions of government. And so the electorate has
00:26:21.600 | stood up and said, "I want more each year, and I want the government to provide it for me if the
00:26:26.560 | free markets are not doing it." And that's really where we kind of got to this point where I think
00:26:30.480 | we elected people to the House who ultimately had this incentive that said, "If I give you this,
00:26:35.120 | I need to get this." And we ended up swallowing this. So I don't know if it cracks with Doge. I
00:26:40.160 | really don't know if people step up and recognize the limits of government and what the limits
00:26:45.360 | should be of the federal government on an electorate basis. It's an amazing moment to see
00:26:49.760 | that Elon went on Twitter and said, "Hey, guys, this is nuts." And everyone said, "This is nuts.
00:26:53.120 | We're not going to elect you if you do this." If that momentum and that transparency can keep up,
00:26:58.000 | I hope that people start to connect the dots, that this isn't a free lunch,
00:27:01.120 | that the federal government spending is not limitless, and it's not unaccountable.
00:27:05.120 | Aaron, I think we have enough people who are notable now speaking up about this excess spending
00:27:14.800 | and the out-of-control debt that it's now in vogue to talk about austerity, to talk about
00:27:21.520 | inefficiency. And that really all comes back to Doge and, dare I say, you know, the conversations
00:27:27.520 | we've been having on this podcast for two years, that this is becoming acute. What are your
00:27:30.880 | thoughts on this vibe shift, this complete pivot where we've gone from, "My Lord, everybody's
00:27:36.880 | saying, 'I got to get mine. You got yours. I'm getting mine,'" to name and shame. They're naming
00:27:43.120 | and shaming now very specific pieces of pork in these bills, you know, including stadiums for
00:27:50.640 | the NFL. And people are like, "Why is the NFL getting this if they're worth $20, $30, $40
00:27:54.640 | billion?" Two just quick thoughts. One, Patrick Carlson had a tweet yesterday that basically said
00:28:00.160 | this big misinformation kind of created by people that want to be slow is that you have to sort of
00:28:07.440 | choose two of fast, good, and cheap. And I think basically, you know, Elon's companies have sort of
00:28:14.560 | always effectively kind of proven the opposite, which is actually, if you just start to ask the
00:28:20.800 | question, "Why does that thing have to cost as much?" You know, if you're building a rocket or
00:28:26.080 | designing a car or developing batteries, if you just do ground-up, why does it have to cost as
00:28:30.720 | much? And so what's interesting is that probably if most people looked at what the government was
00:28:35.840 | spending on, they wouldn't even feel like, you know, it's not even helping them in like the
00:28:40.080 | disaster relief sense of, you know, I think like that there are probably actually people that
00:28:44.800 | actually do experience the benefits of disaster relief. It's actually just all of the overhead
00:28:50.960 | that we've created to getting anything done in the government that could actually make the
00:28:54.240 | government better serve all the constituents. I was talking to, you know, sort of a nameless
00:29:00.400 | individual in the government the other week, where by Congress, they have to hire contractors
00:29:06.400 | to do work. And the contractors, the contracting firms charge them two and a half times the sort
00:29:12.400 | of cost of an individual employee that they could otherwise hire. And so now they have to outsource
00:29:18.240 | the work, and they don't have any accountability mechanism for that contractor. And so I think
00:29:23.760 | there's not a single American that could look at that and say, "This is like actually working well."
00:29:28.400 | Like, yeah, we are spending more money to do less. And the ultimate outcome is actually lower
00:29:35.360 | quality. And so you have to at some point, just kind of do a little bit of a reset moment. And
00:29:40.480 | that's obviously the upside of Doge is like, it's like, it breaks every rule of us thinking about
00:29:45.200 | how you would actually go and attack this problem. We thought you'd attack it through meetings.
00:29:49.120 | And we would do it through congressional, you know, sort of processes and research. And it's
00:29:55.120 | actually just, it is, you know, obviously a much more sort of founder startup oriented way to
00:30:00.800 | approach this. There's gonna be lots of things that are broken glass around the edges. There's
00:30:05.120 | no question. But I think what's interesting about this week's event is, I think that there's been
00:30:09.760 | this underlying kind of notion of like, you know, Elon and whatnot at all, you know, don't understand
00:30:15.840 | the government enough to be able to change it. And it might actually be the case that the
00:30:19.600 | government doesn't understand Elon in the sense of, of like, he will just see this thing through.
00:30:25.440 | And the tools at his disposal and Doge's disposal is sort of, you know,
00:30:30.320 | completely unprecedented in terms of the ability to put any, anybody in Congress on notice if,
00:30:35.280 | you know, if basically they're promoting things that are not making the country better. So,
00:30:40.880 | so that the, you know, the thing that we saw this week was actually that playing out,
00:30:44.640 | everybody's been wondering, well, what are the actual, you know, what are the formal mechanisms
00:30:48.320 | Doge has to accomplish and enact change? And it's like, you just saw it, like, they can just,
00:30:53.600 | they can just create enough visibility and spotlight on the problem that it causes it
00:30:59.280 | a level of discomfort in, in supporting moving forward with, with whatever that thing is.
00:31:04.000 | And so I think it's interesting. I have no opinion on the actual elements of the bill,
00:31:07.920 | other than from a process standpoint and a, and a new kind of case study of how this is going to
00:31:13.680 | play out is I think we're seeing some early indication of what Doge will, will be able to do.
00:31:17.520 | Aaron, how do you feel about this Doge effort?
00:31:20.800 | As someone who is a public Harris supporter. So you've come out, I think you've, you've been
00:31:26.880 | public about this, but I want to understand like why people wouldn't be supportive of this effort,
00:31:33.200 | right? Like, like, what is like, what is the motivation for people saying this isn't a good
00:31:37.440 | thing? We shouldn't be doing this. Like, cause there's a lot of folks that have gone on these
00:31:41.200 | shows. It's like, they, they, they blast Elon, they blast Vivek, but like, aren't these principles,
00:31:46.480 | like, shouldn't they just be universal that we should not be wasting money and stuff?
00:31:50.960 | Sure. Of course. The, I mean, so to give some credit, you know, you have Ro Khanna supporting
00:31:56.880 | it. You have a Fetterman supporting it. Bernie Sanders, you know, everybody has their.
00:32:00.960 | Yeah. Bernie Sanders had a good call out for Elon.
00:32:03.360 | You have, everybody has their thing in the government budget that they don't like. So
00:32:06.800 | assuming that they see that as something Doge can contribute to, you could probably get
00:32:10.320 | actually broad support. You know, there's a classic, you know, sort of reflex within,
00:32:15.360 | within probably the Democrat party on, on at this point, just because of Elon's support of Trump,
00:32:19.280 | that, that if something is a, is an Elon project, they're going to, they're sort of going to
00:32:24.240 | instantly respond no matter what the thing is as a negative without, you know, kind of actually
00:32:29.120 | saying, is this, does this actually support actually something I do agree with? And so that's.
00:32:34.160 | It's all partisan, none of its first principles, right? I mean, for these people, for this group
00:32:39.120 | of people, which isn't everybody. That's going to be true of both sides. Like Michelle Obama,
00:32:43.200 | Michelle Obama was like, let's get kids healthy. And all of a sudden now it's in vogue to do that.
00:32:47.920 | So, so I think, I think we're just in an environment where, where any, anything will
00:32:52.080 | become partisan. What's interesting is, is that because of the, you know, some of the,
00:32:56.880 | the cross party elements of, of, of Trump and now his cabinet is it might pull in more,
00:33:02.640 | more of the Democrat party than, than would usually happen. And, and I think because of
00:33:07.200 | Elon and the people that are surrounding Trump, you probably have a bit more air cover for the
00:33:11.680 | Rocanas of the world to also step up. Because if it was like Steve Bannon and Trump doing Doge,
00:33:17.520 | it would be like, ah, okay, you know, maybe this is not the thing to, to, you know, lend credibility
00:33:21.440 | to for, for pure political reasons. If you have, you know, some of the best entrepreneurs that are,
00:33:25.920 | that are out there actually like literally in the cabinet driving this, at some point, you know,
00:33:31.280 | it's, it is an IQ test if you're, if you're on board or not. Shabbat, this is a sea change that
00:33:36.720 | we're seeing. And to Aaron's point, this administration gets to pick their priorities,
00:33:41.440 | but everything can't be a priority or it's not a priority. Therefore they've picked the
00:33:45.600 | priority of smaller government, more controlled spending over say mass deportation or removing
00:33:52.320 | more rights from women to choose when they have an abortion, et cetera. So this seems like a
00:33:57.120 | distinctly different focus for Trump 2.0, the second term. What are your thoughts on what we
00:34:02.400 | saw this past 72 hours? It was the most, it was the most incredible change in how politics will
00:34:11.920 | be done going forward. I think that people are probably underestimating what happened here. This
00:34:19.360 | was a multi hundred billion dollar grift that was stopped on a dime over 12 hours of tweets.
00:34:30.800 | You would have never thought that that was possible. The point is to put a dagger in
00:34:35.520 | something that big that had so much broad support just a few hours earlier. I think
00:34:43.440 | it's so consequential in how the United States can run going forward. So building on what you
00:34:48.960 | said, Chamath, also very interesting here is the fact that Trump hasn't taken office yet,
00:34:55.200 | and they're having this huge impact. This is occurring a month before he's even in office.
00:35:00.560 | If they can stop this, what will they do when they actually are in power?
00:35:04.560 | I think right now you're seeing the first order reaction, which is about the bill itself,
00:35:09.840 | and I think that misses the point. The bigger issue is going forward, you will have the ability
00:35:15.680 | to, and part of it is because we have a set of tools now that allows us to do this,
00:35:21.120 | to read 1500 pages in a matter of minutes, to summarize it into the key essential elements,
00:35:27.760 | to really understand what's being asked and what's being offered, and then to put it in a
00:35:34.240 | digestible format that normal people can consume. Then all you'll have to do is just connect the
00:35:40.000 | dots and tell your congressman or congresswoman that you'd like or dislike this thing, and what
00:35:44.960 | you're going to see is a much more active form of government. I think that's the really big deal,
00:35:49.600 | the fact that it really becomes the voice of the people.
