A Busy - and Expensive - Summer for AI, with NYT's Mike Isaac

22:14Peter Kafka:And again, they are selling subscriptions right now that people are using. Far away. The ones that are selling them. It's not a profitable business, but I still would argue they're the most successful in consumer. I was talking about this last night, actually, in an Oakland library event. So it was very fun. But like the whole that's the question that is on like average person's mind, which is like these are burning cash or whatever. How do they do it? The funny thing is, back in the day, the exit was always, OK, Google or Facebook will buy you or whatever. And like, that's fine. I think these some of these startups startups are too big to choke down by any of these companies, not necessarily that they'd even want to.

22:56So they're doing their own thing. So like they have to get to profitability, I think, faster because to your question of where is that spent? So much of it is compute. Like people don't really understand how much it costs to build what's called computing power, compute, to train these models to what's called inference and like have them actually running.

23:22Peter Kafka:And we keep hearing about the chips. Is that the primary thing, right? Renting those. These GPUs that NVIDIA sells. Is that the primary cost plus the associated machinery around that? Yeah. OpenAI. So that's the thing. OpenAI doesn't really own. What they have to do is pay Microsoft, pay Oracle, pay these other cloud providers for use of their GPUs. The GPUs, if people don't know, short supply. I, NVIDIA has cornered the market on it, and they are the things that power AI computations the best. So it's literally, it'd be like, we own a factory, but we can't operate the factory without electricity and basic equipment, which we don't own.

24:05Peter Kafka:Yes. And the more we make, the more we have to spend. Yes. But at some point, right, we've been taught by the internet or by tech, right? At some point, they're supposed to clear those over-advancing costs and then mint money. So what is the inflection point theoretically for that? I mean, look, they have to the can being kicked down the road every time. And I think you got probably a flavor of this in the in the briefing was like, we're getting more efficient. It's going to cost come down. These are scaling laws of computing over time as tech improves, like the cost will come down. I don't think we're anywhere near that right now.

24:42Like I really and I think the the entities that are at the risk the most around that are open AI and Anthropic and startups whose entire business model is this one thing. Right. You can be meta and Google and throwing off cash like crazy every quarter and just sink it back into this money.

25:03Peter Kafka:Which the meta and Google people are happy to point out. They're saying we don't you'll notice we're not out there fundraising. Right. We have giant money machines. Yes, it's true. I mean, that's true. And we can put that money into this. Yes. And so far, they have not been penalized by Wall Street because, I mean, look, Meta just had like a blockbuster earnings quarter the other week, right? Like they are. And the thing, the point that they make is like, look, you know, we make all these AI things and whatever, and that's future facing. But it's improving ad targeting for us. It's making Reels recommendations better.

25:35It's keeping people on the platform longer, which I think is a real thing, too. So you can more abstractly say they're benefiting.

25:41Peter Kafka:But also who cares, right? Like as long as our money machine is running. Yes. That's what you care about. Correct. That is growing still. Correct. And we're going to, you know, try to leapfrog with this next tech. But don't worry about it. We're still running the money machine. I mean, it bothers me in the sense that they don't ever really get penalized by Wall Street because of that exact thing. Like they have spent more than$100 billion on the metaverse. And they are niche products at best, the hardware. And the software is non-existent as far as like people messing around in the metaverse. So I don't know.

26:11Peter Kafka:You know, there's plenty of analysts asking the same questions. And then there were some questions, right? For a minute, yeah. Especially with Metaverse. And the first, I think last year, when everyone was starting to throw around a lot of money at AI, there was like, how are you going to make money with it? And then they just said, shut up. We're still making money. And everything was fine. But one last parallel to the old days that you and I used to cover. But there, the plan was spend a lot of money, own all the market share, and there's going to be one winner, right? And so, again, that's why you kind of don't need to worry about the money we're losing now because one day we'll own the whole market, right?

