Reid riffs on antitrust, AI coworkers, and Chinese manufacturing
22:15 Share

I'm Reid Hoffman. And I'm Aria Finger. We want to know what happens if, in the future, everything breaks humanity's way. With support from Stripe, we typically ask our guests for their outlook on the best possible future. But now, every other week, I get to ask Reid for his take. This is Possible. Possible.

So, Reid, there have been a lot of headlines lately on antitrust, and a federal judge ruled that Google had illegally monopolized specific online advertising markets. The DOJ is seeking the divestiture of Google Ads Manager and other assets to restore competition. No surprise, Google is fighting back.

So happy to hear your thoughts on this specific case. But also, more broadly, how will recent rulings reshape the digital landscape or influence the future of innovation and competition in tech? So not surprising. Well, actually, maybe surprising to a number of people since I'm on the Microsoft board and, you know, obviously have to have a certain care in how I talk about antitrust, competitors with Microsoft, etc., even though...

My actual thing is I'm a venture capitalist. I'm on the side of scale tech. I'm on the side of building new things. And so that's where my interest in society, my interest in economics, my interest in how do we create a better world, you know, kind of most align with scale tech. So startups going to scale. And so blocking out the possibility of scale is one of the places where I think antitrust legislation can be very good.

and important. Now, that being said, on the Google specific case, there's a couple of notes. So one, I think probably the most robust note is, you know, where high economics are being used to buy exclusive channels. That's probably a pretty good sign that something is being done to

lock in a monopoly or build a monopoly. The challenge in a lot of these monopoly cases is what do you take as the size of the market? For example, do you take the size of the market as the general search market, in which case you go, okay, Google is massively dominant, or do you take it digital advertising? And because if you do digital advertising, you include meta, you include a whole bunch of other things.

And so, you know, part of the reason why there is smart people arguing on both sides is you get to a artifact of what is the comparable market to determine this. I tend to look at it as

Well, where is it started? Where is it stopping potential scale up competition? And by the way, it's not all scale up competition because you say, hey, I would like to start a new desktop search company using the same techniques that Google did to build it. That's not clear that that's a benefit to society to try to squash Google enough to allow random startups or any large tech company to come in and compete.

Now, the last comment is one that I think is maybe the most unpopular, but I think it's important to track, which is, you know, as we move to a more of a multipolar world and, you know, the kind of classic thing is, for example, a U.S. tech industry, a Chinese tech industry, you know, TikTok, etc.,

And we say, well, only the U.S. is going to be doing monopolist remedies and the others aren't. You have to track this within national competition. And so part of the question is, is to say, is this part of a competition?

global resorting for competition. And that's actually, in fact, extremely important because while we definitely want, you know, the next generation of companies coming out of, you know, Silicon Valley where I'm here, I am right now, you know, kind of talking about this. But on the other hand, of course, if you say, well, the scale ones that have the scale benefit,

We're limiting ours, but we're not limiting China's. We're not limiting other prospective ones. That could be damaging, not just to, obviously, American industry and American prosperity and American society, but also, of course, damaging in terms of the balancing of the world. So you have to also pay attention to that. And I think one of the things that is too often not included in the considerations in these cases, actually, I'll say one other thing, which is

There's always some politics in this, even though the Google case went across different administrations. Yeah, started under Trump. Exactly. Started under Trump, continued under Biden, returned under Trump. Yep. But there's always some political considerations that are not necessarily red versus blue. There's also the, do I look like I have a win because I was fighting like I'm the anti-monopoly division, et cetera. So you always have to pay some attention to this. And so you tend to go after politics.

The targets that have more of a populist or press bent. I mean, the most obvious one for me, like if I was force ranking all the ones that I would consider would actually, in fact, be the Apple App Store. Right. Because you're like, OK, this is hugely locked in. You're not allowed to have other app stores. You're not allowed. Like, it's very highly controlled. Right. And there's some arguments around, hey, we've got to maintain enough security and so forth. But by the way, that's also part of how all of these things are. Right.

