Pablo Torre Finds Out - Share & Bubble & Tell with Mina Kimes and Derek Thompson

Episode Date: March 5, 2026

Is A.I. gonna boom, bust or just become Excel? And how screwed are we either way? Plus: Peyton Manning, Lamar Jackson, Geraldo Rivera's vault, a young Tim Pawlenty, vibe-coding NFL scouts... and Filipino Jell-O. Further content:• "A.I. as Normal Technology" (Columbia University)https://knightcolumbia.org/content/ai-as-normal-technology• "The 2028 Global Intelligence Crisis" (Citrini Research)https://www.citriniresearch.com/p/2028gic• "Nobody Knows Anything" (Derek Thompson)https://www.derekthompson.org/p/nobody-knows-anything• "This Is How the A.I. Bubble Will Pop" (Derek Thompson)https://www.derekthompson.org/p/this-is-how-the-ai-bubble-will-pop• "Fear of A.I. Eliminating Jobs Makes Its Way to Football (Mike Florio)https://www.nbcsports.com/nfl/profootballtalk/rumor-mill/news/fear-of-ai-eliminating-jobs-makes-its-way-to-football• Previously on PTFO: How Artificial Intelligence Is Already Changing Sports, with Daryl Morey and Sendhil Mullainathan• Subscribe to "The Mina Kimes Show featuring Lenny"• Subscribe to "Plain English with Derek Thompson" Hosted on Acast. See acast.com/privacy for more information.

Welcome to Pablo Torre finds out, presented by eBay Live. I am Pablo Torre, and today you're going to find out what this sound is. It's a really strange form of marketing to predict that if your technology is successful, if the effect of that success will be a calamity for the United States. Right after this ad. I want to explain Mina and Derek how you guys know each other, because Derek, this is the first time that you and I are really getting to hang out together in this context. The origin of this episode is that you guys have been beefing off air,

but I want to understand the back story. Yes, you guys are beefing. That's not at all accurate. We're confronting Derek. I would not characterize it as a beef. I would characterize it as a constructive disagreement. I wouldn't even go that far, personally. A deconstructive disagreement.

Wow, even worse. We used to call it Satan, fake beef. Wow. That's great. Derek and I have been friends for like 20 years. Is that too long? No, less than that. Almost. Maybe 15? Yeah, maybe like 15 years.

Because I met you the month I moved to New York City. And I moved to New York City in 2012. So in the summer of 2012, is when we met. I think we were co-honorees at some award thing for young journalists, whose name currently escapes me. And for some reason, I recall, like, we were like the two most maybe like overdressed people at this particular ceremony. And so Game Recognized Game, at that point, you were working for Fortune and doing investigative reporting, and we got lunch.

We were both business journalists at the time. Yeah, we had a lot in common considering like similar jobs and kind of things like that. And yeah, we were early 20s, though. So that was a long time. And then we stayed friends throughout the years, kept in touch. And before I even started working in sports, I remember Derek shocked and awed me with one of his, wildest takes, which was that he doesn't root for an NFL team, he would just root for a quarterback and then follow him. Peyton Manning being, which just like NBA happens all the time.

Not a thing, really, in the NFL. I was ahead of my time, Mina. I was way ahead of my time in terms of following players over teams. I mean, now with fantasy, there's a lot of people who I feel like don't follow a team at all. Like they de facto root for the players in the fantasy teams because that gives them way more emotional feedback than whether or not a team wins. And as you said, in the NBA, I think following players over teams is pretty standard. So it is, or maybe was an embarrassing position in 2012.

I did fall in love with Peyton Manning, because he and I both shared the biological curse of high foreheads. And so, again, that was just game-recognized game. And, yes, since Peyton has retired, I've basically been an absolute free agent. And so, by the way, I was around in the time-life building on a different floor of Sports Illustrated when Mina Kimes, rising star, award-winning business investigator reporter

was also this character who showed up on like Bloomberg looking very goth. That's just how I was going to imagine. Before I knew how to do my makeup on television, I looked like an insane person on C-SPAN. Anyways, that is our lore. That is Derek and I's lower. So there you go. And so when did you guys start hating each other and arguing offline?

I will answer this question directly. It's probably now where Mina's going to go. before we started hating each other offline about artificial intelligence. After the first or second lunch that we had, Mina sent me a photograph of young Tim Pawlenty, former rising star in the Republican Party, and said, Derek, do you realize that you look exactly like young Tim Pawlenty? And I think one really has to like remember the world of 2012

to even like live inside of this reference. It's genuinely one of the most hateful things you can possibly accuse your Democratic friend in being. It's attested it really to just how charming and smart mean it is that even despite this just horrendous offense, I was like, you know what? She's still pretty fun to hang out with. I guess we can still get tricks. Okay.

One of my worst attributes, and I have many, socially, is once I see a looks like, I cannot physically restrain myself from pointing it. Even when you are like, it's a high wire act, you're, because, you know. It is a risky thing to say to someone unless it's someone really good. I've done it to Pablo, Derek. I mean, I... Instead of getting a former governor of Minnesota, Tim Pawlenty, I got Andrew Lopez from the bear from the Forks episode. It's just undeniably true.