00:35:54.000 | Now, the alternative can also happen. Imagine there is a piece of legislation
00:35:58.240 | that is very controversial, but it turns out that people actually want it. Then the opposite will
00:36:06.640 | also happen and can also happen now. I think that it's a very nuanced and interesting way that
00:36:11.920 | governance can happen. The other thing that I'll say though, which is funny, is we should have a
00:36:16.000 | segment called Today in Hypocrisy. If I was running the segment, what I would say is, "Today
00:36:21.360 | in Hypocrisy, you have a group of people, i.e. the Democrats, who are very upset and who now point to
00:36:30.800 | Elon as some shadow cabinet minister or some shadow president-elect or some shadow first buddy."
00:36:37.760 | First buddy. I love that.
00:36:41.120 | Except then I thought to myself, "Well, hold on a second." There was something untoward happening
00:36:46.960 | in the shadows and I thought, "Well, actually, this is the exact opposite." The guy is tweeting
00:36:51.440 | in real time his stream of consciousness. You absolutely know everything that he wants because
00:36:57.200 | he just lays it all bare. At the same time, I thought it was really interesting. The same people
00:37:02.160 | who were saying that were the ones that finally admitted that they were hiding Biden for the last
00:37:07.120 | two years. I thought, "Did we just miss this?" The same people that are like, "Hold on, Elon. I don't
00:37:12.960 | like the fact you're telling us what you actually want on Twitter transparently while we hide our
00:37:18.640 | President of the United States in a box." Well, yes. You're referring to a Wall Street
00:37:24.160 | Journal story that broke this morning.
00:37:26.320 | This week in Hypocrisy, Jason. Take it away.
00:37:28.560 | Yes. As we said on this pod, we knew that they were hiding him. Now, the cover-up is worse than
00:37:36.480 | the crime and the cover-up is a cover-up. They were not letting him take meetings. They were
00:37:41.360 | limiting access to him. Dean B. Phillips came on this program. Shout out to him. Congratulations
00:37:46.560 | on a great run. He just told the truth here. He's not up for it. He's sundowning. Terrible
00:37:54.480 | situation. People get old and people have cognitive decline. The end.
00:37:59.760 | Question. Hold on a second. Just to contrast and compare, Trump is not even in office.
00:38:06.560 | Elon is not a member of the cabinet per se. These are effectively today still private citizens.
00:38:13.520 | There's all of this noise about what happened over the last two days to stop an absolutely
00:38:17.440 | ridiculous pork barrel bill. When are we going to double-click into what decisions have been
00:38:22.560 | made in the last two years that were actually Biden's versus surrogates that just decided,
00:38:27.680 | and who gave them the authority to make those decisions?
00:38:30.320 | If they're going to do, Aaron, and you and I are left-leaning moderates, I think that would
00:38:34.640 | be the most accurate way to describe us. I mean, if they want to do an investigation,
00:38:39.360 | there should be an investigation, too. Did they know that he was in massive cognitive decline and
00:38:44.240 | let him stay in charge of the nuclear codes? Do you agree or disagree?
00:38:48.720 | This is sort of not the part of politics I think about as much, so I'll leave that up to you as
00:38:55.920 | the other left-leaning moderate. No, I just wonder if this is a crime that they... What if it turns
00:39:02.400 | out he actually has Parkinson's or Alzheimer's, like a diagnosis? It's not out of the realm of
00:39:08.800 | possibility that they covered up an Alzheimer's diagnosis, Chamath, and if they did, is that
00:39:14.320 | criminal? I mean, it's unethical. It's deeply unethical. It's very dangerous.
00:39:21.440 | Yes. Again, it just goes back to
00:39:24.400 | the people elected Joe Biden. He won fair and square. He ran on a specific mandate that the
00:39:31.600 | people endorse. I just think it's devious, though, that certain figures in that White House took a
00:39:38.800 | level of power and decision-making authority that they were never entitled to. If they wanted that,
00:39:45.200 | they should have run and they should have been elected. That's what we all sign up for,
00:39:49.520 | and so I think that idea that we let that happen or that that happened to the American voting public
00:39:55.760 | is, I think, very unfair. That's why I think you have to realize that, and you've said this before,
00:40:03.120 | we need these checks and balances going forward. I think the way that you have these checks and
00:40:07.760 | balances, again, is veer towards transparency. The more transparency there is, and again,
00:40:13.280 | this is where I'll say you may or may not like Donald Trump, but the one thing you will never
00:40:19.520 | have to be worried about is whether you will not be able to hear from him in first person.
00:40:26.240 | You're going to hear from him. That is absolutely true. I mean,
00:40:29.440 | simple suggestion here, Chamath.
00:40:31.120 | If you're ever in doubt, you will be able to know very quickly where he stands
00:40:36.320 | on whatever topic is important to you, and I think that that idea is very important,
00:40:41.280 | because then if it's filtered through somebody else because of some health issue that's then
00:40:45.760 | covered up, we're making decisions that impact the entire world. We're making decisions that
00:40:51.440 | impacts the economy. We're making decisions that touch hundreds of millions of American lives.
00:40:56.640 | Who are making those decisions?
00:40:58.320 | Yeah, it's kind of crazy. I have a simple suggestion here with these bills, by the way,
00:41:02.000 | when they're 1,500 pages. How about for every 100 pages you release, you have one day of review?
00:41:07.440 | So if you want to release 1,500 pages, 15 business days review,
00:41:10.800 | does that sound reasonable? Maybe three weeks, and then just stop doing 1,500 at a time.
00:41:15.920 | Break these things up into two or 300 pages at a time. I just love the fact that people
00:41:21.280 | are motivated and they have the will and the desire to focus on these focus and desire,
00:41:28.400 | because people have to have the will and people did not have the will to get in the weeds
00:41:33.040 | and examine the spending. And now it's becoming like in vogue. It's becoming a
00:41:38.400 | pastime to question the spending. This is a really great moment in time for America.
00:41:42.480 | The other thing, Jason, that we haven't done is I think that killing the bill was step one.
00:41:47.200 | The thing that America has not yet seen, and I think Aaron brought this up with the Patrick
00:41:52.800 | Collison tweet, which is just excellent. Nick, if you just want to put it up there, it really is
00:41:56.960 | true. America still believes that the more you spend, the more you get. We do at the core.
00:42:05.520 | That's why there are 1,500 pages of spending in here, because people want things because
00:42:10.000 | they want things to be better. What we need to train people to understand is actually that it's
00:42:16.560 | the lies that are told that make you think that with more money comes a better outcome.
00:42:21.520 | And we see it every day in Silicon Valley. We all start with next to nothing as a startup,
00:42:27.840 | and we outmaneuver and we out execute companies all day long with way more resources. So it has
00:42:33.040 | nothing to do with the resources. >>Contraint leads to great art. Aaron,
00:42:36.960 | your thoughts? >>No, I can't add that much more to this. I think there's probably a little bit of a
00:42:42.400 | disconnected times from, let's say, the voting public and broad constituents from then those
00:42:49.440 | that have sort of seen this in real life being inside of a company having to do a startup and
00:42:54.480 | scaling up and just the perverse incentives to build bigger teams, spend more. Your project
00:43:02.720 | then is more important, the more dollars it gets. We have all of these systems in place,
00:43:07.280 | which is the stuff that gets attention are the things that you spent more on.
00:43:10.880 | So you have all these weird incentives to actually have your thing literally cost more,
00:43:15.680 | to have more overhead because you've brought in more contractors into the project that then
00:43:21.520 | you're going to get some kind of future benefit from in some way. So you have a lot that is sort
00:43:27.520 | of fully broken in this. And it's hard to imagine any other way to veer off from that path other
00:43:35.520 | than something that does shake things up as much as Doge is doing. >>All right, listen, we can't
00:43:40.880 | do any worse than being massively in debt. So just let's have a culture of- >>Yes, we can. Yes,
00:43:45.840 | we can. You could have runaway debt. >>I think we already have that. I mean,
00:43:50.000 | for people who maybe are rooting against Trump in the audience or rooting against this because
00:43:56.240 | they're super partisan, all I'll say is I know these individuals around Trump, root for them
00:44:01.440 | and root for this process, please, because this is a path to fixing the most acute existential
00:44:07.680 | problem we have, which is our debt. You don't have to like Trump to like Elon, to like Vivek,
00:44:12.880 | and to like these other individuals, Sachs included. There are great people who are being
00:44:17.760 | called to serve. Let's judge them based on their performance. That's all I ask for the people who
00:44:24.080 | hate Trump, who hate these individuals, judge them on their performance. They've come out with a bold
00:44:29.680 | plan. Let them cook. Once they've cooked, then judge their results. That is what I'm telling
00:44:36.000 | everybody who hates Trump and who hates this administration and who's partisan on the other
00:44:40.720 | side. Let them cook, judge the results. >>I'll just throw out one more thing on this,
00:44:44.480 | because I think Doge has, the branding of Doge is often the efficiency side, which people always go
00:44:50.960 | to the spend side. But the corollary to that is the regulations that obviously are expensive to
00:44:57.280 | maintain. That's what creates layers and layers of overhead on reviewing everything that's coming in
00:45:03.040 | then to the government. Unfortunately, we have great examples in California where literally,
00:45:10.480 | we spend more to do less. It's because we've ratcheted up these layers and layers of regulation.
00:45:16.480 | I have friends literally doing climate tech. In climate tech, you couldn't imagine something more
00:45:23.040 | probably left-leaning Democrat that they can't actually get things done in California, the state
00:45:28.240 | that you would imagine to be the most climate-first friendly state, because of the amount of
00:45:32.800 | regulation that prevents them from getting things done. There's actually this total combination of
00:45:40.560 | actually fewer regulations. You'll spend less money in government. You'll actually grow the
00:45:45.360 | economy faster, which will create more jobs. All of the things get solved, the more efficient you
00:45:51.280 | get, kind of writ large on all of the topics. >>Absolutely. So this is spending. Great point,
00:45:56.400 | Aaron. Regulations next. Let's see what they do there. I was in the room when Antonio Gracias
00:46:02.240 | was doing zero-based budgeting at Twitter, now X, and Sax and I were looking at roles for people.
