AI is saving the economy. For now.

0:00I'll be honest, Eric, I'm having an annoying day. I have been trying to reach sources in the AI world because I really need to get my story out. And no one wants to talk because everyone is paying attention to the GPT-5 release. Well, it's like a bank holiday in tech. It's just like everybody's... And they complain. I feel like the competitors all complain that OpenAI's people all tweet in advance and they're like, we got something big coming. They're good at building up the sort of excitement on Twitter. And then everybody clearly is excited. It feels like all the other companies are orbiting, you know, these opening.

0:36They got to be excited about something. Cause remember last year, Sam had the 12 days of shipmas where they were putting out a new product every day, which I think culminated in four, four something. I remember which model, but the fact that I don't remember clearly means that it didn't build towards something huge. Whereas, you know, this one is supposed to be really meaningful. Like this is supposed to be the moment where GPT-5 and open AI reasserts itself. as the best model that's out there. And specifically on coding, they in their embargoed press conference, and then later in the public press conference today basically said, we have the best model out there for coding.

1:16And also deep learning. And on coding, everybody would like them to get better, right? People are spending a ton of money on it. Anthropics is dominating, and competition is good for consumers. So we'd like them to succeed. Not clear yet, if they have. I mean, the prediction markets, I think there's a betting pool on who will have the best model at the end of the month. And Gemini is leading and I think gained a lot of ground when GBD5 came out. So the initial reaction I think is not so positive. This podcast is supported by Google. Hi folks, Paige Bailey here from the Google DeepMind Debrel team.

1:57For our developers out there, we know there's a constant trade-off between model intelligence, speed, and cost. Gemini 2.5 Flash aims right at that challenge. It's got the speed you expect from Flash, but with upgraded reasoning power. And crucially, we've added controls like setting thinking budgets so you can decide how much reasoning to apply, optimizing for latency and costs. So try out Gemini 2.5 Flash at aistudio.google.com and let us know what you don't. And I feel like a real loser. I mean, I keep opening chat GPT waiting for Christmas to arrive and mine still says 4.0 and I'm paying you guys.

2:33I've got, you know, a plus account. I asked chat GPT like, oh, are you, what do I do? And it's like trying to do like logic puzzles with itself to see if it thinks it's actually GPT-5 or not. It's like, well, if I was 4.0, I wouldn't be able to solve this problem if I was 4.0. So I must be 5. You know, it's like, why can't they just program it? Doesn't it? I thought they give these things prompts where they say, like, doesn't your prompt tell you what model you are? Oh, I have. Yeah, I guess it should be able to say that. That's a fairly basic feature. That's just like the about page on your iPhone.

3:09Like, yeah, I forgot what kind of iPhone model I have. But you've actually plunged this thing into like a deep existential crisis. if you're like have you been outmoded yet you know did they have they had like a funeral recently for claude uh for one of the early no i didn't see that wait yeah yeah uh kylie robison uh our buddy over at a wired now i think uh she had a piece about this talking has she been on the show yeah we need to have her on the show no we should have she lives down the street for me too we should have her in studio but uh yeah she did this story about how when they retired one of the early versions of the Claude model.

3:41They had like a funeral for it, which is a particular type of AI researcher. Very anthropic. Yeah. Oh yeah, for sure. Anthropic, by the way, is definitely feeling the heat. Like they rush released, it seemed like their updated version of Claude, like Claude Sonnet 4.1 or something this week. So it's been, and then also OpenAI had the open source model that came out, which everyone for whatever reason declared, not everyone, you know whatever twitter's algorithm wanted to show me like declare it as like a kind of a bust uh it didn't seem like it certainly wasn't met with like the deep seek like ferocity but deep seek was about other people being able to play catch up to open ai right so open ai you know there's such high expectations i mean the thing that open ai has that everybody else doesn't have is users you know just like there's the inertia that consumers will use whatever we have Like we can say, oh, the 4.0 announcement was underwhelming, but it's like, I'm going to chat UBT and just waiting for the thing to update.

4:42But there's like a more powerful model, which I think there's no debate as to whether or not 5.0 is better in most regards to like the earlier OpenAI models. Like, is that going to drive more growth? Like, who are these model updates really for? I get the like coding side of it. Like, that's a very competitive field and Anthropic basically owns it now. So like OpenAI has business reasons to be a real player there. But, you know, what is – chatGPT is at like 700 million weekly users now. Like, is the thing that takes them to a billion, like five being that much more powerful? Or it's just like generally, like there's still more word that needs to leak out to the non-initiateds about like chatGPT as a software rather than the fact that like five can do whatever logic puzzle is slightly better.