Accurate. I met Andrew, and we both ultimately had to agree that Me and a Kimes was right. Which is all to say, by the way. Also, the guy in Miss Rachel. There was Spider-Man meme. He looks like the guy in the Miss Rachel universe, Derek, which I know is a parent of a young child, you can also, I pointed this out to you. It's the guy, I don't know if he's Filipino or Filipino. He is Filipino. Yes. You know you instantly knew what I was talking about.

As the father of two-year-old who just saw the episode where he talks about being Filipino, I can tell you. Angelo Soriano often referred to as Jello is very Filipino. By the way, there's a real Filipino Renaissance. The blues clues guy, the new guy, Josh, the guys who are secretly driving your Waymo, obviously Filipino, as we've, learned in the testimony from the chief safety officer of Waymo. An ongoing theory, by the way, is that AI is just 60 Filipino guys that are just operating your stuff from a call center.

Really productive Filipino guys. Yeah, incredibly. Are they? The amount of market cap that is hanging on those 60 Filipino guys is something truly impressive. And that's why we're gathered here today. I do want to understand, though, because Derek, the other reason I brought you on here besides your desire to argue mercilessly with Mina about AI is you do all of the reading. That's why I consume your work, and part is because I know you to synthesize lots and lots

of research and text and you do a lot of reporting that helps inform what the state of the union is right now. And so before we get to the argument specifically that you guys were having, if you were to say to a person who feels like they know what the fuck's happening with AI and gets the idea that, like, maybe it's God, Maybe it's the devil. Maybe it's the thing inside of Geraldo Rivera's vault. Maybe it's actually all of the riches in the world. How would you characterize where we are right now?

It's March in 26. What is the log line of what's up right now? One line that I think is fair to say is that there's no way that artificial intelligence isn't going to be one of, if not the most important, stories of the 2020s. Because either one of two things is true. Either artificial intelligence is a bubble, in which case companies are supposed to be.

spending $700 billion per year. That's two Apollo programs per year. The Apollo program was $300 billion spent over 10 years inflation adjusted in the 1960s. So either it's a bubble because we're spending all of this money and the revenue is never going to catch up and what's going to happen to AI is exactly what happened in the dot-con era, is exactly what happened in the railroad era that takes down the stock market. It takes down the economy. It takes down banks. It certainly, if this happens the next 18 months, transforms the 2020. election picture because any incumbent party running on an economic recession is facing an enormous risk. Or it's not a bubble. And for it to not be a bubble, when these companies are spending $700 billion

a year requires that the AI companies make hundreds of billions of dollars a year in the next 24 months. That would be the fastest growing business in history. And already, by some, measures, Open AI and Anthropic, are the fastest businesses to take their annualized run rate to $30 billion collectively. They're already maybe the fastest growing businesses in modern capitalism. So if that happens, like if the revenue actually does keep up with the spending, I mean, for this economy to find half a trillion dollars per year in extra spending that's going to something that's competing with labor, well, that's the biggest story. of the decade. So from my standpoint, there's no off-ramp here for AI being the most important story of the

decade. It's either a bubble in which case you have to pay attention or it's not a bubble in which case you have to pay attention. I think, Derek, what I find so tricky about the bubble or not question and like the actual economic impact, which goes back as a business journalist was something I thought about a lot. I didn't work at fortune during the dot com. But we had all of the issues from that period prominently displayed the pets.com. And then I did join during the financial. crisis, which is a very different kind of bubble. I think what I find so tricky here is if AI and the actual impact on the technology is more akin to what they call a normal technology, right?

There's the famous piece that I think it was in the Columbia Review written about AI being normal technology. If it's neither personal computing in terms of being a transformative, completely revolutionary thing that's going to affect everything and make things better, but it's also not NFTs, where it's just, you know, bullshit. If it's somewhere in the middle, which is kind of where I land, based on my reading, but valuations right now and the markets are so out of whack, what does that actually mean for the next 10 years, both from an economic perspective, certainly, but also an industry

perspective, right? Like if this thing is actually useful and it's important and it's threatening, but it's not that. What does that mean if we've so dramatically overvalued? Because from where I stand, I see a technology that where most of the money is just from other companies paying for it, or for the services or the chips. I don't see an economic impact at the ground level. And I don't see humans paying for it right now. So it's so hard for me to...