00:46:10.960 | What this team can do in terms of making things more efficient, it's amazing what can happen when
00:46:16.880 | you do zero-based budgeting and when you just think from first principles, well, do we even
00:46:21.360 | need to do this? Does this need to exist? Take all these regulations, put a 20-year clock on
00:46:27.520 | half of them, a 10-year clock on the other half. >>Can I just get one more random example, feel
00:46:31.280 | free to edit it out. Chamath, you'll like this because it came out of Meta. Have you followed
00:46:35.680 | the Z-Standard compression library from Meta? So this open-source library, kind of a next-gen
00:46:42.800 | compression on data. It took us probably too long, but the team worked insanely hard,
00:46:47.840 | implemented this compression algorithm. Our uploads and downloads got faster.
00:46:51.840 | We spend less money on networking and compute. It took some re-engineering of the system,
00:46:58.800 | but that's just not a concept that people go into problem-solving with, which is like,
00:47:02.960 | what if the thing actually was cheaper to run and it was better? And so think about all the systems
00:47:08.240 | of government that could just be upgraded, and then you would spend less money actually
00:47:12.400 | maintaining them. I mean, we spend hundreds of billions of dollars, way too much, on legacy
00:47:17.360 | infrastructure, technology. We could automate more. You could spend way less money and then
00:47:20.960 | get better outcomes. So this is just happening everywhere, and I don't think people realize
00:47:25.520 | the scale of the opportunity. >>And I'll just give an example. When we went into Twitter,
00:47:30.560 | nobody was coming to the office. There were people who hadn't come to the office, who hadn't
00:47:36.080 | committed code in six months. So what were they doing, right? So you start looking at the data.
00:47:40.640 | Then they were spending an enormous amount of money on SaaS software that nobody had logged into.
00:47:45.120 | And then they had desk software for route people around that was costing $10 per day, per desk,
00:47:51.920 | per location, whatever. Nobody was coming to the office, but they were paying for software to route
00:47:58.800 | people to the right desk in office suiting. The waste, when I tell you the waste and the grift,
00:48:04.800 | and I'll just call it what my interpretation is, stealing. They were stealing from those
00:48:09.360 | shareholders. The government is stealing from taxpayers. It has to be fixed. Let our boys cook.
00:48:15.760 | Freeberg, you get the final word before we go on to Conspiracy Corner. You got nothing? All right,
00:48:21.200 | let's go straight to Conspiracy Corner. Everybody put your tinfoil hats on. There were drones over
00:48:25.920 | New Jersey. Thursday morning, the FAA banned drones. This is breaking news in parts of New
00:48:32.080 | Jersey until January 17th due to "breaking news, special security reasons." They also warned that
00:48:39.760 | the government may respond with deadly force against drones posing a threat. This thing is
00:48:45.200 | getting crazier by the day. There have been thousands of reported drone sightings in New
00:48:48.880 | Jersey and bordering states over the last week. Here's some examples. Play the clip, yada, yada.
00:48:55.200 | Until Thursday morning, White House officials have been dismissive, saying repeatedly that
00:48:59.600 | nothing significant is going on. One of the theories was that the drones were looking for
00:49:05.280 | nuclear material, a.k.a. a dirty bomb or lost radioactive material or, the ultimate, a lost
00:49:13.520 | nuclear bomb from, like an actual nuclear warhead from Ukraine. On Tuesday morning,
00:49:20.640 | the mayor of Belleville, New Jersey suggested the drones could be looking for that radioactive
00:49:24.480 | material that went missing on December 2nd. That was radioactive material, germanium-68. We'll get
00:49:30.640 | details on that from Freeberg in a moment. And then, of course, conspiracy theories are going
00:49:36.240 | wild. It's Iran, it's UFOs, and my favorite, Project Blue Beam, which is that NASA is using
00:49:42.320 | holograms and other technology to create a new world order and religion and projecting Jesus
00:49:47.200 | into the clouds. You can look that stuff up, or we'll have Alex Jones in Sachs' seat one week.
00:49:53.360 | Okay, Freeberg, you're the genius here. Tell us what's going on.
00:49:56.320 | I think there are three likely explanations. The first is that the government's got some
00:50:01.440 | activities that can't be disclosed, so we don't know and they can't talk about it. No one can
00:50:05.280 | neither, you know, say yes or say no to it. So, you know, that's kind of one that's pretty
00:50:09.760 | possible. The second is, these are just individuals with a bunch of drones messing around, having fun,
00:50:14.960 | trying to wreak havoc. Could be fun. Bunch of kids. I think we've all been there. The third
00:50:20.160 | is what I would call kind of a bit of a more nefarious, like, this is my conspiracy theory.
00:50:24.960 | I think it could be considered a PSYOP. Okay, so right now, the US has a significant regulatory
00:50:35.840 | burden on drone utilization in a commercial setting. And it's very hard to use drones,
00:50:41.680 | you have to have line of sight to the operator, these things aren't supposed to go on their own,
00:50:44.720 | there's all these rules and restrictions and so on and so forth. Meanwhile, you've got countries
00:50:49.520 | like China rocketing ahead. So I don't know if you guys know the company Meituan in China,
00:50:53.440 | you know, the food delivery company? If you guys, do you know that they do a large amount of drone
00:50:58.320 | delivery of their food now? I was not aware of that. We're testing that here in the US.
00:51:02.240 | The drone delivery business in China is already $30 billion a year. And they're
00:51:07.760 | also launching a pretty significant fleet of what we would call kind of the eVTOLs or flying cars.
00:51:12.480 | The expectation is that by 2030, there'll be 100,000 flying cars,
00:51:18.160 | moving people around in China. And these are huge efficiency gains. In fact, with Meituan,
00:51:22.160 | you can now order food while you're on the Great Wall at one of the ramparts. And it'll bring the
00:51:26.080 | drone will bring the food to you while you're walking the Great Wall as a tourist. And in the
00:51:31.200 | US, the reason that these things haven't taken off, and the reason we don't have a large kind
00:51:35.040 | of drone industry, which is clearly emerging, it's going to be a huge economic driver for China and
00:51:39.920 | others around the world is simply the regulatory restrictions. And so if you were going to try and
00:51:45.280 | mess with the US's ability to move forward with the drone economy, you would probably try and
00:51:52.320 | wreak some havoc, stoke some fear, and get people to say, "Hey, this doesn't seem cool. What's going
00:51:58.480 | on? I don't like that there's all these drones in the sky. I'm freaking out." And try and get
00:52:02.000 | the regulators to come in and say, "Hey, we're banning drones." And set up everyone, including
00:52:08.400 | the people in the government to say, "We should take a beat. We should think a little bit before
00:52:12.880 | we deregulate." We should... Who would do this? Who's the motivation? Uber Eats and Postmates?
00:52:17.600 | No, no, no. People driving? No, no, no. It could be the other government. That's my psyop point.
00:52:22.960 | This could be a psyop. Oh, you're saying this could be China doing this to try to slow down
00:52:26.000 | our economy? Think about it. If you're going to pass a bill in Congress to make drones more
00:52:31.120 | freely roamable in the skies, you're going to reference back this crazy story in New Jersey.
00:52:35.760 | Everyone's freaking out about it. And you're going to say, "Hey, wait, what about all that stuff that
00:52:39.440 | happened in New Jersey? Maybe this doesn't make as much... People are scared about it. We shouldn't
00:52:43.200 | rush. We shouldn't rush. We shouldn't rush." That would be my psyop theory. That's my conspiracy
00:52:49.360 | theory tinfoil hat. I don't often do them, but that's what I would think about. I think the
00:52:52.720 | first two are probably more likely. Alex Jones would be proud. Aaron, why don't you jump the
00:52:56.560 | fence and tell us your best conspiracy theory in this? Jump the fence. No, no, no. That was
00:53:01.520 | all crazy pills that I just heard. Yeah, take some. It's entertaining. I think there's 10
00:53:07.280 | higher psyops you would do if you wanted to get us to have a collapsing economy than going after
00:53:12.320 | drone deliveries. What would they be? First of all, I think you'd have AI do a robot. I think
00:53:21.920 | you'd have a robot AI thing that runs amok. That's a good idea. Rogue robot. Robocop. Yeah.
00:53:28.080 | I think that would be way sooner than you worry about food delivery. No, you have 10 self-driving
00:53:31.840 | cars hop the curb and crash into a storefront. Self-driving is on ice for two years. That would
00:53:38.480 | be an example. Chamath, your thoughts? You got some conspiracies here. What do you think's going
00:53:43.440 | on here? No, but I thought that the most credible thing was that they were looking for radioactive
00:53:52.320 | material, that somehow some got lost. Why would they only look at night? Actually, it's interesting
00:53:59.440 | you say that. There was somebody on Twitter, X, who claims to be an expert in this field,
00:54:05.120 | and there's a startup actually that I just talked about. Which one? Canacoa the Great?
00:54:08.880 | I think it might've been Canacoa the Great. Shout out to Canacoa the Great in his mom's
00:54:11.600 | basement eating pizza bites. Why do you hate Canacoa? He's a nice guy. I don't hate Canacoa.
00:54:15.520 | I don't hate Canacoa. Canacoa tried to get me bad because he said I was trying to dox him.
00:54:20.160 | You always can't Canacoa. He's great. No, I just think it's hilarious that people are
00:54:24.000 | retrading Canacoa the Great as if he's this great journalist, and he's in his mom's basement eating
00:54:30.480 | Hot Pockets, or he's working for Putin. He's a really good-
00:54:33.680 | I don't know. Do I want my news from Jake? It's J. Cal or Canacoa. I don't know.
00:54:38.000 | I don't know. Who do you take? I mean, your conspiracy theory is pretty good.
00:54:41.280 | I subscribe to Canacoa. Aaron Levy, you have a favorite? Who are you? You Autism Capital or
00:54:47.760 | Canacoa? Who are you? I like Autism Capital, too. I think he's really good.
00:54:50.480 | Absolutely. I'm liking Geiger these days.
00:54:53.280 | Oh, you're into Geiger? Geiger Capital is good. Geiger Capital is good.
00:54:56.080 | I read them all. I read them all.
00:54:57.200 | VC Braggs? VC Braggs?