26:47Peter Kafka:We'll own search. We'll own, you know, ride share. We'll own user-generated video. We'll own social network. Whoever wins that wins everything. Do people imagine that is the same dynamic with AI where it's going to be either open AI or Google or Meta or one of those companies owns it and everyone else ends up using it? Or is there where there's four or five or a dozen of these things? I think so. Me speculating. But like, I think one of the biggest stories out there right now is Google because of what I said earlier, which is like this could be the death of traditional search. Like, it really depends on how people navigate.

27:28What these chatbots are essentially a new way to navigate the Internet, right? Like, and I think it's a generational thing because, you know, I'm 40, you're probably 70. I think it's like one of those things where we don't maybe experience this way of navigating the Internet as much because you and I grew up on AltaVista, Yahoo Search, the portal era to Google. Now we're trained on Google, even though Google Search kind of fucking sucks. And now there's younger folks who are finding this way into search through chat GPT, perplexity, Claude even.

28:01Peter Kafka:Yeah, no, I'm not me. Yeah. Listen, listen, whippersnappers. I actually I I periodically am like, is is is chat GPT like a like an intern that doesn't know anything, but is trying to be helpful or is it better Google? I can have used both of those things in my daily life and I'm frequently using it as just better Google. Right. And that's that's the bet. And so to your point about share, I think Google's freaking out because they sort of see this as like the entry point or whatever to the next thing. I think it I do think actually something becomes a winner take most in this market. Probably there's probably going to have to be some differentiation on like what the chatbots look like, how you're going to incorporate it into your life.

28:48I think Meta's whole thing. I think Meta is kind of a mess right now, but their whole thing is like, we want to make it personal and this is going to be your companion. and blah blah blah but at the end of the day like people kind of get habituated to different forms of search and navigation of the internet and i feel like open ai is the one to beat in that regard and like all of these other players are using the weapons that they have which all day long are scale and money and they can jam their still nascently used products into every sort of surface they have and set money on fire trying to get it going.

29:28Distribution, right?

29:28Peter Kafka:Correct. All these, the mega, primarily Google and Apple, but in theory, Amazon, sorry, not Google, not Apple, primarily Google and Meta. I mean, Apple too. Apple should be able to do this if they wanted to pursue it, right? So you use our stuff. That stuff is now baked in with AI. We own it or we're licensing or whatever, and that's how we're using it. And that's also one of the arguments about why you don't need a winner take all is it doesn't matter whether meta has the best AI. If you're still using meta's core thing and the AI is in there, meta doesn't believe that. Yeah, yeah. It leads me to the next thing because you have been writing, along with a bunch of others, but you've been writing about these talent wars.

30:05Peter Kafka:This is where they're also spending money. People are vaguely aware of this, right? You're just making gazillions of dollars. It's not a new thing to be, to one, have giant tech companies throwing money at engineers. That has gone on forever or to retain them. And in AI, I know that has gone on for a while, right? Especially a while back when people were very interested in self-driving cars. Like anyone who could do anything with that field, which is a lot of AI. Basically, Uber was essentially trying to buy Carnegie Mellon University at one point. Oh, my God. I remember that. So they were spending a ton of money.

30:38Peter Kafka:But now that it's all supersized, you had a lead in a story you wrote recently about Mark Zuckerberg personally trying to recruit a guy not from OpenAI. He was a startup guy. And he said, I'll give you 100 million bucks to come work for me. And the guy said, no, thanks. And then Zuckerberg came back and eventually offered him$250 million over four years. These are, as you pointed out, NBA offers. And the guy took it. So is all of the spending, is this primarily meta saying there's an arms race, but we're just going to go over the top? We're going to 10x everything. we're going to intentionally price out other people who can't or won't spend this much.

31:20Peter Kafka:Yes. Or is every, but there's other stories, right? Oh, you know, these, these giant aqua hires. Yes. Google is buying a company for a handful of people. Yes.

31:32Peter Kafka:It's, what is going on here? How long can it last? So, yeah, no, you're totally right. It's not just meta. It's everyone. And it's a lot of different things. I think a few things are going on. One, I'm talking to executives across the industry who are pissed off at Mark Zuckerberg because he has fucked up the entire market for this stuff, right? Intentionally, though, right? Oh, 100%. Like, he has, like, this is, again, like, one of his weapons. He has a checkbook. And, like, this is something that's painful for OpenAI, I think, because they can't, like we just covered, they're raising constantly.