5:30At first, I was like, this is an insane question because it's like, obviously, you know, for consumers, they have to go find how the model can deliver value. And the smarter it is, the more value they'll get out of it and the more reasons they have it. If they're using it a lot, OpenAI will find more ways to charge for it. So obviously, you know, a better model means a lot of stuff. And then, you know, real businesses that will build on top of the API will find use cases and be able to charge for it and handhold users to the value. But I do agree with a piece of the question, which is just like sort of astonishment that these models can be as impressive as they are today.

6:11And there's still lots of people who don't use them at all. And then some of us that do use the models, we're not really trying to apply them in all the cases we should. But I think that's just because it's like the sort of like, you know, businesses are going to need to build on top of these models to handhold us to exactly the value. And so getting smarter is great for the API business. And that's where I think a lot of the business could be long term. Yeah, but that implies that people were using earlier versions of the models and were underwhelmed by it because it was technically not as powerful as they wanted when really could just be like they haven't been exposed to it.

6:48I'm talking about normies, right? just people that would be using chat gpt i just think you know some hallucinations and you don't trust somebody you know somebody lies to you a couple times you're like don't trust you you know i i think sort of consistency is really valuable in building trust and so if these models can can become just more reliable and i think that's part of what open ai is saying here it hallucinates less uh and that that could help it a lot i mean it's worth saying i mean while we want to get much more intelligent, part of the big development here is just that this is a unified model that OpenAI is going to route you to, oh, this is a pretty easy question.

7:26We don't need the model to waste all its resources thinking about it. And, oh, this is an intense question. We'll think about it. And so that could allow the model to operate more efficiently and also just make it life easier for consumers to get great answers without having to pick the appropriate model. Yeah, I think it's both like a model selector, but also my understanding is that the models themselves have been improved. So it's not just like we've routed to the most useful thing to answer your question. But, you know, the tool, the tool itself is more powerful. Yeah, I don't know. I mean, I don't go you use this more than I do.

8:02I don't yet pay for the plus version of it. $20 a month is just I'm not there yet. I'm not sorry. I'm not even against it. I just like I haven't found$20 worth of value. for me that is like that much better than the free version i just like proofread it feedback run my stories through it what do you think what am i missing like it always gives me a couple useful thoughts like all right it's a new model i mean i think the last word on it to me is just we always try to like decide whether it's a good model on the first day and i think nobody knows right away it takes takes some time we we write off models and then they become sort of our daily drivers and then other times we get excited and then nothing.

8:43So I think we'll see people play around with them. Yeah. I also think we can't understate how significant it is for open AI in terms of, you know, it needing to be some kind of a leap for them because a lot of the conversation has been, you know, GPT, everything has been more or less iterations on four. We came out like a little bit over two years ago. So this is like a, like, there's no way around this one. This wasn't, And, you know, the backstory of building this, you know, this model was that for a long time, they were working on this giant model. Like, I don't even know how many trillion parameter model that was, you know, they called it Orion internally.

9:19And they realized most of the way through, it kept being delayed and delayed and delayed that it wasn't showing off large enough improvements and capability for it to be for it to merit being called five. And at the same time, you know, this breakthrough came with like deep reasoning and like logic, you know, the ability for it to like handle more complex pain of thought. And so they ended up calling Orion GPT 4.5 and it's basically unused as a model. No one really talks about it. But five is the real deal. Like it has to be something. It can't just be written off as like, yeah, we're still iterating on something.

9:51And this does not feel it does not feel like what the four was the iPhone for the one where it's like, oh. new mark or it's like they have the app store and well 10x right the iphone x the first time they had like the full screen oh yeah exactly it's not like that it's not one of the big iphone moments where it's like oh they really changed it this is like oh yeah they they come out with new phones every once in a while and sometimes we we try to get ourselves excited about it that's how this feels to me and i agree yeah we've been talking about five for a long time thought five would be an insane moment.