00:54:59.040 | I think the nighttime thing would be, it would be at least typical of this government to do
00:55:02.640 | a Streisand effect of just like, maybe if we cover it up, nobody will see, and then
00:55:07.920 | obviously it's the biggest thing, so. Yeah, and they have blinking lights on them.
00:55:11.040 | Yeah, they put blinking lights on them.
00:55:13.520 | But if they, there is a startup that makes, we really looked it up, and there is a startup
00:55:20.560 | that makes drones to do this specific task, to look for dirty bombs in ports, and ports
00:55:25.840 | obviously look for these things. It's a known threat. This is not rocket science.
00:55:30.080 | I'm sorry, and why, and why only at night, to free folks' questions?
00:55:33.040 | Oh, because that they can read the signatures better was the theory, that at night you can
00:55:37.680 | read the signature, which doesn't make sense to me. Science guy, you want to come in here?
00:55:41.600 | I don't see how night, I don't see how nighttime you'd get a better read on radioactive material.
00:55:47.840 | It doesn't make any sense, so.
00:55:49.040 | That made no sense. That sounds like a crazy thing. But anyway, we're living in crazy times.
00:55:52.960 | Speaking of crazy.
00:55:54.960 | By the way, that was my first conspiracy theory in 208 episodes of the All in Pod, so.
00:55:58.960 | Very nice. I actually can participate in Conspiracy Corner now.
00:56:03.120 | I'm coming back next week with a better one.
00:56:06.000 | Okay, Jake, I'll read us some more news. Go to Google News and read us some more news,
00:56:10.640 | so we can do another topic.
00:56:11.360 | Jesus Christ, man.
00:56:12.080 | Read a f***ing article for us. Go.
00:56:14.080 | No, you do it. Go ahead. No, no, you get the docket in front of you.
00:56:16.400 | Okay, so today the Dow Jones Industrial Average is up 23,000 to 44,000.
00:56:21.680 | I try to highlight you guys, make you look good.
00:56:24.320 | Gary Gensler said he's super happy.
00:56:25.760 | I'm going skiing. That's enough. I'm going skiing.
00:56:28.080 | I thought it was cool how he read the entire congressional bill earlier.
00:56:32.720 | Yeah, all 1,500.
00:56:33.840 | J. Cal loves to filibuster.
00:56:36.160 | I'm not filibustering. All right, listen. Here, let's do another story. Let's see if
00:56:39.600 | you can contribute something. Open AI update, Mattis Lander. Your contribution was amazing.
00:56:45.840 | I was actually using chat GPT to go into the founding father's papers.
00:56:51.440 | I was reading the Federalist Papers with Gemini.
00:56:55.600 | 13 and number 44.
00:56:57.200 | I'd like to quote from Hamilton the Musical.
00:57:01.680 | I was comparing the lyrics from Hamilton the Musical.
00:57:08.880 | Oh my God.
00:57:10.800 | Hey, Nick, can you send J. Cal another Yahoo News article? Let's get going.
00:57:13.760 | Come on.
00:57:14.960 | I can't wait. I can't wait to see you in the morning. Let's go.
00:57:24.640 | Can we just end it here?
00:57:28.400 | No, do it, do it, because I want to talk about OpenAI.
00:57:32.400 | You want to talk about OpenAI? You do?
00:57:33.760 | Yeah.
00:57:34.160 | Give him his red meat. Give Chamath his red meat.
00:57:36.080 | All right, here's an update on closed AI, flipping,
00:57:39.360 | and Sam Altman, supervillain Sam Altman, secure in the bank.
00:57:43.840 | Meta wrote a letter to California's Attorney General-
00:57:46.640 | That was your best open yet.
00:57:47.840 | -on OpenAI's for-profit conversion two months ago.
00:57:52.960 | OpenAI announced a $6.6 billion funding round, $157 billion valuation.
00:57:59.360 | They're projecting $3.7 billion in revenue this year, pretty great revenue growth.
00:58:04.400 | But there's a poison pill in the deal.
00:58:07.440 | OpenAI must convert to a for-profit within the next two years,
00:58:11.920 | or investors can ask for their money back.
00:58:15.600 | And right before they announced this round, you remember CTO Mira
00:58:18.960 | Mirati and two other top researchers resigned. Many people saw this as a protest.
00:58:24.640 | Elon, who put the first $50 million in and co-founded OpenAI, is currently suing the company
00:58:29.440 | and seeking a court order that would stop the for-profit conversion.
00:58:33.840 | Now, Zuck is joining Team Elon.
00:58:37.680 | Elon and Zuck are in some weird pairing up. They're not in the ring, not in the octagon, no.
00:58:43.120 | Last week, Meta sent a letter asking
00:58:46.320 | California's Attorney General to stop OpenAI's for-profit conversion.
00:58:50.400 | What do you think of this?
00:58:51.360 | We've got them now not in the octagon, but they're in the political arena.
00:58:55.840 | Chamath, your thoughts?
00:58:57.120 | I think that this is so interesting.
00:58:58.560 | So I was looking at all of these things.
00:59:00.880 | If you put them all together, it paints a really interesting story.
00:59:04.320 | So you have Elon filing an injunction, which basically says you should not allow this
00:59:10.400 | conversion to happen until we sort out all of these details, because if you allow it to convert
00:59:16.320 | and then I win, it'll be very messy to go backwards.
00:59:19.920 | I think that that's a pretty credible legal argument.
00:59:22.320 | Then you have Zuck basically say, "Hey, Elon's right. This company should not convert."
00:59:30.080 | But the more interesting thing that really got me thinking about this was a chart that
00:59:34.080 | Menlo Ventures put out.
00:59:36.240 | Nick, can you just show this?
00:59:37.360 | What this shows is just what's happened in the last year.
00:59:41.920 | What do you notice?
00:59:43.680 | What you notice is the market share of OpenAI has changed pretty meaningfully
00:59:50.000 | from half to about a third.
00:59:53.360 | What you see is Anthropic doubled, Meta roughly staying the same, Google picking up steam.
00:59:59.440 | It started to make me think this is very similar to a chart that I would have looked at in 2006
01:00:09.280 | and 2007 when we were building Facebook, because we had this huge competitor in MySpace that was
01:00:15.040 | the de facto winner and we were this upstart.
01:00:18.640 | It made me think, "Is there a replay of this same thing all over again?"
01:00:23.840 | Where you have this incumbent that pioneers an idea and they start with 100% share.
01:00:31.440 | Then all of these upstarts come around from nowhere.
01:00:35.040 | Then I thought, "Well, what is better today if you are a company that's just starting
01:00:41.120 | versus if you were the incumbent?"
01:00:43.120 | I think that there's a handful that are worth talking about.
01:00:45.600 | The first is when you look at what XAI has done with respect to NVIDIA GPUs,
01:00:52.960 | the fact that they were able to get 100,000 to work in one contiguous system and are now
01:00:59.920 | rapidly scaling up to basically a million over the next year.
01:01:02.880 | I think what it does, Jason, it puts you to the front of the line for all hardware.
01:01:07.360 | Now all of a sudden, if you were third or fourth in line, XAI is now first and it pushes
01:01:12.560 | everybody else down.
01:01:13.440 | In doing that, you either have to buy it yourself or work with your partner.
01:01:18.000 | I think for the folks like Meta, that translates and explains why they're spending so much
01:01:23.680 | more.
01:01:23.840 | It's sort of like this arms race.
01:01:25.600 | If you can't spend with my competitors, I'm just going to prefer my competitor to you.
01:01:29.600 | I think that that creates a capital war.
01:01:33.280 | In a capital war, I think the big companies like Google, Amazon, Microsoft, Meta, and
01:01:40.480 | brands like Elon will always be able to attract effectively infinite capital.
01:01:45.760 | Their cost of capital goes to zero, which means they'll be able to win this hardware
01:01:50.480 | war.
01:01:50.960 | Okay, so put a pin in that.
01:01:52.160 | Then the second thing is what happens on the actual data and experience side on the training
01:01:57.600 | side.
01:01:57.840 | If you listen to Ilya Tsutskever, if you listen to Andrej Karpathy, what they effectively
01:02:02.800 | are saying is there's this terminal asymptote that we're seeing right now in model quality.
01:02:08.880 | What happens?
01:02:09.440 | A lot of these folks are now experimenting on the things around the model, the user experience,
01:02:14.800 | how you use it.
01:02:15.520 | Can I keep things in memory?
01:02:16.640 | Can I cut and paste these things from here and there?
01:02:18.640 | Because what it says is the data is kind of static and brittle, but it's actually not.
01:02:24.960 | That's what we said before, because you have this corpus of data on X that's pretty unique.
01:02:29.360 | I suppose if Elon fed in all this kinetic data that he controls through Tesla, that's
01:02:36.000 | very unique.
01:02:36.720 | Does that all of a sudden create this additive pool of information?
01:02:41.120 | Possibly.
01:02:43.920 | Then the third thing is when you look at that chart, what that chart says is, "Hold on a
01:02:47.440 | second.
01:02:47.680 | Why are corporates moving around?"
01:02:50.480 | What I can tell you just through the 80-90 lens is we are completely promiscuous in how
01:02:56.880 | we use models.
01:02:57.760 | The reason is because these models offer different cost quality trade-offs at different points
01:03:05.200 | in time for different tasks.
01:03:06.560 | What we are seeing is a world where instead of having two or three models you rely on,
01:03:12.000 | you're going to rely on 30 or 40 or 50, and you're going to trade them off, and you're
01:03:15.520 | going to use effectively like an LLM router to-
01:03:18.320 | Load balance them, yeah, and to route them.
01:03:22.000 | Or just to manage and route them.
01:03:23.440 | Then there's an intelligence above that that's constantly tasking and figuring out prompt
01:03:27.280 | optimization across these models.
01:03:28.800 | It's this thing where we were very reliant on open AI.
01:03:32.800 | Now we're reliant across three or four.
01:03:36.480 | Ideally, we'll be reliant on 30 or 40.
01:03:40.400 | I just see this world where it's all getting commoditized quite quickly.
01:03:43.600 | I'm just sort of scratching my head.
01:03:46.800 | Where is the market value here?
01:03:48.880 | Aaron, your thoughts?
01:03:49.840 | I know you're very promiscuous when it comes to LLMs as well.