32:04They can't shell out a quarter of a billion dollars for one dude, you know, multiply that over and over by whatever. So I think everyone's very mad at Mark for doing that. Folks I've talked to internally who also openly admit this is the most insane people who have been there for decades, the most insane talent environment they've ever dealt with. You know, Mark was used to like going up to Tahoe 10 years ago and like romancing a AI researcher with like a million dollars, which was which was and still is a lot of money at the time. But like now it's just ridiculous. I think the other dynamics I would say that people don't really think about is.

32:46To your point, AI has been a field, a competitive field for talent for a long period of time. Some people have been inside a deep mine, for instance, for a while and they got hired or even OpenAI or really deep mind and like some of the or Microsoft. And they got hired being paid a good salary, but certainly not what Mark Zuckerberg is offering right now. And they also got hired inside of Google, which is this giant bureaucratic mess. London is the power center for deep mind because that's where it's all based. So like if you're getting paid only a million dollars and you're a deep mind is Google's Google's Google's AI thing, although they have a Google brain, which is competing.

33:28And like this whole. Yeah, every one of these is kind of messy. But if you're there and you're like, hey, Joe Schmo just got paid 10 million dollars in a day to go create like whatever is Mark Zuckerberg has on his mind. people are getting antsy and i think that's part of it the other thing you alluded to is these weird um their acquisitions in all but name basically and like the companies know it too and they're so full of shit when they deny it but it's like them like saying we're making a 14 billion dollar investment in this company and taking a minority stake and rating all of the top talent but the

34:03Peter Kafka:company is still standalone and like we're and by the way when this we're talking we're talking now about scale scale.io and by the way we know that when we buy this company or take a minority stake in it all of its other business goes away because no one none of because it makes its money selling this services to all our competitors who will no longer want to use it once we own it so we are essentially shutting it down it's it's crazy yeah these companies protest almost too much but uh So do people see an end to this or do they say this is going to go on basically until we get this until there is a winner?

34:39Peter Kafka:We're just going to keep ratcheting up the prices. Yeah, I think the rationale that I've heard is so one important distinction is a lot of the talent they're approaching is is what's called like people who are working on, quote unquote, frontier research, meaning like small areas. Correct. Stuff that was not where most of the people in the field were looking at and understanding. a 22 year old let's say got his PhD at MIT in an area of AI that no one cared about at a time and suddenly he the talent of the pool of talent for that is very small and he suddenly becomes very desirable right so the rationale that I've been told is like yes we overpay for x amount of years they are thinking maybe two or three years it's taken that long for the PhD field the current AI PhD field, which is, by the way, like so many people going into that over the past few years because they've seen the talent pool for AI slowly expand.

35:39They think it's going to even out to that point in a few years, like the market will become bigger. There will be just market dynamics. There will be more people to do it. You can pay less because there's less insane competition. I don't know if that's going to bear out, but that's the calculus.

35:54Peter Kafka:And what do people make of the fact that So much of this is coming directly from Meta and Mark Zuckerberg, who basically, again, as you've reported, concluded this spring after saying, we've got this great idea and we're doing open source and Lama and it's the best. Basically saying more or less. Actually, none of that really worked. It's crazy. We thought it was. We need an entirely new set of people. And that is why we have to go spend 15 billion dollars on scale.io to hire basically one guy who's not even really a technical guy. he's going to be like sort of the marketing guy for all this. Like I would, you know, there's, it seems like you could make an argument saying, wait a minute, you've been at this for so long.