Can I just put the meme, Derek, in front of you? I think it's the image that someone posted of like a surge projector plugged into itself. And it's like invidia. And the question that means is circling to summon 2,000. and eight when all of us were doing some form of print journalism. Is it now too big to fail? Is that what we've boxed ourselves into? Well, Mina brought up a scenario that I think is really easy to describe. If AI is anything close to NFTs and OpenAI is currently valued at $600 billion, and Anthropic is currently valued at $400 billion, it's a bubble, period. It's over. These

companies are doomed if this technology is NFTs. So along that timeline, you are, looking at one of the biggest technological bubbles of the last few generations. It's not NFTs. And the reason it's not NFTs is that NFTs did not transform anyone's job, unless you were like a collector of digital tokens, digital artifacts, essentially. Look at what's happening in the software industry. The job of being a software programmer has been completely changed. Let me give you a sense. There's this company called Meter, NETR, that does these famous benchmark studies of AI, that does these studies that are quoted throughout Twitter, throughout the AI discourse of how powerful this technology is. And last year they published a study that was held up very strongly by the AI Doppers,

because it said that if you ask the best software programmers to use artificial intelligence, they think it makes them about 15, 20 percent more effective, but if you have a third party grade their work, they're actually significantly less productive on an hour-per-hour basis. And all these people who are doubters of artificial intelligence pointed this study and they said, look, this proves that AI is just vaporware.

This year, Meeter announced that AI for coders with technology like ClaudeCode and Codex from OpenAI has gotten so much better that they can no longer do this study because they can't find developers who are willing to work without AI in order to be the control group. So this is already a technology

that is completely transformed at least one major industry, which is coding. But I guess what we're going to talk to, I'll talk about a little bit more, is that the underlying value of this technology, the underlying skill, I think, is its facility with data.

And there are so many jobs that are, whether they're in data analytics or they're in research, they're putting together PowerPoints, they're working with Excel. There are so many jobs where the moment-to-moment tasks are so easily reproducible or accelerated by the best AI tools that exist, that I think it's far more likely than what you've seen in the last few months is going to continue, which is that anthropic and open AI have, in the last 13 months, gone from a combined ARR, a combined annualized revenue of about $3.5 billion to a combined annualized revenue of $35 billion. It's grown by 10x. That's practically

impressing in the history of capitalism. And it suggests at the very least that there are ordinary people, not just 60 people in the Philippines, but millions of people in America that are choosing to pay significantly for this technology month after month. People who were paying it for it out of pocket? Or you mean in companies and in their jobs? Because I think that's an important distinction here. It's a great question. And it's hard to decompose. exactly how much of this spending is coming from individuals who are choosing to pay for it versus companies that are choosing to pay for it. But I would say this, there are a lot of companies,

a lot of major Fortune 100 companies that prohibit their employees from using Claude or Open AI and force them to use other models like Gemini. And that suggests that a lot of the money that's going into Anthropic and Open AI are from individuals choosing to use this technology rather than from companies forcing them to use this technology. And I think it's important to say, you go back to what the folks at Meeter said. It's not that the folks at Meeter have found that companies are forcing software programmers to use AI.

What they found is that the individual software programmers themselves will not enroll in this study because it requires that some of them be selected to be in a control group where they can't use this technology, and they're saying we can't do that. I don't dispute that there's use. utility for software programmers. And I think that, Pablo, my feeling about AI generally and why there's so much public backlash and it's so like politically polarizing, whatever, is it strikes me as

unsexy technology being marketed as sexy. Maybe sexy is the wrong word. But like, I see a lot of B2B use. I see a lot of data use. I see a lot of industries where it streamlines processes. What I think, where I kind of get my like a little bit suspicion about putting it more in the personal computing, completely transformative side is I would like to see evidence of consumer use rising to that, to meet that as opposed to like people using it in their job, which I do, like Derek, I believe that. Like I've heard stories about that. I know people in software who say that. I just don't think, I just want to see some evidence that people are paying for it, normal people. And a lot of this for better and for words does happen on Twitter, on X, the

everything app. It's a lot of where, frankly, the industry is like arguing. amongst itself. The AI industry. Yeah, and the stories are told, right? And I think storytelling is such a huge part of this still. We just had, I'll bring in, Derek, something you're familiar with the Satrini Research Substack, that moved markets.

And what that literally was, was a fictionalized account of what Satrini, the author of the post, who I weirdly kind of have interacted with before, and I'm eager to actually talk to maybe on the show one day about. But it was this looking to the crystal ball of like this is what it's going to be like. This is my vision for how maybe AI might be revolutionary

and also still horrific for the economy. And so in real terms, as the storytelling contest was being contested, you saw American Express and DoorDash and these specific companies that he cited as like those who may suffer from the agentic, the revolution in which maybe the ability to process data in this seemingly unsexy way, but with this frame around it that feels, if nothing else,

like it is cosmically impactful, it led to these real-life monetary hits, which is kind of insane. It tested it to the fact that as much as there is this economic reality that is being debated, we're also still here for the best story someone can tell us. There are lots of persuadable voters, even in the market, it seems like. Let me give you two answers to that.

So first, Gallup in December 2025 published its latest estimate of AI use at work. And according to its latest estimate, approximately half of the American public, 45% of U.S. employees report using AI at work at least a few times a year. The number who report daily use has also increased essentially quarter after quarter. So is this being forced on employees by corporations? But that would require a separate question to know for sure. But we're looking basically at a technology whose adoption is rising faster than any technology that's been measured, including the computer revolution. So I think it's important to say to Mina's question first that every indicator we can see shows rising use of AI that basically has no historical precedent.