01:03:52.240 | We have a very similar model, which Ma said, which is we're agnostic, so we work with multiple
01:03:58.800 | AI vendors.
01:03:59.360 | I think a friend deep in AI land a couple of years ago, right before Chatterbot said,
01:04:07.200 | "There's no secrets in AI," and I didn't totally understand at the time.
01:04:10.800 | It hadn't registered what that meant, but very quickly it became obvious, which is the
01:04:14.880 | research breakthroughs propagate insanely quickly across the AI community.
01:04:21.600 | Back to this MOS framework, if you just think about it, if the research effectively becomes
01:04:28.080 | open at some point in time quickly enough because either the researchers move or people
01:04:31.920 | publish it or whatnot, then it really is a compute game and then maybe a data access
01:04:36.160 | game, and that means that there's four or five at scale players that can fund this.
01:04:40.080 | I think as we've seen in other areas where it's an infrastructure play, with enough competition,
01:04:50.080 | you have the underlying service eventually trend toward the cost of the infrastructure.
01:04:53.440 | What we should expect is that the price of a token in AI land basically will be whatever
01:05:01.120 | the price of running the computers are, and maybe with a plus 10%, 20% margin.
01:05:07.040 | Did you see the same thing happen with storage?
01:05:08.800 | I remember in the early days with Box and Dropbox and YouTube, you all had this major
01:05:13.920 | innovation with storage.
01:05:15.600 | How did that play out?
01:05:16.480 | Let me give you a fun stat.
01:05:17.520 | We give our customers unlimited storage.
01:05:19.360 | We have 82% gross margin.
01:05:21.760 | What happened was the price of the underlying storage has gone down by hundreds of times
01:05:29.520 | since we started the company, and then all our value is in the software layer on top
01:05:33.840 | of the storage.
01:05:34.960 | We've benefited by this incredible, ruthless competition between Western Digital, Seagate,
01:05:40.480 | other players that are just trying to pack more storage density into these drives.
01:05:48.080 | Every couple of years, they have a new breakthrough.
01:05:49.600 | We're heading toward maybe a 50-terabyte hard drive.
01:05:53.680 | When we started the company, they were 80 gigabytes.
01:05:56.480 | How much of your time in the early days was spent on dealing with this infrastructure
01:06:01.680 | issue, and then how much of your time and your leadership team's time is spent on this
01:06:05.920 | issue now?
01:06:06.420 | Back in the day, if we had 10 people in engineering, 80% of them was doing pure infrastructure
01:06:13.760 | work.
01:06:14.240 | If we have 1,000 people, it would be inverted in terms of the ratio.
01:06:18.960 | You get more leverage, both as you get the advancements in the technology, but then also
01:06:25.200 | as you get scaled.
01:06:26.320 | But all of this is to say, you should basically anticipate a world where -- and I think Zuck
01:06:31.200 | is this interesting counterbalance on all of this because of open source -- if at any
01:06:35.200 | moment you know that Zuck will basically provide an open-source model that is at best-in-class
01:06:41.520 | benchmarks and at the frontier, then there is a limit on how much you can charge for
01:06:47.040 | the tokens of your hosted model because anybody will then be able to go host the open model
01:06:51.200 | and be able to provide infrastructure around it.
01:06:53.440 | So if you always have that counterbalance, and the tokens eventually kind of look the
01:06:57.680 | same, the output tokens kind of look the same...
01:06:59.920 | Isn't it also the truth that major enterprises -- Fortune 500s, 200s -- 20 years ago weren't
01:07:08.960 | interested in open source, and now that's kind of their default?
01:07:11.520 | They want to buy into open source because they don't want to be locked into a vendor?
01:07:15.120 | It's actually not even necessarily the case that the customer has to pick the open-source
01:07:18.880 | vendor.
01:07:19.200 | They might buy it through an abstraction layer that is letting them get the benefit
01:07:22.960 | of open source, but still buy through a proprietary or commercial...
01:07:26.080 | So you believe open source wins?
01:07:27.600 | I believe...
01:07:28.960 | In AI?
01:07:29.520 | I believe open source causes pricing to always be extremely low.
01:07:34.400 | Right, but in this case, do you think open source is going to ultimately win the day
01:07:40.000 | in models?
01:07:40.880 | No, not necessarily.
01:07:41.520 | You don't?
01:07:42.000 | No, because only meta has the scale to be able to provide...
01:07:46.640 | I think what Aaron is saying here -- let me maybe try to frame it -- I think what he's
01:07:49.760 | saying is there'll be open-source models, there'll be closed-source models, but the
01:07:55.520 | price that Aaron or me or anybody else pays these model-makers will effectively go to
01:08:00.720 | zero.
01:08:01.520 | Got it.
01:08:02.400 | And it'll go to the cost of the compute, to be clear, with a little bit of margin for
01:08:06.800 | the work of...
01:08:07.040 | The energy, the physicality of that compute.
01:08:09.520 | Got it.
01:08:10.080 | Now, it's important to step back and say, "I still think you could probably have the
01:08:13.280 | entirety of the model providers make 10 times more revenue than they do today because we're
01:08:17.440 | just literally in the first percent of the total TAM."
01:08:20.640 | So, it'd be a mistake to think that that has some kind of downward pressure in terms of
01:08:25.840 | the long-term economics of these businesses, especially because I think OpenAI's revenue
01:08:29.920 | stream is increasingly looking like a SaaS revenue business as opposed to just the API
01:08:35.200 | kind of token pricing.
01:08:36.160 | So, none of this is to provide any sort of color on what would you bet on today.
01:08:42.000 | I agree with Chamath that you're going to have maybe not 30 providers, but let's say
01:08:44.800 | you at least have five to 10 good choices all competing heavily for the next breakthrough.
01:08:48.960 | Like literally this morning, Google had a breakthrough in sort of this reasoning-oriented
01:08:53.600 | model from their Gemini family.
01:08:55.760 | They're 2.0, yeah.
01:08:57.200 | Yeah.
01:08:57.840 | And so, what's incredibly kind of great is it's sort of the best time ever to be building
01:09:04.240 | software, assuming that you have a play in the market that lets you remain differentiated.
01:09:10.320 | And the key there is just do enough on top of the AI model that there's enough value
01:09:14.320 | there.
01:09:14.560 | You know how much the world spends on software and software-related things every year?
01:09:18.720 | It's about $5 trillion.
01:09:20.800 | So, there's like, call it like a trillion and a half of software licenses, a trillion
01:09:27.920 | and a half of consulting, and a trillion and a half of IT folks inside of companies, plus
01:09:32.800 | or minus a little bit more, you get about $5 trillion.
01:09:34.800 | And that's compounding 13% a year.
01:09:38.880 | I'm pretty sure that the market here shrinks by an order of magnitude.
01:09:44.800 | And instead of fighting over $5 trillion, I think we'll be fighting over $500 billion.
01:09:49.920 | What do you think, Aaron?
01:09:51.840 | You buy that?
01:09:52.880 | No, not at all.
01:09:53.520 | I mean, I don't know if you want to make the case more, but...
01:09:57.680 | Well, the only reason is that I think that as we de-lever the software development process
01:10:04.080 | from humans, I think the unit cost of creating code effectively becomes so cheap that it
01:10:08.800 | is going to be very hard to differentially price these products the way that they are.
01:10:12.800 | So, an example would be that, let's say you use, I don't know, pick your favorite piece
01:10:19.520 | of software.
01:10:20.160 | I don't want to pick on anyone.
01:10:22.960 | That's why I'm not saying anything.
01:10:24.400 | I just say an office suite.
01:10:25.600 | Let's pick an office suite.
01:10:26.800 | Everybody's got one.
01:10:27.600 | Sure.
01:10:28.820 | Let's pick on Excel because maybe that's like...
01:10:31.520 | Sure.
01:10:31.920 | Excel, Google Sheets.
01:10:33.200 | Yeah.
01:10:33.840 | It's not that Excel isn't valuable.
01:10:35.360 | It's incredibly valuable.
01:10:36.640 | It's what is the marginal cost of creating the Excel equivalent that is good enough that
01:10:40.560 | people switch.
01:10:41.520 | The marginal cost today is very expensive.
01:10:44.080 | And you can see that because it's what it costs Google to make Sheets, but that's humans.
01:10:50.800 | So, the real question is, if you have a legion of bots that works 24/7 incrementally and
01:10:58.560 | increasingly more accurately every day, the question is, what is the marginal cost?
01:11:04.240 | And I think the marginal cost of that is going to be very cheap.
01:11:07.120 | And when you do that, it's very difficult to price it anywhere near the same.
01:11:12.480 | And the reason is that other companies will then replicate it and say, "Hold on, if Excel
01:11:18.080 | wants to charge $100, I'll charge 50 and I'll take a lower margin."
01:11:21.840 | That's just supply-demand economics.
01:11:24.240 | Yeah.
01:11:24.740 | So, I just don't agree with the TAM compression because I think there's another kind of counter
01:11:33.200 | event that's happening that AI is really going after services.
01:11:36.880 | And so, that then conversely expands the TAM software where IT budgets weren't usually
01:11:42.640 | applied to those types of things.
01:11:43.680 | But we can get into that in a second.
01:11:44.880 | But I think we've already seen this, though.
01:11:47.760 | And it hasn't exactly played out as you're saying.
01:11:50.000 | So, Zoho is this really interesting business.
01:11:54.240 | It's probably a couple billion in revenue at this point.
01:11:55.840 | And it's basically a suite of extremely low-cost, affordable software products by category.
01:12:02.240 | The cycle time of Zoho is poor.
01:12:04.480 | But that's not been the reason people don't switch, though.
01:12:07.600 | No, it's all levered to humans.
01:12:09.680 | I just think it's not a good product.
01:12:11.440 | It's decent enough.
01:12:12.880 | Why do you think people don't switch, Aaron?
01:12:14.400 | What's your thesis there?
01:12:15.840 | I do agree with you, Aaron, by the way, that when you bring that whole offline services
01:12:20.880 | category online and you automate them with AI, I agree with you that that TAM could be
01:12:25.360 | ginormous.
01:12:26.000 | All I'm saying is the traditional software TAM today, what people spend $5.1 trillion
01:12:30.960 | on, I think people will spend $500 billion on.