10:24And so far, I'm not seeing that level of hysterical euphoria that I would have expected six months ago for ChatshpD5. Yeah. Which again, back to what you said earlier, is maybe why people are expecting Gemini to leap ahead because OpenAI's best shot just didn't blow people away as much as maybe unrealistic expectations led them to think. They're in a big fight in terms of model quality now. You've got Anthropic very capably putting out stuff, you know meta i'm assuming will come out with something uh of value and you know google is very serious microsoft's still irrelevant by the way like i'm never not going to say until they do something real here yeah like it's just a very different landscape than when four came out let's talk about a related but different topic so uh the capex build out data center build out we we're in this moment and i keep telling this to other people ai adjacent to like man ai is apparently like propping up the entire economy.

11:24It's like, thank God, you know, we're in the orbiting artificial intelligence because that's, that's the thing saving the economy. And, you know, the Wall Street Journal, uh, when I was covering AI. So you're working for a company with AI conference. So you got even closer to it. Yeah. That was the problem. I wasn't close enough to AI. Exactly. But the, the things that, you know, it's the infrastructure build out that there's real capital expenditures going into this and that the big tech giants, you know, are spending aggressively. Yeah. Basically what I saw happen. And it's kind of funny when like an idea just percolates through, you know, the discourse, but you had at least three different articles come out in the last two weeks pointing out.

12:07We were earlier than some of this, like, sorry, there, there is this, what Kodrosky or whatever he, he sort of initiated it. Is it Paul or Peter? Yeah, there's Paul Kodrosky. I saw. So we riffed on that like a couple of weeks ago. And now, yeah, the journal, Noah Smith, a bunch of people are like, oh, maybe everybody loves a good, oh, the world could end in financial collapse thesis. And that's sort of what this is, right? Yeah. And essentially, and what I think Kodrosky's article did a very good job explaining was like, And we had some new GDP numbers coming out. It's also earnings season, so people can more easily calculate what the CapEx spending looks like for all the biggest companies.

12:51But a non-insignificant percentage of it, what was it? Almost 1 % of GDP growth came from AI CapEx spending. And that is the chart that Kodrosky put together showed that outside of the building of the railroads and the telecom build out during the dot-com bubble, have you seen that much of GDP growth confined to one particular sector? And it's an infrastructure build out that, at least in the case of AI, is well ahead of its broad-based adoption. And I guess the next question to all of this is like, well, is this a bubble? Is this something that could potentially bring down the economy if AI doesn't end up becoming what a lot of people are expecting?

13:42And I want to take the bear case on this one because it's more interesting to me and say like, oh, yeah, we're absolutely circling a Bear Stearns-like event where AI doesn't pan out. But I just don't see it. Like I can't follow the train of thought that leads me to AI. You know, CapEx spending is going to somehow when AI doesn't get as widely adopted result in like the collapse of 10 % unemployment. This is like meta going on on the metaverse. Like they spent tons of money trying to do that. And like, well, didn't work. The key here is that the Mag 7, they make tons of money. And so they can spend a bunch on CapEx.

14:26For everybody but Meta, it's still, I think, less than their actual sort of cloud revenue, right? And even if it were bigger, it's just like they stopped doing it one day and they say, okay, well, we'll print money again. Like, I don't see the debt story. I know some of these things are like finance, but they're super valuable companies that could raise unlimited sums of money that produce real cash. Like, I just don't see what's what do you like if you're a monopoly, what are you going to do with that? Right. If you have just an incredible margin, you have cash flow, like crazy cash flow. The money's got to go somewhere.

15:00You could do stock buybacks. You could. And this is a good thing. I feel like the sort of left, it's like, oh, share buybacks. Exactly what you're saying, share buybacks and dividends, they're bad. They're not doing anything. It's like, here they are. They're excited about something in the future and they're investing in the real world infrastructure to run at it. I mean, I think if you believe AI will be good for the world, then it's like, great. Yeah. So what if they're spending a little bit ahead of, you know, themselves and they find out they sort of overbuilt too soon? Yeah, I guess like there's an argument that like the actual technology that you're building is something that's going to be like job displacing.

15:40So like there's a medium to long term impact of building up the technology there. But like, I mean, that gets into like the, you know, kind of technophobic argument of like, well, what do you want to do not invest in this technology that's going to be transformative? Like I never, and I, again, I'd love to take the bear case here, but I just can't follow that logic of like, let's, let's not do the thing that could potentially create like God. And the people making those arguments are always on their iPhones. You know, they're just like, it's, they have once like, I mean, what, it's the choppers of the world who are like anti-tech, but then they're clearly like tech addicted.