All these people are using it for free, right? a lot of them are using it for free, but also the fact that anthropic and open AI are two of the fastest growing businesses in history suggests that there's a lot of people that are paying for it as well. So you have both rising, you have evidence both of rising adoption and evidence of rising revenue. And so I do think that when you put those together, you have a phenomenon that is like clearly showing penetration throughout the U.S. economy. But there's a second point here, or there may be a better way to put this is there's an interesting tension, I think, between Mina's point and the power of the Satrini argument. The Satrini post presumed not that AI was vaporware or that AI was being a technology pushed on employees by corporate overlords when the underlying technology didn't do anything. In a way, the Satrini post that moved markets by a trillion dollars, as Pablo said, presumes that AI is so useful that it displaces trillions of dollars of labor activity, that it is used so frequently and at such mass scale that you essentially have so many people thrown into unemployment that demand in the economy

crashes. And because GDP gross domestic product is equal to gross domestic income, if income in the economy declines significantly what you have as a recession. So the The Trini Post was essentially asking people to imagine not a world in which AI's impact on the labor market is being overrated, but rather a world in which we are underrating just how fast AI is spreading. And if it grows so quickly and companies use it so often and individuals using AI are seen to be so productive, their productivity will displace other workers who aren't using AI who will be laid off, they'll be unemployed, they won't spend money. the fact they aren't spending money leads to a recession, and overall you have the irony of productivity growth leading to a demand-side recession. That's the Satrini picture these painting,

and I'm not here to endorse it. I'm not here to say it's right or wrong, but I just want to point out the tension between these two questions that you and Paolo asked because they are making opposite assumptions about the future of this technology. I think what I hear me is saying, and I appreciate the call for receipts, right?

Is like, show me where, if you're talking about this as a matter of real, money, you're seeing customers actually forking over real money for it as opposed to these enterprise contracts, which, as we all know, have a different calculation in terms of what it suggests about its adoption. But look, it's funny that the journalist perspective actually can come back into the conversation here because on Twitter, where, again, all of these things are being contested, the response from so many people from Silicon Valley for years was learn to code.

That was the insult of like shitting on journalists. And now, of course, the number one thing that everybody seems to agree is that coding has been the most disrupted by things like vibe coding. But Claude Cod Code is a matter of like a technology that is seemingly doing stuff that we've never seen before at a speed with a facility we've never seen before. That seems beyond argument at this point that coding has been the first thing. There's no question. There's like a profound irony here that like these folks in AI, they, they talk about themselves as like inventing God, and the first thing they do is like saw off the branch on which they sit, right?

The first thing that it's done is turn English into the universal language of coding, thus not only sort of completely changing the job, but also potentially open the door to folks who, you know, our word cells, now suddenly maybe they're able to build apps that displace the apps built by the shape rotators that were supposedly talking about how we need to learn to code or otherwise accept, you know, permanent status in the underclass.

There's a profound irony here. You can say it's totally fitting that essentially if what these architects are trying to do, if the architects of AI are fundamentally trying to use the corpus of the internet in order to allow a technology to do all human work, well, of course the first thing they would understand how to automate is their own work.

I'm here trying to point out ways in which my perspective on AI and the economy is a little bit different than Minas. But I think it's important to say here, Like, the last column that I wrote for my substack was called Nobody Knows Anything. And that's a famous line from William Goldman, who's a screenwriter for all the president's men and the Princess Bride, and the first three words of his autobiography about the ability of Hollywood to predict hits was nobody knows anything. And my big thesis about AI right now is that anybody trying to make any macroeconomic prediction about what this tech is doing to us doesn't know what the hell they're talking about.

It's very, very hard, in fact, to see any effect that AI is having on the economy right now, whether you're looking at productivity or employment data, an employment rate is still 4.7%. Even the hiring rate, it's difficult to show that AI is having a significant effect. So I'm a believer that we're looking at a technology that's growing quickly, that people are paying for, and absolutely has the potential to transform the U.S. economy and the way we work and live. And also, I do think that, staring at this thing as clearly as possible, it's very, very hard to say, like, where are the macro-canonic receipts? I don't know where they are yet. I guess my question, before we get to the other side of this, which is what Derek and I originally went back and forth on, which is AI's actual utility for our jobs. But sticking with the actual bubble slash impact question, when should we know by, man? Because I am not someone who's here arguing the impact of this thing isn't going to grow or that it won't affect certain parts of the workforce. My position has just been generally like I have yet to see consumer, the evidence that the

consumer facing impact or that side of the industry per se has demonstrated proven financial growth commensurate with how it's being discussed by some parties. But my question for you is this. I think you're correct in saying we're not seeing it the actual economic impact at the moment. if we are still talking about this in two years and people still aren't demonstrably paying for this by their own volition, would you change your position? Do you think that this is something that actually has to start to show its work, per se, in the next two or three years?