01:12:34.400 | And barely if that.
01:12:35.440 | There may be other things that people spend money on that are wrapped in AI.
01:12:41.760 | Yeah.
01:12:42.240 | I guess the counter, and maybe you'd look at an ERP system or a CRM system or something
01:12:47.760 | else, that is sort of--
01:12:50.160 | Those things are totally screwed.
01:12:52.400 | No, but this is my point.
01:12:54.560 | The opposite.
01:12:55.200 | The last thing you want to touch is the system that is powering your supply chain.
01:13:00.000 | The companies I talk to, they're consistently like, "Rip it out.
01:13:03.920 | Get us to a point where we can rip it out."
01:13:05.600 | And the reason is because what they've realized is they'll spend $50 to $100 million a year
01:13:10.160 | for five features.
01:13:11.520 | And they're like, "Just give me these five features as workflows.
01:13:14.640 | Give me a simple CRUD database and just get out of the way."
01:13:17.520 | And it's like the trade-off for that makes a lot of sense because, look, let's face it,
01:13:24.000 | when you have to build one piece of software that has to sell to 50,000 companies, the
01:13:28.560 | reality is that that piece of software is trying to do everything and then some, and
01:13:33.840 | it's trying to solve two or three use cases, plus around five or six common use cases that
01:13:38.720 | are generalized for 50,000 customers.
01:13:41.520 | So you end up with 50,000 features.
01:13:43.120 | You know, it's really interesting because, Aaron, you kind of alluded to there's a TAM
01:13:48.080 | expansion moment there, and I'm seeing this on the front lines.
01:13:51.200 | We run a program, Founder University, I've talked about it here before, where we see
01:13:54.800 | people pitching us their year zero startups, two or three-person teams, and what they're
01:13:59.440 | doing is they're not going after existing legacy software.
01:14:02.480 | They're kind of going after jobs.
01:14:04.080 | They're looking at a position or a job somewhere.
01:14:07.760 | This is an accountant.
01:14:08.960 | We're going to take an accountant.
01:14:10.080 | We're going to make the number one accountant in the world that's an AI agentic, an agent.
01:14:15.520 | We're going to make a podcast producer with podcast AI.
01:14:18.480 | We're going to make a virtual assistant with Athena.
01:14:20.800 | We're just seeing it over and over again.
01:14:22.720 | That's a whole nother category where you study a person's behavior as they work, a social
01:14:28.640 | media manager, and what they do, and then you replicate it with AI, and that's something
01:14:34.320 | that just hasn't previously been done.
01:14:35.680 | So there could be two things occurring here, Chamath and Aaron, at the same time, which
01:14:40.000 | is a deflation in legacy systems, and they'll be replaced.
01:14:44.720 | And then additionally, human capital and jobs that are easy enough for AI to do as an agent
01:14:51.600 | will also expand the TAM, two things at the same time.
01:14:55.200 | You are completely right about the usage.
01:14:58.160 | All I'm debating is when you have to price something, you have to look at the total cost
01:15:04.480 | of what it took to make it, and then you want to try to build in a reasonable amount of
01:15:09.120 | margin and some reasonable expansion, and you discount it back, and this is what you
01:15:12.720 | think it's worth.
01:15:13.280 | Even though you may not think you're doing that when you implicitly price something that
01:15:17.520 | way, that is what's happening underneath the hood to the unemotional buyer of that good.
01:15:22.960 | And all I'm saying is, if what Aaron says before, which I agree with is true, which
01:15:28.880 | is the cost of using a model to get to an output effectively goes to zero, somewhat
01:15:35.440 | what I've been saying before, the marginal cost of compute goes to zero, the marginal
01:15:40.320 | cost of energy goes to zero.
01:15:42.320 | The real question is, what does it take to make a good in the digital world?
01:15:46.320 | >> Let me ask Freeberg a question here.
01:15:47.920 | You're back in the CEO slot at Ohalo.
01:15:50.720 | You're doing it every day, Dave.
01:15:52.320 | You're making these choices as to what SaaS products and how you're going to solve problems.
01:15:56.640 | Are you looking at it saying, "I'm going to hire developers who can work 10x because
01:16:00.720 | of all these new tools, and I'm going to build my own internal systems," or are you looking
01:16:04.960 | and saying, "I'm going to buy off-the-shelf SaaS products"?
01:16:08.480 | What are you doing when you make your own decisions every day, David Freeberg?
01:16:12.160 | >> Well, I try and encourage the teams to build stuff that better meets our needs, and
01:16:20.160 | then it can actually be a better solution, and it can be built in-house.
01:16:25.200 | I think I mentioned this in the last episode, but we did a hackathon where we brought in
01:16:29.440 | people to learn how to use Cursor and ChatGPT to build software that had never done it before
01:16:35.840 | to try and create this as a capability for people broadly in the organization, and there
01:16:40.400 | were great tools that came out of it.
01:16:41.840 | So I think that that's really the future.
01:16:44.400 | And this is kind of like, I'd say, early generation, but as we get to further later generation
01:16:48.480 | capabilities, you could see the instruction to a ChatGPT or like interface, "Hey, I'd
01:16:55.360 | love to do the following things with a piece of software."
01:16:57.680 | It shows you a couple options.
01:16:58.960 | You pick one.
01:16:59.960 | It shows you a couple UX.
01:17:00.960 | You pick one.
01:17:01.960 | It does user testing automatically for you.
01:17:03.120 | It does QA for you, and ultimately, it puts it into production for you, which is the biggest
01:17:07.760 | challenging step right now.
01:17:09.040 | You still need engineers that can load stuff into production and do QA, but if all of that
01:17:12.440 | gets automated as well, now any user in a company can actually stand up software to
01:17:18.120 | do something for them.
01:17:19.120 | In the same way-
01:17:20.120 | A non-developer you're saying for you.
01:17:21.120 | Non-developer.
01:17:22.120 | To stand up software.
01:17:23.120 | Maybe.
01:17:24.120 | So Chamath, just to clarify, there's two different ways to approach the lowering the cost of
01:17:28.160 | developing software.
01:17:29.160 | One is that it just creates more competitors in each of these categories, which then lowers
01:17:33.080 | my price because now there's some downward pressure.
01:17:35.320 | Dave's bringing up a different example, which is I'm going to build my own software at effectively
01:17:39.000 | the price of zero.
01:17:42.080 | The challenge on this, especially the second one, is most organizations don't want to be
01:17:46.600 | in the business of having to think about building their own software.
01:17:49.760 | They just want it done for them.
01:17:51.240 | It's not a core part of their...
01:17:54.960 | Your business would be very unique relative to the broad economy, which is I want a place
01:18:00.080 | to put my CRM records.
01:18:01.280 | I want a place to just have my HR get managed.
01:18:03.520 | I think the downward pricing pressure due to software cost lowering makes total sense.
01:18:08.440 | I don't think it's a 10x factor, but I think that we've always had a long tail of applications
01:18:16.280 | that enterprises build.
01:18:18.360 | You did it at Microsoft Access, now you do it in Retool.
01:18:21.440 | So the next era of that will be obviously AI built and there'll be 10 times the amount
01:18:25.520 | of that software, but it's not obvious to me why that would go after the core systems
01:18:30.880 | of running a business because just most companies are not looking to reinvent the wheel of that.
01:18:34.800 | We're working with an aerospace partner, won't say who it is, and we sat there and they walked
01:18:40.160 | us through what they deal with to make the things that they're making.
01:18:45.320 | It's convoluted and this is not a software problem, it's that there is no piece of software
01:18:50.600 | that understands how they want to run their company.
01:18:53.760 | So instead, they have to morph their org chart to the tools that are available.
01:18:58.720 | So I think what Freeberg is saying is some version of that as well.
01:19:02.400 | There were probably people there using the tools that were available and as a result,
01:19:06.760 | at some point, some HR person said, "We need to hire this other person and this other person,"
01:19:11.440 | and instead, what it allows companies to do is just completely reimagine.
01:19:15.680 | How do you want your company to work and what is the business process you actually need
01:19:20.320 | to implement and then let's just get that built.
01:19:25.680 | I don't know if you guys saw this, but Satya Nadella had this clip that's gone pretty viral
01:19:30.720 | in the last couple of days where he effectively said the same thing and what he's effectively
01:19:34.200 | saying is all these big software systems are business rules wrapping a database plus an
01:19:42.400 | incredible go-to-market team.
01:19:45.960 | This is what Alex Karp points to as sharp knives, steak dinners, and basketball game
01:19:53.240 | tickets.
01:19:54.240 | I don't know if you guys saw that clip.
01:19:55.240 | I did see the clip.
01:19:56.240 | Yeah, he likes to win based on product, not on sales effort.
01:19:59.840 | Nick, you should find this clip.
01:20:01.520 | We don't play golf.
01:20:02.880 | What we do do is we play software.
01:20:04.920 | We will put, if you want to actually compete, compete on your product.
01:20:09.360 | And what's very special, and yes, do I enjoy humiliating people who have better steak dinners
01:20:14.240 | and sharper knives and better golf swings?
01:20:16.480 | Yes, I do.
01:20:17.480 | You know what?
01:20:18.480 | I really like that we win in that way.
01:20:19.840 | It makes me very happy and it makes our clients happy.
01:20:22.960 | But it really points to the heart of how the software industrial complex, that's what I
01:20:27.120 | call it.
01:20:28.120 | How this $5.1 trillion, this software industrial complex, how has it evolved?
01:20:33.560 | I think it's people that build good point products, but when the market, meaning the
01:20:38.440 | shareholders and the public investors, demand growth, you have this natural expansion.
01:20:44.080 | And what do they do?
01:20:45.240 | They inflate the feature set.
01:20:47.680 | They inflate the cost.
01:20:49.520 | And then in order to sell that, they inflate the set of incentives that they give to the
01:20:55.880 | buyer.
01:20:56.880 | The CIO inside of these large organizations, they control budgets.
01:21:01.840 | They're equivalent to like state level budgets in some cases.
01:21:05.560 | You know, what do you think the CIO of a top five bank is spending?
01:21:09.000 | It's probably $15 to $20 billion a year.
01:21:11.960 | They are getting wined and dined to a way that you could not even imagine.
01:21:16.640 | I don't even think the CEO probably understands.