16:14You know, it's like they never choose not to use the products that they say, oh, we shouldn't we shouldn't build the next one, even though I love the current one. Yeah, it never makes. Yeah, I don't want to go to let's we I would want to save that part of the conversation for a different episode, because I do think there is something to say about like the social impact of AI and the effect that it's having on like intelligence and like sociability and like, you know, trying to think about stuff. I think people not reading. I mean, we talked about this on the Cerebral Value podcast. I made a defense of reading and writing.

16:47I do worry about that. But to me, that's more of a social media phenomenon. AI, as it exists right now in a chatbot, actually makes you think and write and engage. I think it's much more intelligent than the lean back. You're forcing me into this argument now. Yeah. What? Yeah, I think the intent aspect of chat GPT is much better than the passive and engagement of social media. There's no question to me that that is an improvement. I think the idea of outsourcing... And social media is built on engagement, whereas at least for now, chatbots are predicated on getting you truthful answers and appearing smart on tests.

17:26So obviously, they're much more virtuous than feeds. I think outsourcing writing and all thought to an AI is like problematic for a society, like rather than cogitating over an issue. Yeah. Or like I need to fire these employees. How can you please write this for me so I don't have to deal with like the emotional labor of what it means to communicate to someone is like a net negative on society. Like, I think there's all kinds of bad things that like asking someone to imitate human emotion and logic and having that be like then put in your own words, like has on a society. But I don't think that's what Kudrosky was getting at in his article.

18:04No, we've driven pretty far from the sort of dead overspending abuse of it. Yeah. Look, I think there's also like, you know, yeah, for the Mag 7, I don't know, remember exactly who's in the Mag 7, but for all these companies. Microsoft, Alphabet. Yeah. These are like cash machines. Like they have no issue investing in infrastructure. They're not taking on huge amounts of debt. There are companies in the AI mix that I'm a little bit more concerned about. Oracle, which I pay attention to, is taking on debt so they can finance the Stargate shit that they're doing with OpenAI. But isn't Oracle just, they're like professionally rich and good at selling techs.

18:43I mean, I haven't dug into them as much as you, but they're doing the financing because they have money, right? Yeah. And like so far, no one's like freaking out about Oracle because they're taking on debt to like, you know, buy the GPUs that are going to be in the Stargate data centers. But they're just in a different class of company than the Mag 7 because they're not in the Mag 7. And, you know, there's also ones like CoreWeave, which, you know, is a crazily valued stock right now. And they're taking on, they keep going back to the debt markets to do build out. So there are some like, there is some concerning leverage, I think, for some of these companies that are taking on debt.

19:20But it doesn't seem like a load bearing beam that if those things go under, like you're going to see a 2008 like event. I'm definitely open to a world where here's the bull case. Sorry, the bear case to me, which I'm not necessarily subscribed to is all the tech stocks are massively inflated because they're investing in the technology that's driving some of their revenue and driving their multiples. Because basically it's like we're going to, you know, big tech companies are the biggest investors in AI startups. Those startups are turning around spending on cloud computing and foundation models and stuff where they're generating revenue.

20:08So they're spending money to make money. It's that old revenue round tripping issue. And then the opportunity of AI is also increasing their multiples. So I could see a world where, all right, suddenly AI is not delivering as much as people thought or AI is delivering, but there's no margin to be had because open source is so successful. And then all the stocks crash and they crash because they're sort of benefiting in multiple ways, both revenue and multiples. And so that would hurt them. And I think given that they're so important to the performance of the S &P 500, it would just be like a bleak.

20:47The American economy would all of a sudden, everybody would feel much poorer because the value of the stock market has gone down so much. But it doesn't seem it seems like, oh, we're all just like a lot poorer. I don't see this sort of like ripple effect. They're companies that make a lot of money. They make way less. They're way less valuable. And then suddenly America just feels weaker relative to the rest of the global economy. But I don't necessarily see like who's going bankrupt. Yeah. And like, let's just also remember that like these companies all crushed earnings. Correct. And I'm mostly bullish.