So the single biggest podcast that I did last year, the single biggest article that I wrote last year for my substack, was an interview with the investor Paul Kodroski called, This is How the AI Bubble Bursts. So in a way, my most profound contribution to this debate has been to outline exactly why I think it's plausible that AI turns out to be a bubble in two years. That said, just because something's a bubble doesn't mean it isn't transformative. The railroads, the transcontinental railroads in the 19th century were four different bubbles that crashed the economy. It was also transformative.

The dot-com boom, the fiber optic cable build out, was an enormous bubble, the famous dot-com bubble. internet also transform the world. My guess is that AI could absolutely fit into this category of something that is both bobbleicious and also transformative. And here's basically the way it would work according to Paul Kedroski. Paul basically says, look, AI is real, the adoption is real, the revenue is real, companies are paying for it, consumers are paying for it, that's all real. But this is not a question of if people are paying.

It's a question of scale. Yeah, exactly. The first fact you have to deal with with AI as a bubble is that $700 billion are being put into this thing every year by the hyperscalers. That's meta and alphabet and open AI and all those. Again, that's two Apollo programs a year. The Apollo program took 10 years. It requires an enormous amount of revenue in order to pay that back.

And basically, Paul's case is this. Right now, you have these companies spending hundreds of billions of dollars on GPUs, on these chips from Nvidia and others. And they're incredibly expensive. And right now, it seems like, or in Nepal, these chips lose their value after a few years. So you have to reinvest in the chips that go into these data centers. Well, if you have to keep spending $700 billion a year,

every year in a perpetuity, eventually that level of spending is going to eat into the revenue that's coming into your company. And yes, meta's rich now and Alphabet is rich now and Microsoft is rich now. But eventually, if you keep spending this much and the revenue isn't coming in, your operating income, your profit margin is going to go from here, which justifies, say, a $3 trillion evaluation to here, which justifies evaluation significantly less. That's a market correction. That's a pullback in infrastructure spending on AI. That's a re-evaluation of this entire space.

And that is a world where we would plausibly say AI is. an industrial bubble, something that even Jeff Bezos has said. That's a world where essentially we get in AI what we got in the internet. An enormous overbuilded out of this technology, followed by a drawback, followed by over time people making uses of the fiber optic cable in the case of the internet such that you were running YouTube on it and Netflix on it and all these other things on it. And the bubble ended up sowing the seeds of a transformation So there are two thoughts that I have as I listen to this. One is that the thing that is sawing off the branch that Silicon Valley has been sitting on, right, which is programming. Like the whole, the costume that it wears is the language model. And I do think that from the marketing of this, like the storytelling contest part of this, like the fact that you interact with it verbally gives it a sense of I'm talking to a genie. And I think there is something to that, just from the packaging of this that makes it feel special and different. So there is that part of it.

But then there's the other part of this, which, and I just want to get back to this because from this macro, the economy is relying upon this now. And not only is the economy buying into this, but America is suddenly in a position where we have to root for it because if it is, in fact, a thing that might collapse, our country's f***ed. And so the question I have is strategically, if you're, whether you're, you're, anthropic or Open AI or any of these other, you know, ostensible winners, are they still playing this the right way? Because maybe it is 60 Filipino guys sort of, again, doing that thing,

or maybe it's the real thing. But either way, we're all so far in on this that we will bail out the industry. Well, I think you're hitting on why so many people have such a knee-jerk aversion to this and not because of there's cultural aspects, politically-coded, whatever. The feeling that it's also inorganic. It's just so overwhelming and pervasive. You know, a lot of revolutionary technologies do start top down, but Derek, has anything felt more top down to you? And that's not even saying it's not useful, right? But like so many of the consumer-facing applications have been forced on us. And that's before you get to even the content side of this, the slop of it, the actual generative output that we see. The commercials are so stupid and

for applications that nobody actually wants or uses. And that kind of goes back to what I was saying. It feels like the actual utility of what's happening is kind of out of step with the public face of it in a way that it does feel like we are just being told, like Pablo said, like this has to work. You have to do this. Everyone's in it, guys, and it's too late to get out. And these are great questions.

I'm not a builder professionally. I'm a talker. And so I'm going to stick with my expertise here. I don't know if they're building it right, but I know for sure they're not talking about it. Well, to use your language, Pablo, what are they giving folks to root for? If the idea is this has to be built, it has to be built. And if it isn't built, the economy will crash. But also, if it is built, as Dario Amadea said, the CEO of Anthropic, tens of millions of people will lose their jobs. Exactly. Sam Altman has written essays that have said this technology is going to be

so transformative and so disruptive, we need to think about universal basic income because so many people are going to be thrown out of work permanently. What are they giving people to root for? It's a really strange form of marketing to predict that if your technology is successful, if the effect of that success will be a calamity for the United States. I almost thought like, I was like, what is the historical metaphor for this? What historical analogy? What is it? if Henry Ford in 1910, as he's getting his assembly lines off the ground and started the Model T, what if Henry Ford said, if this car thing takes off, more Americans will die of car accidents every decade than the total number of Americans who died in World War I? That'd be the strangest