01:21:19.680 | And that filters down through the org.
01:21:21.560 | And so they make these-
01:21:22.560 | Hey, don't like totally ruin our whole game here.
01:21:24.360 | Well, I mean, it's not, it's not, I'm just pointing it out.
01:21:26.960 | Like it's like-
01:21:27.960 | I mean, I'll say the counterbalance to that.
01:21:28.960 | It's the game on the field.
01:21:30.440 | But here's another part of the game in the field.
01:21:32.400 | If these tools make your team 10, 20, 6%, whatever you can quantify, more effective
01:21:39.440 | at their jobs, then it's a de minimis amount for most organizations who have created very
01:21:45.400 | profitable businesses.
01:21:46.720 | I see this where, I was mentioning some companies earlier, and their SaaS pricing is broken
01:21:52.400 | because the number of employees at a company has been going down because people are getting
01:21:56.200 | more efficient.
01:21:57.360 | So now they're looking at saying, well, what value did we provide with this podcast, you
01:22:01.840 | know, producer in a box and whatever it does?
01:22:04.560 | They say they were starting with like $99 a month or $49 a seat.
01:22:08.400 | I said, listen, just charge a minimum of $500 to get the software.
01:22:12.400 | Nobody complained.
01:22:13.800 | And they got rid of the bottom tiers.
01:22:15.640 | So there's a value being created here that is so great that I don't think everybody's
01:22:22.120 | going to roll their own like Friedberg because they're going to be like, you know what?
01:22:25.240 | This is so great.
01:22:26.560 | It's $6,000 a year.
01:22:27.560 | F*ck it.
01:22:28.560 | YOLO.
01:22:29.560 | I'll just buy it.
01:22:30.560 | It is today.
01:22:31.560 | And in five years, just like we're all going to, and we're all using chat-like interfaces
01:22:37.200 | more frequently, you go to your chat-like interface and eventually you realize you can
01:22:42.120 | ask it to build some tool for you that does something.
01:22:45.480 | And it renders the tool and it makes it possible and it stands it up in production.
01:22:50.080 | And suddenly you're using it at your company and then you ask it to do another tool and
01:22:54.080 | it does it exactly to spec and you define the UX you want, you define what you want
01:22:58.400 | it to do and it works and it interoperates, but you and really clearly and really cleanly
01:23:03.920 | with the other tool.
01:23:05.200 | And there's a big aspect of this where you can start to build all of the software infrastructure
01:23:09.760 | that you as an organization want.
01:23:11.480 | Right.
01:23:12.480 | But I think you and Aaron, you and Aaron mentioned this though, the extreme difficulty is not
01:23:17.720 | that front and part.
01:23:18.800 | That's easy.
01:23:19.800 | It's like, I think it's less than three or 4% of the work.
01:23:23.920 | The 96% of the work is how you actually integrate it in the backend and how do you provision
01:23:29.240 | it?
01:23:30.240 | How do you have controls and how do you do security?
01:23:32.200 | Because if those things fail and those are implemented in a bot in a highly regulated
01:23:36.600 | market, as an example, you may not actually be allowed to operate.
01:23:41.280 | Let's go through that example.
01:23:42.280 | And let's say that you're in a bank and you tell the AI, figure out all the security and
01:23:47.520 | permissions and authority rules that are necessary for me to operate as a bank, and it can actually
01:23:52.200 | render it.
01:23:53.200 | It sounds far fetched today, but there's no reason that in seven years that is not the
01:23:56.960 | standard du jour that I don't have the ability to say, go look at all the software that's
01:24:00.640 | out there in the world today.
01:24:01.720 | So that helped me build a tool that meets compliance standards, that meets all of my
01:24:05.160 | security standards.
01:24:06.160 | And you can actually instruct the AI to operate like a large number of software engineers
01:24:10.720 | need to operate today.
01:24:12.520 | The practical issue where the rubber meets the road there is when there is a penetration
01:24:18.000 | and a regulator who's a human, because it will not have been a bot, comes and knocks
01:24:21.920 | on your door and you're like, well, here's this immutable log of the things that I did.
01:24:26.840 | And they're like, I don't want an immutable log.
01:24:29.320 | Who the hell tested this?
01:24:30.600 | Show me the unit test that you created and signed off on.
01:24:33.680 | So there is a critical human in the loop problem on that other 95%, which is where I agree
01:24:39.640 | with Aaron in, in, in, in like stage one, like we're in stage zero, no, no, no, no.
01:24:46.120 | I think it's in stage five and six.
01:24:47.720 | I think you'll need governments to take an entirely different risk posture for highly
01:24:51.100 | regulated markets.
01:24:52.360 | I don't see that happening in any, in any time in the near future.
01:24:55.360 | In life sciences, if you want anything in a clinical system, I don't need to tell you
01:24:59.120 | guys this, but you have to, you have to do a QA test on every single change that ever
01:25:03.080 | happens and be able to prove that you tested every single thing.
01:25:06.440 | So, so the idea of a probabilistic AI system generating the kind of code for you, you know,
01:25:11.600 | in a clinical trial, you know, workflow, like I just think, yeah, it's going to take a lot
01:25:15.640 | of change from, from regulators.
01:25:17.120 | Yeah, I mean, we got another decade of evolution here to make these things sustainable and
01:25:22.480 | have a high quality.
01:25:23.640 | I think it'll happen faster than everyone thinks.
01:25:25.600 | Yeah.
01:25:26.600 | And I think actually that's, that's like, there's actually no reason why that wouldn't
01:25:29.520 | be a good thing if that did happen to be clear.
01:25:31.360 | I think though, back to, to, you know, Jason's earlier point though, like then the counterbalance
01:25:35.920 | on the TAM compression is, is just now the sort of thing that we think of as software
01:25:41.360 | that market is now is, is much larger.
01:25:43.280 | So if you're, if you're selling, if you're selling, yeah, so, but like, but like that
01:25:47.720 | means that if you're selling, let's say $50,000 of security software, you know, maybe that
01:25:51.720 | goes down a little bit to, you know, $25,000, but, but you might, yeah, but you might sell
01:25:55.720 | $100,000 of the new agent.
01:25:58.520 | Yeah.
01:25:59.520 | So I agree a hundred percent with that.
01:26:00.840 | My only comment is the software industrial complex today has to shrink because the stranglehold
01:26:06.240 | that it has on how companies run is incredibly high for an experience that's incredibly poor.
01:26:14.040 | Hmm.
01:26:15.040 | Nobody, nobody raises their hand and says, gosh, this piece of enterprise software that
01:26:18.680 | I use in my day to day life is as good as Instagram or tick tock.
01:26:21.800 | Nobody says that.
01:26:22.800 | Well, I mean, except about box, I was leading the reviews earlier today and I was on a G2
01:26:28.160 | or one of those sites and it just box was just a rating for a tremendous, so.
01:26:32.640 | I love box.
01:26:33.640 | I love box.
01:26:34.640 | It's incredible.
01:26:35.640 | Well, I mean, I mean just to bring it all.
01:26:36.640 | What's the ticker symbol?
01:26:37.640 | Is it dollar sign?
01:26:38.640 | Well, we try to make it very simple for people, so.
01:26:41.320 | Is it dollar box?
01:26:42.320 | Is that what it is?
01:26:43.320 | It is dollar box.
01:26:44.320 | I'm putting a J trade in here because I may have to J trade this and it sounds like a
01:26:48.680 | really great opportunity for me.
01:26:50.280 | All right, listen.
01:26:51.280 | Aaron, you just got Kramered.
01:26:52.280 | What was your last J trade?
01:26:53.280 | Yeah, you got Kramered.
01:26:54.280 | My last J trade is I'm just, I'm loading up on MSTR and Bitcoin.
01:26:59.160 | I'm doing a, I'm shorting Bitcoin and I'm buying MST.
01:27:01.800 | I don't know what that.
01:27:02.800 | I'm not, I'm not trading stocks right now.
01:27:04.520 | I'm focused on investing in a hundred startups per year.
01:27:08.560 | One hundred new startups per year.
01:27:09.560 | I'm putting the first check into them.
01:27:11.360 | That's what I do.
01:27:12.360 | All right.
01:27:13.360 | Listen, enough.
01:27:14.360 | This has been a great episode.
01:27:15.360 | Aaron, you're amazing.
01:27:16.820 | Thank you for coming on constantly.
01:27:18.560 | Everybody follow.
01:27:19.560 | What is your Twitter handle?
01:27:20.560 | Sorry, before we wrap on AI, can I just tell you guys, are you at Levy or at Aaron?
01:27:23.560 | Yeah.
01:27:24.560 | Yeah.
01:27:25.560 | I love you.
01:27:26.560 | Livia.
01:27:27.560 | Is that the French pronunciation?
01:27:28.560 | L-E-V-I-E.
01:27:29.560 | L-E-V-I-A.
01:27:30.560 | Have you guys seen the VO model from Google?
01:27:31.840 | Yeah.
01:27:32.840 | Have you guys talked about how Google has like gotten some religion?
01:27:35.240 | Google's incredible.
01:27:36.240 | They are.
01:27:37.240 | They're firing on all cylinders.
01:27:38.240 | They're back in the game, baby.
01:27:39.240 | I literally paid.
01:27:40.240 | Shout out to Sundar.
01:27:41.240 | Yeah, yeah.
01:27:42.240 | And Sergei's going to work every day, by the way.
01:27:43.440 | He's at work.
01:27:44.520 | What can Brown do for you, Jekyll?
01:27:45.840 | Sundar, back in the group chat.
01:27:47.160 | Okay, there we go.
01:27:48.160 | By the way, did you guys see Genesis yesterday?
01:27:49.160 | Well, okay.
01:27:50.160 | Let me just put this up here, and then I'll tee it up to you, Freeberg.
01:27:54.000 | Google has been putting out new models.
01:27:55.660 | They have this new Google Gemini.
01:27:58.800 | And they have one with Deep Reasoning and their 2.0 model.
01:28:01.440 | I did a side-by-side test, same prompts, with our friend Sandeep Madhra on This Week in
01:28:06.040 | Startups just this week.
01:28:08.160 | We did multiple tests.
01:28:09.880 | He has the O1 Pro, the $200 per month.