36:33Peter Kafka:You're, you're such a big brain genius. And you stood up three years ago and told us the metaverse is the future. Now you're saying it's super intelligent and you need to spend all this money. Maybe, maybe we should hold you accountable for the fact that it shouldn't take, you shouldn't be scrambling in 2025 to acquire this town yes i mean yes absolutely i think um again to the reasons that we just talked about like they're not going to penalize them until they start seeing uh until their money stops printing like it is you know remember a few years ago when they had a shitty quarter in their stock tank like by like half or whatever like that is how twitchy that was because yeah it was a shitty quarter also because they said this is because apple has changed the way ads work on the internet so it seemed like this might be the existential and they actually solve that problem they fix it yeah that's a whole other really interesting thing but like that's the point i think i keep in my mind a lot which is like you're golden until you're not you know and like i think mark can get away with this sort of sleight of hand every few years saying here's the next big thing and then like when it doesn't work i mean look before metaverse it was Libra and crypto.

37:45And that clearly went nowhere. They they abandoned that. There was NFT sort of thoughts inside. There was like he has a history. I think my criticism of Mark would be he's like he has a history of being a little ADD and like chasing the next hot thing that's kind of going around. To his credit, he's the AI thing is like parts of it, I'd say, are real enough. And everyone does realize that they have to sort of figure some version of this out. But yeah, what happened was like six months ago, essentially, Llama, which is their open source sort of strategy to build AI, hit a wall in its development.

38:22Like he kept promising we're getting bigger and better. And for a while it was kind of working. Like there was some pretty impressive versions of Llama and everyone would get it for free and he was flooding the market. And that, just for people who don't know, was very different than OpenAI, which is a quote unquote closed model. And they weren't like a lot of people have it. But like they just hit a wall. They couldn't get better versions of Llama and they had to pivot very hard. And that's why I would argue he's scrambling and opening the purse, the pocketbook to like make up for it and get the talent that he feels like he didn't have basically.

38:56Peter Kafka:And the pro the pro argument is besides he's making a lot of money, he has pivoted successfully before. Sure. And also this is like if you believe all of this, if you believe that this really is coming. And by the way, like someone's someone's sharing a thing on Twitter today showing that that chat GPT can't do this really basic math problem. Yeah. You should follow the guy Colin Fraser. Yeah. He's very good because he just puts it through the it's trappings every time. Yeah. And it's still really wild. It's very funny. Ask it to spell strawberry. Or no, it's something about R's in strawberry.

39:33Peter Kafka:But presumably, if we all believe that this is the future and we're going to get there, then it is worth basically anything to make sure that you either win or are not left behind. Yeah. And it's existential. But I also just the caveat being is that I really do feel like everyone talks about it as inevitable and I just don't feel comfortable talking about it as inevitable. And I feel like that needs to be remembered when we're talking about it. We'll be right back. But first, a word from a sponsor.

40:07Peter Kafka:And we're back. I've saved the sexiest part for last. Obviously, we talked about the fact that we're not wearing pants. You talked about me already. So I don't know if we can get the second sexiest. Mike, I want to talk to you about U.S. tech policy. Oh, fuck you. Boing. All right. Why did you have me on for that? I didn't tell you I was going to ask about this. All right. It can be a quick section. So Donald J. Trump, President of the United States, had a big event a week or two ago, unveiled his new U.S. tech policy. I couldn't make a lot of heads or tails out of it. I know lots of some people in policy world are excited about it.

40:47Peter Kafka:Mm hmm. Can you can you try to sum up in a sentence or two what it means? Why it matters? No, I think. Yeah, I think put simply not they didn't get everything they want because ideally they would have. Sorry, the tech Silicon Valley, let's say, or let's say pro AI people who want no strictures on how they develop it. This is which is a which is a big switch, but it's also a big switch over the last couple of years. Right. Yes, correct. Because prior to this last election, if you talk to AI people and you asked about things like regulation, they would all say regulation is very important. We're very excited to be regulated.

41:26Peter Kafka:We're working with Washington and all the stakeholders to become regulated. We are very excited to be part of the framework. And then the criticism was, oh, you just want that so you can regulatory rules you want. But they had all sort of learned the lesson of social networking, which is the government's going to be involved in your stuff and they're going to want to. So you should work with them in some way or at least say that Donald Trump gets elected, brings in a cohort of people who say not only should there be no regulations, we shouldn't have trust and safety. We should we should there.