way to advertise the success of your product. It happens to be true. 35,000 Americans die of a car action every year. You multiply that by 10, 350,000 Americans die every decade of car. accidents, that's, I think, more than died in World War I. So it would have been an accurate prediction, but why on God's green earth would you ever tell the American people that this is why they should root for the success of the Model T? It's an unbelievably strange marketing strategy. So I think you're absolutely right that beyond the economic debate that we've had for the last few minutes about the degree to which there's like a sharp takeoff here in usage and revenue, from a marketing standpoint, from a talking standpoint, I think it's a very, very good question

to ask, what are they giving people to root for if they're saying, if this fails, the economy collapses, and if it succeeds, your labor market collapses? What is the outcome to hope for? Even beyond the pessimistic worldview stuff, which obviously helps them in terms of like inflating their own valuation and sense of importance, they're marketing the dumbest, most useless aspects of it, right? Like, there are absolutely useful applications, varying degrees professionally. I would argue. to a lesser degree in people's personal lives. But the way most of these, the companies, when, you know, the average American is presented with

it is, hey, you dumb idiot, you don't know how to not book a reservation in the rain. Guess what? Our world-changing technology is going to come in and save you. And, like, I think that has led to some of the aversion that people feel is you are being presented with the stupidest possible use cases for this constantly. It just makes me think that their actual intended audience, their target demo for their marketing is not actually the American people, the customers. They're dealing with an administration that is all to incentivize to, of course, unchained and deregulate, or rather more accurately, never begin to regulate. The corporations that are standing to profit by virtue of a strategy that says, once we are so all sunk, we're all so sunk costed into this, there's no backing up the truck.

So all we got to do is just keep going forward and the administration will help us. And I don't think the strategy is wrong because big picture, they're getting exactly what they want whenever they are asking for it, which brings us to why it's so fascinating that Dario Amadeh, the one guy who seems to have a post-Trump view of his company's trajectory, maybe that is his market efficiency, is that I'm going to take a longer view that actually transcends whatever the f*** political climate we're in right now. But I also want to be mindful that I have not done. delivered on the debate that I've been teasing for like I know we actually even talked about what Derek and I was originally texting about we're there now because there is there is the big picture view that we've now described which I think helps set up the premise of like okay

Mina has been saying now over and over again when does this help me kind of yeah that's kind of actually where we are now so give us the backstory on what you guys have been again viciously viciously fighting about. The sheer number of Google image Tim Pawlenty photos that have been sent my way in the last 14 days is genuinely criminal. No, I mean, I was texting with Mina

and some other folks in the NFL analysis space who had been reaching out to me to basically say, Derek, I don't know how this technology makes my work better. And, you know, as I was talking to Mina, I presented Mina. I went through some, you know, pro football reference data and I basically showed her,

like this is how Claude, you know, could break down the history of passing seasons as made available by publicly available pro football reference data. And I mean, this basically basic response was like, this is incredibly basic and I can do all of this in 1.5 seconds with ESPN research. And that wasn't empty bragging because I asked me, you know, okay, hot shot, if you're so good, if you're so good at whatever, data analysis, You tell me what quarterbacks have the highest adjusted net yards per attempt passing against a slith shell defense in the last five years, and I want them ranked by total passing attempts. And like literally one point five seconds later, she not only texted me the answer, she had like the screenshot of what she had looked up. And there, behold, was exactly the data.

Sorry, it was the Mar Jackson in 2024, right? It was last year. I think so. Yeah, I think it was. And then I think Alan was, Alan 25 was also in the top five. In any case, she gave you the data 1.5 seconds. And I want to make an importantly and sincere point here, which is what I learned from that interaction with Mina.

And then I'd love to know what Mina learned that interaction other than that I was wrong. I want folks to imagine, like, a spectrum of legibility of data in the world. Like, there's some data requests that are, like, entirely ill. legible to humans or machines. My parents, I've talked about this a lot and why I'm so interested in the future of science. My parents would pass away from cancer. If you ask a large language model, look up drugable protein targets for pancreatic cancer. That data doesn't exist. There is no internet of human biology. The data doesn't exist. It's entirely illegible. And so large language

models are completely worthless. And that's not just my opinion. I just interviewed the CEO of Eli Lilly. He said large language models for these purposes are basically worthless for now. But then you take the other end of the spectrum of legibility. And that's what Mina has with ESPN research. She has access to so much data so quickly, the same way that a typical person has access to calculating what is six times three, that there's no use for AI in this circumstance because the data is too legible. But now there's like, thinking about like a goldilocks zone between sort of on the one hand, something totally opaque, like drugable, protein targets for pancreatic cancer and something way too legible, like something you can look

up on ESPN research on dashboards. That's something like economic data, BLS data, a ton of government data, which is famously, famously impossible to access quickly such that I have to send a request to an economist and tell him, go away for a week and then come back to me to tell me how has father time for children under two changed in the last 20 years, according to the American Time Use Survey. That's the perfect zone for artificial intelligence. That goldilocks zone of the database exists, but it's an enormous schlep to go through it. And that's what I learned from our interaction, is that because different people have wildly different jobs, some people are working in biomedicine, some people are working at ESPN doing football commentary, and some people are working in the middle

doing economic commentary. And this is the sweet spot for AI right now, and it's not at the other polls. And the important thing here, and then I'm done with Staradish Spiel, I think when people think about artificial intelligence as related to other technologies, it's so fucking weird.