01:28:13.800 | Gemini was beating that, hands down, for me.
01:28:17.400 | And I think he conceded that with their free model and their $20 model, if we just gave
01:28:21.120 | it decent prompts.
01:28:22.680 | I think Google now has reached parity, and they have an app out.
01:28:26.320 | Freeberg, what are your thoughts?
01:28:27.720 | Because people were counting them out, and here we are.
01:28:30.600 | I think not only have they caught up, I think they have exceeded.
01:28:34.040 | What are your thoughts?
01:28:35.040 | They were late, and the compounding effects are playing out, and it's only going to continue
01:28:37.720 | to compound.
01:28:38.720 | And I will say the data repository at Google, the engine that they have, the infrastructure,
01:28:42.320 | the team, everything down to components is advantaged.
01:28:45.160 | And so everyone's been kind of counting them out, but it's clear they're in it to win it.
01:28:49.000 | They're here to play.
01:28:50.000 | Freeberg, 70% of the usage of open AI is consumer.
01:28:53.920 | 70% is consumer.
01:28:55.240 | What does that mean for them as Gemini and Google get momentum?
01:28:59.440 | I think you'll end up having a Gemini that has ads, eventually.
01:29:05.200 | And you could pay and have no ads, or you could not pay and have ads.
01:29:08.640 | No, meaning do you provide enough value and of enough quality where folks don't need to
01:29:13.080 | then folks stay on Google?
01:29:17.000 | I'm asking you whether it impacts open AI, the quality of Gemini as it increases, or
01:29:21.280 | do you think that these usage cohorts are roughly set?
01:29:25.080 | I jump back and forth personally.
01:29:26.800 | My experience is I don't feel like I've got embedded data on open AI or on Gemini that
01:29:31.240 | makes me stick with one.
01:29:32.920 | Just like with Google, I'm going to go to the best interface, the best engine.
01:29:36.600 | So I do think that there's fungibility here and people will move over as they realize
01:29:40.560 | better results, better performance.
01:29:42.140 | But I will say that what we're seeing now with the data advantage at Google, so the
01:29:46.400 | internet, all these LLMs are language models trained on text from the internet, right?
01:29:50.800 | There's call it 50 billion words on the internet.
01:29:53.840 | And I think if you estimate the data repository in these training sets, it's like probably
01:29:58.720 | a couple terabytes, one to five terabytes or something like that.
01:30:01.760 | But if you look at the video data that's out there, there's hundreds of billions of hours
01:30:07.080 | or a hundred billion plus hours of video data, a large amount of that is sitting on YouTube.
01:30:12.040 | And by some estimates, there's a thousand exabytes of video data on the internet.
01:30:15.660 | So about a billion times more video data than there is word or text data.
01:30:21.480 | And I think we just saw that play out with the VO model that launched yesterday.
01:30:24.640 | But here's some examples of VO.
01:30:27.100 | Google has all of this YouTube data, you know, whether or not they're using it to train,
01:30:30.120 | I don't know the answer.
01:30:31.120 | I don't think they're allowed to, is what I heard from the insiders.
01:30:33.600 | They have to redo their terms of service to get explicit permission to use it.
01:30:38.560 | But you're right.
01:30:39.560 | These models are tremendous.
01:30:40.560 | It's amazing.
01:30:41.560 | And it's basically rendering physics or it looks like it's rendering physics.
01:30:44.080 | Now, Genesis came out yesterday and it's open source.
01:30:48.000 | So you can actually go play with this Genesis model, which similarly renders the actual
01:30:52.600 | 3D objects into a 3D environment.
01:30:56.280 | So rather than rendering a two-dimensional set of pixels to look at a video visual with
01:31:02.040 | Genesis, as you can see here, you can type in a prompt and it renders this extraordinary
01:31:06.400 | video that also has underlying it the three-dimensional objects that make up the video.
01:31:13.160 | Which means you could change the angle, right?
01:31:15.280 | And you could work with it.
01:31:16.920 | And also with Google and this, you can start to implement three-dimensional models based
01:31:22.520 | on some prompt that says something like, I want to have the camera angle at this point
01:31:26.680 | in the room.
01:31:27.680 | I want to have the room look like this.
01:31:28.760 | I want to have this color, this wallpaper.
01:31:30.880 | And suddenly everything starts to prompt in a way that you can actually render in real
01:31:35.120 | time a video game, a movie, a visual experience.
01:31:39.180 | And it goes to this point that we're unleashing the capacity for human imagination and creativity
01:31:44.360 | with these systems because it's no longer just a lookalike two-dimensional image.
01:31:48.520 | It's now an actual three-dimensional object that then renders the visuals to make this
01:31:52.080 | happen.
01:31:53.080 | So we're starting to see, I think, the next era of these models that goes beyond just
01:31:56.920 | text prediction.
01:31:57.920 | What do you think, Aaron?
01:31:58.920 | Yeah, there was another release last week of an experimental project for browser use
01:32:04.080 | by Gemini, which is another advantage because YouTube has obviously an incredible amount
01:32:07.680 | of screen-sharing data and videos.
01:32:10.640 | So you have, like we think about visual as just like, okay, it's going to be for developing
01:32:17.060 | CG characters or something.
01:32:19.160 | But it's actually just like all of the use cases of a computer are also on YouTube.
01:32:25.260 | Project Mariner is the name of it, right?
01:32:27.720 | Where your Chrome extension AI can control your browser to do things.
01:32:31.120 | And if you go to the AI studio from Gemini, they have a mode where you can turn on your
01:32:35.560 | webcam and then basically, it has full visual access to anything.
01:32:44.440 | And so they're cranking.
01:32:45.640 | And it's super exciting because you can tell that Google's woken up and they are just on
01:32:52.840 | full assault.
01:32:53.840 | So I mean, just in 10 days, the quantum, the AI, open source, Gemini updates.
01:32:59.460 | It's like every morning, you're waking up to a Sundar tweet that is some new breakthrough.
01:33:03.320 | Well, yeah, it's amazing when you fire 20,000 people who weren't doing any work and were
01:33:07.560 | involved in DAI nonsense.
01:33:09.960 | And then you say, "Hey, that's not the point."
01:33:14.200 | The point is they've had this compounding and infrastructure advantage, data advantage,
01:33:19.560 | personnel advantage.
01:33:20.560 | And now they were just a little late to the game.
01:33:22.800 | But when you have a better engine...
01:33:23.800 | But they were also distracted is my point.
01:33:25.360 | They were distracted with nonsense.
01:33:26.600 | You know, that's true.
01:33:27.600 | Come on.
01:33:28.600 | I mean, the evolution of this came out.
01:33:29.600 | It was making Abraham Lincoln, African-American and Betsy Ross was Asian.
01:33:35.360 | They were distracted.
01:33:36.360 | Okay.
01:33:37.360 | A big part of Google's orientation historically has been don't mess up the machine, a classic
01:33:43.320 | kind of big company dilemma, where if I do something that's either disruptive to my core
01:33:47.280 | business or if I do something wrong where I will invite regulatory and consumer scrutiny,
01:33:51.640 | I'm going to get wrecked.
01:33:52.800 | And so they avoided launching stuff early.
01:33:55.040 | And what they've done is they've changed the posture in the last two years.
01:33:57.800 | And now the posture is launch early, launch aggressively, push hard.
01:34:01.620 | We have to win this battle or we're going to get eaten alive.
01:34:04.240 | And it's amazing to see the founders and Sundar lead this organization.
01:34:08.800 | I think there were a lot of question marks a year ago, but I think that the questions
01:34:12.640 | have been answered.
01:34:13.640 | I agree.
01:34:14.640 | I'm saying it right now.
01:34:15.880 | I think we've hit peak open AI in the market.
01:34:19.080 | I think they're going to be the number three or four player.
01:34:21.280 | I think Gemini, Meta and XAI are going to lead them.
01:34:26.800 | If we're sitting here in three years, I think open AI is number three, four or five, not
01:34:32.000 | one or two.
01:34:33.000 | I don't think they get the one or two slot.
01:34:34.880 | What do you think, Aaron?
01:34:35.880 | Open AI?
01:34:36.880 | Who's in the one and two slot?
01:34:37.880 | I don't think you want to bet against Sam and Greg.
01:34:40.480 | You don't want to bet against Elon.
01:34:42.000 | You don't want to bet against Sergey now in office with Sundar.
01:34:46.520 | You don't want to bet against Zack.
01:34:48.240 | So I think the rankings are kind of less interesting as much as just the fact that it's like game
01:34:53.160 | is on.
01:34:54.160 | You do not have incumbents that are sleeping at this point.
01:34:57.680 | That's just going to push everybody forward.
01:34:59.120 | So it's just an incredible time for everybody.
01:35:02.440 | What a time to be alive.
01:35:03.440 | All right.
01:35:04.440 | For the sultan of science, David Freeburg, the chairman dictator, Tamal Paya-Hapatia,
01:35:08.720 | and I don't have a nickname for you, Aaron Levy, but we will have one soon.
01:35:12.440 | This is the All In podcast.
01:35:14.760 | We miss you, Saxie Poo.
01:35:16.240 | We miss you.
01:35:17.600 | Come back soon, Saxie Poo.
01:35:19.240 | Come back soon, Sax.
01:35:20.240 | We miss you.
01:35:21.240 | All right, everybody.
01:35:22.800 | Have a great Christmas break and we'll see you again shortly.
01:35:24.920 | Go sign up for our spam at allin.com.
01:35:27.600 | Sign up for the spam newsletter.
01:35:28.600 | We'll send you some updates on the button.
01:35:31.200 | It's a good thing to sign up for.
01:35:32.200 | I'm joking.
01:35:33.200 | I'm joking.
01:35:34.200 | Love you, boys.
01:35:35.200 | Sign up at allin.com.
01:35:36.200 | Love you, boys.
01:35:37.200 | Bye-bye.
01:35:38.200 | Bye-bye.
01:35:39.200 | All in.
01:35:59.200 | All in.
01:36:23.800 | I'm going all in.
01:36:24.800 | I'm going all in.
01:36:24.800 | I'm going all in.
01:36:32.800 | ♪ I'm fallin' in ♪
01:36:35.380 | [BLANK_AUDIO]