41:56Peter Kafka:These things should spit out whatever they want. And if there's bad consequences, you know, you have to break some eggs. That's and like this is my so I my take on this is that everyone was talking. Everyone seems to be talking to an audience of one depending on the presidency. First, it was Biden. And to be clear, they had a bad when I say they I mean all the tech CEOs had a bad relationship with Biden. I got so much in my ear about how much they hated him, how much they hated the administration, because there was like really nothing but disdain for Facebook, for Google, like from the Biden administration.

42:34I think that was.

42:34Peter Kafka:And also they threw, even though Lena Khan and the antitrust stuff didn't really work to stop most big mergers, it did basically choke off the environment for M &A in the Valley. And so there are a lot of people who are not running Meta and Google who are still very upset about that because if they're a VC, you can't get your company bought. If you work at those companies, you can't get it. Like, it was a really important thing for them to be able to buy and sell companies. And that went away for three or four years. Although Lena Kahn vindicated with Figma, I guess. I don't know if that's the meme.

43:09I don't know if that's entirely true. We'll have that for a separate argument. I was going to say, do that. But no, you're exactly right. Like, the joke in the day was, oh, you don't need a business. You just need to sell the Meta or Google or Facebook or Google eventually, right? I think under, so like the Biden era is over. I think the one caveat I would say under Trump is that I don't know how much in lockstep the Trump presidency is with the folks at the FTC and the DOJ who are heading it because there is some ideology there. And let's say Josh Hawley, like there are people or Marsha Blackburn, for instance, there are people who hate big tech who really do like feel.

43:50I think it's probably more motivated by what they call like, you know, wokeism and like content moderation stuff. But like people like Hawley are really true believers in terms of hating big tech.

44:02Peter Kafka:I would argue. Yes. But there's asterisks with that, too. Right. There's this guy who's a lawyer activist, MAGA activist, guy named Mike Davis, who's in Trump's ear when it comes to him saying like, you should not, you know, don't let Google or Meta get away from these lawsuits that they're asking you to stop. you know we need to keep going after that need to keep going after that then and we're going to get really in the weeds here but there was a hard to follow story that broke last week where basically two people from one of the antitrust wings of uh i think it was doj i'm going to mess up some details they left it was a whole mystery you couldn't really understand and what turns out they were complaining they were reviewing a hewlett-packard deal that basically they thought shouldn't go forward.

44:46Peter Kafka:And it turns out that one of the parties in the deal hired this guy, Mike Davis, to advocate for them. They're basically saying this is a Trump friendly guy who is making the deal go forward. You know, he's he's coloring the whole thing. And then, you know, basically the conclusion is. What matters is, does Trump like you or your representatives, not is there competition? That's totally secondary. So there's no ideology except does Trump like it. No, I agree. I agree. Now we've gone very far. But I agree with that. And I think that's the so like my point of view would be tech is kind of getting everything they wanted in that regard.

45:27Like if anything, I feel like Trump presidency has been a great presidency for them. Honestly, they they get to lay people off. They get to do these like fake acquisitions that haven't been challenged yet that i've seen you know they get to have few strictures on ai compared to what uh they had to sort of you know telegraph to buy tariffs if they give trump a golden glass bobble yeah what what was that it was it was the isr run god that's the most tech bullshit ever but you're right like that's like that literally i mean i kind of admire someone who sort of acts like an old school real estate guy.

46:08Like this is how you play the game, right? And if you come in, or at least openly, like there are versions of this, you could argue under democratic regimes where like you get the right lobbyists, they're talking to the right people and similar stuff happens. But like - That happens everywhere here,

46:21Peter Kafka:but here you have to do a whole public demonstration plus money. That's a whole thing. But back to the tech policy, is there anything meaningful in that policy that we should pay attention to? And assuming that we have elections in the future and the pendulum swings and you can get someone who's more centrist or left in the White House. Is this stuff that can be undone? These aren't laws. These are just sort of a framework for things the U.S. believes in. What are the practical consequences of that thing being unveiled? I think the content moderation stuff and, like, elimination of bias stuff is something really to think about and closely watch because what you're essentially telling the meta, let's say meta who's like Joel Kaplan, their policy head guy is on Twitter basically, like agreeing with all the Republicans all the time and saying, yeah, we're going to do this.