Like, when you think about, like, what does a light bulb do? A light bulb turns on. That's what it does from you and me, Mina, Pablo, everyone listening. The light bulb turns on, and it's the same lumens and watts for everyone who pulls down on that light bulb cord. But AI is like,

the light bulb that when different people try to turn it on. For some people, the light bulb turns on dark. For some people, the light bulb is one million watts. And for some people, it's somewhere in the middle. And so because this technology is so weird and jagged and incredibly, exquisitely sensitive to the individual's prompt, it's very hard to explain what it does for everybody because there's no single universal answer that you can give. Yeah, I would say this. So the original question was like, is this actually useful for our jobs, right? And I, the Goldilocks zone, you're describing, there are applications even in my current job. There certainly would have more applications, I think, back when I was a writer. But even now, like I was at the Combine and a prospect

ran the fastest 40 time ever for a tight end. And it's quicker to me to open clod and say, what are the fastest 40 times ever by a tight end over six foot, whatever, than for me to try to use Google, which is, of course, useless now, to figure it out. So as like a super Google quick, it is absolutely useful. And I do think a lot of people experience that same utility in just trying to research things in their ordinary lives. But I think an important distinction, especially when you think about generative AI, is does this help me do my job versus does this help me do my job better or well? And that's where I think is another layer to all of this because my experience is that there's not only a limit in terms of actually. getting the answers, as we talked about, because of the access I have to databases.

But if I was to lean on it excessively, I would be bad at my job. And this is why when I look at data, when I have a question, and I'm like, let me see, you know, if I tweak this and this and this, or I want to see which running back in this scenario, whatever, the process of looking it up is as important for me to do my job well as the output. When I see the numbers and I start toggling the filters and I start ruling guys out and I go down rabbit paths and I add extra, you know, that is how I actually arrive at my insights, is not by entering a question and getting an answer. Similarly, when I was a writer, I could have used generative AI to spit out an outline and then write off that.

But the process of outlining, I feel, is what made my stories better and allowed me to work through my thoughts. And I think that is something that people are figuring out when they use these technologies is, yeah, if you want to just do a crappy job, it could do it. But most of us should want more than that, I think. And that's been my experience in playing around with these tools. Pablo, can I jump in there? Because I just love this point that Mina made, and I think it's like one of the most important points

that someone can make about this technology, which is this capacity for overuse of AI to lead to a kind of cognitive atrophy. The most memorable thing that I heard here is, I think it was a sub-sac essay, where someone was pointing out the distinction between a job, job and a gym. And they said, with a job, the point is to get the work done. But at a gym, the point is to lift the weight. You can't go to the gym and ask somebody else to, you know,

bench 135 and tell yourself that you sort of at bench 135 because you ask somebody else to do it. No, like, your muscles will atrophy day after day, week after week, if you turn going to the gym to going to the gym to ask somebody else to lift the weights. I think lots of people use AI to help them do their job to just get the job done faster. But, I mean, it's totally right. There's no question that especially we can already see this at the high school and college level. People are using this to lift the weight. They're using it to write the essays. They're using it to do the research. Journalists, I'm sure, using it to write their outlines. And that line between when is it a job and when is at the gym, it's not like there's a law. It's not like there's a formula. You have to kind of feel it for yourself.

Like, when am I leaning on this technology to do my job in a way that's keeping me from building the kind of cognitive muscle that's necessary to get better month after month? And when am I using it to just help the work that I'm doing become richer? I do think strategic thinking is going to be that much more of a premium skill, right? The question of how do I use this and how do I deploy it? and in a world in which everything can be instantaneous, what am I using that advantage to do? I mean, it reminds me when I was at the Sloan Conference last year, I got Daryl,

president of basketball operations at the 76ers and the co-founder of the Sloan Conference to basically admit, like, yeah, I have an LLM in the room, basically, when I make decisions. And immediately, it ecotectually went viral and everyone laughed at Daryl. Like, Daryl's asking chat GPT on how to, like, you know, fix Joe L&B, like, good luck. And on some level, legitimately funny.

But it's also just the question of, if you're in a room with advisors, do you want the sum total of the most cutting-edge AI technology to be there telling you also what it thinks? And so, Mina, one of the funny subplots, I think Mike Floreo had this at pro football talk, I think that's where I saw it, was there was a subtext at the combine this year in which scouts and front office members were actively wondering, is AI company? for us. And can we be replaced? And I think it speaks to the exact conversation that we just had about that. Have you guys seen that commercial, the co-pilot commercial that was like worthy?