21:21Bullish. Yeah. It's like, it's clearly this is where things are going. We spend all our time on the internet. These companies are the main entities innovating in the world. They're very sophisticated businesses. Yeah, I don't know. I'm pretty bullish. Yeah, I guess the part of it that I think is worth being concerned about or paying attention to is the fact that it is a non-marginal part of GDP growth, spending on these things. And you have to just pay attention to that. Whenever one sector is that responsible for, absent all of these things, we would have fairly anemic growth. And the argument also is like, well, that's fine because this is transformative.

22:04We should be investing in these things. And this is all going to later lead toward even more growth because we're going to have all this efficiency and people will be more productive. But it's getting into the territory where it's interwoven into the broader economy. It's not just like some tech experiment of, you know, some companies building something that could be cool. Like this is now part of it. Like you're, you know, you're woven into the fabric of what's making the economy work and problems within that. I don't mean we won't even foresee what the problem could be, but problems within that will have ripple effects.

22:37And that's it's just worth paying attention to. Like I don't see the warning signs yet, but like it's, you know, it's built into it. Well, it's just also just it's been too long, you know, in between crashes, right? In startups, there was a downturn. You know, 2021 was a peak. We came down. There are all these software companies are overvalued. We did have a minor correction because of the pandemic inflating the value of tech stocks and driving forward revenue that then slowed down. But AI came around at just the perfect time. And like, there's no like, there's no morality play there. It's like, oh, we should have, we should have really taken our comeuppance more.

23:18It's like, no, we got lucky that there was like another really cool idea. Just as tech was about to take it pretty hard from getting overexcited about everybody working remotely. And so that sort of saved things. And that is the tech idea that it's like, oh, we'll keep up coming up with new ideas. And the new ideas are real things and that will create value. And it's not just like financial games like on Wall Street. right and it's also just kind of the way things work like i see this argument being applied to san francisco and be like man if it wasn't for the ai boom san francisco would really be fucked right that's his job they're all hard at work like coming up with a new thing going where it's kind of what they do here there's always something i mean it's like man if it wasn't for the gold rush san francisco would really be right right right yeah it's it's like that's not really the argument you think it is because that's just like that's how a boom economy tends to work And the fact that like San Francisco keeps coming up with the new thing that matters to the economy is good at is they all left crypto because crypto stopped making them money and they shift over to AI.

24:20And honestly, you want things to die so that people can go to the actual productive parts of the economy. And, you know, yeah. Well, you're closer to the crypto people than I am. But like, like, are they all just like riding a false high right now? Like knowing that the value of all of their holdings is like not really logical at all. What do you mean Bitcoin? I mean, the dollar is eroding at the moment. So, I mean, Bitcoin is like, there is, I mean, I don't hold a lot of Bitcoin. I have what, a thousand dollars worth or something. But like, definitely, I think, you know, Bitcoin is a psychological investment thing.

24:59Just like, you know, so many things. And it's a bet against the dollar. And I think the dollar is in a bad place. So I don't know. I think the other thing in crypto, which this is, you know, it's just that it's like fraud is legal and there's tons of fraud in crypto. And so it's like you can do whatever you want, seemingly. And so I think that's going to open up some activity. Right. Well, I'm just thinking about it because, you know, I was reading our buddy Mike Isaac's story about San Francisco. and like, you know, this moment, the, you know, the AI moment and how we've moved to the hard tech era in Silicon Valley.

25:36And, you know, he had this like one, the clause in there of like San Francisco is used to like things like blowing up. And he's like, like crypto. And he's like, also things tanking crypto, which is like a funny structure, but also like, like crypto is not tanking right now. Like just in terms of like the value of the currencies, it's actually really high. So that wasn't the best analogy. Did you read that piece by the way? yeah i read i read some of it i mean i yeah i don't get you know it's just the the trend story sort of thing you know it's like this is the new york times you know it's like this this is where we are i mean i think it's generally correct you know it's just like i don't know it's not it's it's too high level i guess yeah i mean first of all you know mike's a buddy and i always you know want to support my friend's work and you know the new york times you know loves calling a moment and explaining to the normies what's going on here.

26:30Um, you know, there's, there's always like kind of a combination of like things that have been going on for like years that we just call is like, Oh, this is just happening now. It's like the year of efficiency, I think was like 2022. So it's hard to say that's like specific to the, a lot of the things about tech are just things humans do. And a lot of things people hate about the media are just things humans do. And a thing humans do are say, what is, what is the mood of the moment? And you know, there were tech buses. We have these narratives in the New York Times. And I think Mike Story does a good job of trying to coalesce.