47:07We're going to eliminate bias. Like training models like that has lasting repercussions, right? Like if you train the meta AI.

47:14Peter Kafka:The policy is have neutral, you know, don't have bias. Then you have a whole debate about biases. But these aren't laws. Right. So what do you have a sense of like. When does law matter? But people, well, there's that. And there also is, it doesn't need to be a law for it to be influential and also if it's influential for three years let's say there's a different president in three years that's a lifetime in ai like you could really yes you could really make consequential moves because of this even though there is not literally a law i mean there's there's a this morning as i was staring dutifully at my crappy twitter i saw kaplan do a joint announcement with some, you know, MAGA influencer guy.

48:00They reached a settlement.

48:01Peter Kafka:His name is Robbie Starbuck. Okay, can you tell me who this person is? I had to Google it because I was just saying that the Wall Street Journal headline says Robbie Starbuck and Meta reach an agreement. I'm like, is Robbie Starbuck a company? Like, what are you talking about? That is the first barista. He's a, and he may make a fine latte. Apparently he's made a name for himself railing at companies for having DEI. And that's it works, though. You have the head of policy on Twitter saying, releasing this joint statement saying, yeah, our bad. We're going to figure it out. We're going to work with you, Robbie Starbuck, to eliminate bias.

48:39Peter Kafka:They literally say that. All you do is like yell loud enough on Twitter that you had like, you know, that that meta AI told you to dye your hair blue and suddenly like the woke left is ruling tech. Right. But again, you don't need a tech policy for that. You just need to be able to work a system that gets into Donald Trump's ear. And that's the system. That's the system we're living in. Because the one thing I'd heard that the tech companies wanted in this policy and didn't want was something in effect of there is no copyright anymore. You can go ahead and ingest whatever you want for these LLMs and the New York Times where you work should stop suing Microsoft and OpenAI, etc that's all done that wasn't in the policy which they were disappointed about but the same day they unveiled the policy trump basically said that on stage like also there should be no copyright stop suing these people so to me it kind of underlines the idea that like you kind of don't need a policy you just need a president who is willing to insert himself into this stuff i mean that that is trump as always is the wild card here like maybe tomorrow he'll tweet fire the intel CEO and also this other CEO or whatever, you know, like I think that I agree with you.

49:53Like a lot of this is theater that's played out in the media and just kind of how it works. And my suspicion is like, look, you can't stop media companies from suing. I mean, ostensibly, you can't stop media companies from suing the companies themselves. It goes to a judge. There is some sort of lawful process. Who knows if Trump gets involved in that? So if it's not in policy, you know, the media companies can take these tech companies to court. But with the sort of tacit endorsement or even overt endorsement of the president saying, hey, do whatever the fuck you want, like, are they going to start pushing the line even more?

50:30My guess would be they were already pushing it forward. So my guess would be yes. And then I think they see legal costs as a line item or as a line item in their expenses, essentially, as they continue to just hoover up a lot of this stuff.

50:45Peter Kafka:That would be my guess. Like, Isaac, you rock. When you're not writing about meta and AI, you write about crude metal bands. I've never heard of them before. Oh, I'm doing another profile next week, actually. Can you tell us who it's about? I can't tell you, but I am going to L.A. if that helps. Is there competitive pressure on this metal band story? This is the new one. Actually, this is a pretty big band. I think you will have actually heard of them. Okay. Look for that story and many others from Mike Isaac coming to you via the failing New York Times. Thanks, Mike. There you go. Thanks, buddy.

51:17Peter Kafka:Thanks to Mike. Love that dude. Thanks again to Charlotte Silver, who produces and edits this show. Thanks to our advertisers, and thanks to you folks for listening. See you next week.

51:32Thank you.