It's like the bucks. And they're like, we're scouting NFL players and it narrows them down or whatever. That commercial is actually, so for those who don't know, it's about like it is actually a hypothetical version of what Pablo is describing, which is this team, the scouting department

is looking at a bunch of players and narrowing them down with like AI promise, basically. It's actually a perfect encapsulation of what you're describing because there is absolutely utility in having a tool that can go through data and sort players and sort plays and sort just you can find hidden gems through data. But then there's a moment when they're like, and we're looking for someone with leadership skills and it narrows down to a guy.

Right, AI is just Googling five articles written about leadership and they just happen to find someone. They don't know shit, right? But it's an excellent representation of like, there is absolutely applications, but they're all absolutely limitations. And I think, again, like the people like me just, like our hackles get up because the limitations seem to be alighted constantly in public representation. I'm so glad you said that. And I think it's important to go back to the very beginning.

Like what chat GPT? What does that stand for? Genitive pre-trained transformers. Pre-trained. On what? On the internet. So these machines are very good at mimicking human language. Why? Because they're trained on the universe of human language. They're very, if you ask it to, you know, write a sonnet in the style of William Shakespeare about the experience of waking up in the the morning, well, guess how many sonnets are on the internet and how many descriptions of waking up in the morning on the internet? It will do an A-plus-plus job. If you ask you a question about cellular function, you're shit out of luck. There's no internet of human anatomy. It doesn't exist. There's no internet of something as ineffable as, or maybe highly specific as like

the leadership qualities of middle linebackers going back in the last 35 years. That is a data set that doesn't perfectly exist. And so I think I'm really happy that we've landed on this spot, because I think it's very important for people to understand. I think about this sometimes in the language of shapes. It's very important for people to understand the shape of AI intelligence, which the additive that is often used to describe it is jagged. And that jaggedness explains the fact that it might be very good

at, say, researching the history of people, tight ends who've run under a 4.4, a 40-yard dash. It's not going to be very good at a lot of questions, where the corpus on which it's trained is basically bereft of information. So there's a shape to AI intelligence. But there is also, and this might be the more important piece, there's a shape to human work as well.

And understanding how those two puzzle pieces come together is going to be really important for using this technology in a way that is effective but also responsible. I think it's going to take a lot of reps, honestly, to go back to the gym metaphor. I think people are just going to have to use this stuff over and over again to feel out. like, where am I using this in a way that makes me more productive and where am I using it in a way

that offers the illusion of productivity, but I'm really just making myself dumber week after week. Yeah. My big fear is just that responsible and productive usage is just going to become so bifurcated by class and, like, running out of time, we haven't gotten to the education part. Like, you know, I can send my kids to schools where they're very thoughtful about this. But how many kids are going to be going to schools where they're not thoughtful or are going to be put in situations? where they're not being given those same considerations, and that's a big concern of mine as well. I think what we're landing on is the premise

that what we're being sold is not actually what we should want to buy. And like the test that I want to do here is like in 10 years from now, 15 years from now, how are we going to think of this? And one of the articulations of this, I want to land on this video, it's the philosopher Tim Dillon responding to Sam Altman. And he sort of translates it in a way that I feel like maybe this is, if we're not going to regulate this federally,

maybe we should just hear how maybe this could also be presented. Sam Alton. One of the things that is always unfair in this comparison is people talk about how much energy it takes to train an AI model relative to how much it costs a human to do one inference query. But it also takes a lot of energy to train a human. It takes like 20 years of life and all of the food you eat during that time before you get smart.

And not only that, it took like the very widespread evolution of the 100 billion people that have ever lived and learned not to get eaten by predators and learned how to figure out science and whatever to produce you, and then you took whatever you...

So here's what's going on. So in this quote AI arms race, we are both racing the U.S. and China to give birth to an ancient Sumerian super intelligent demon and your aunt Connie in Phoenix who's a risk

receptionist is f***. I just feel like the demon political rhetoric might be more effective than everything that we've said in this episode, unfortunately. I'm also concerned about how to calibrate that. I'm now really kicking myself for not having any ancient Sumerian references in any of my AI answers, because I think that always enlivens the answer. I mean, look, I think AI is an incredibly powerful and very likely transformative thing on the level of, let's say, the personal computer. we work with personal computers. They didn't displace us.

They didn't displace workers. The unemployment rate still under 5%, as it has been for the last few years where computer penetration has been higher than ever. It's very, very likely to me, it's very, very plausible to me that 10, 20 years from now, generative AI is something a little bit like Excel.

It's something a little bit like PowerPoint, something a little bit like the laptop. Now we're agreed, Derek. I should start with that. That's how I've been feeling this whole time, my guy. But it's this powerful thing that is like, I think I think he called it. It's like the home screen of the knowledge order. Wait, wait.

So it's powerful and it's everywhere. Hold on, what I'm saying. What you're saying is that this technology, this entire time, has been a lot like Peyton Manning. Maybe he doesn't have the arm to really throw all the way down the field. But strategically it can dink and dunk, pick apart defenses, and very safely deliver you to something like a winning season. People also say the same about a little politician named Tim Paul Nt. This has been Pablo Torre finds out a metal arc media production. And I'll talk to you next time.