But they're also taking a cut of every transaction and it makes using Amazon mobile apps sort of unusable because you have to go off app to buy certain things. And so it's actually not good for consumers. Yeah. Well, and also, like, for example, OneTel that I started with Google is saying, hey, you're spending massive amounts of money to lock in an app.

you know, effectively exclusive position. Well, that's actually a tell. But another tell is you're charging 30 percent. Right. Right. Off the top to everybody and making a whole bunch of money from that. And it's like, OK, well, that's another tell. So those are kind of tells where you say, well, those are things that that that you should examine carefully, look at this and potentially consider doing, you know, remedies on. So I guess the overall thing is I think it's good to do these things to enable people

you know, scale competition. But we don't want to lose the kind of gems we have as an American society. Like, for example, probably most Americans don't realize that these companies, Google, Apple, etc.,

get over half the revenue externally. Mm-hmm. Outside the U.S. Yes. They're one of our, you know, massive trade benefit companies. And you're like, okay, that's important to us as a society. Doesn't mean we shouldn't do antitrust things, but it means that we don't want to, it's not just the kind of the, you know, hit hard with hammer as a approach. It's be careful about maintaining the

Right. Probably not the best. And to your point, like geopolitics does matter. And so we can take that into account when we're doing these things. But there are certain tells for the position. And so, you know, the Chinese market continues to embrace AI as an accelerator and Chinese tech companies are growing their global footprints. We've seen that BYD, which is China's leading electric vehicle manufacturer, is really

rapidly expanding its footprint in Europe, especially as Tesla really tumbles. And so, especially with this increasing tariff regime, there are going to be some real problems with U.S. auto manufacturers selling to Europe, which could mean that China gets a greater foothold there. What do you think that most people are still getting wrong in how they think about competition with China? Well, there's a couple things in terms of competition. So, for example,

you know, already implicit in what you said, one of the real damaging to American prosperity, American quality of life, both in purchasing of things and in jobs and everything else is actually in fact, trade partners matter. And so it's part of the reason why there's been different trading blocks and a U and a NAFTA and those trade blocks actually matter because being part of them gives the people in them advantages and edges against the people who are not. And so when you say, hey,

I'm just going to go apply terrorists to everyone. And, you know, and then the absurd thing is I'm going to apply terrorists to islands that have no people on them and penguins only. But, you know, I mean, that's just the incompetence part of the whole clown show. But when you start doing that, you're going, okay, I'm going to kind of declare trade war on everybody. Whereas what you'd want to be doing is saying, hey, I'm getting closer to my partners and allies and I'm competing with the people that I'm competing with. And so by saying, hey, we're going to declare war

trade aggression with Canada, trade aggression with Europe, the natural thing for Canada and Europe to do is say, great, we'll go trade with China. Thank you very much. And your so-called compete with China policy is literally a gift to China. And by the way, the BYD product is very good.

So I think this is something that is highly harming of American society that starts from the general prosperity of our society to our functioning of our industries to the prices and engagement of consumers in this. And

And I don't just mean consumers in, you know, wealthy cities. I mean, across the entire country. And so this is the thing that is kind of, call it most obviously wrong about thinking of competition in China. Another interesting thing you said was that the BYD product is very good. And I think some people's conception of the world is still sort of China 10 to 15 years ago, where it's like, oh, we're going to flood the market with cheap knockoff Chinese goods. Right.

Well, no, they're doing advanced manufacturing. This isn't just, you know, low-level T-shirts. Yeah. We have to think about the prosperity of our society. We're like, we want to return to manufacturing. That's great. By the way, China, which currently has the best, well, one of the best in a vector, massively important scale vector, one of the best manufacturing capabilities, societies, cultures in the world. Like when I...

go to Shenzhen or have gone to Shenzhen is the only place I've gotten the experience where someone coming to Silicon Valley must feel like, holy shit, like I am seeing part of the future in terms of speed and how people are operating. It's more manufacturing there. And I think that the thing that we don't realize is even though they have this advantage, they're going full on AI, robotics,

By the way, that's what we should be doing, too. Those will be the new manufacturing jobs of the future. You know, we'll be the ones working in robotic factories. That's, I think, really, really key. The Chinese know that even though they have an edge with all their human labor right now, they're building like BYD specifically is building is intensely roboticizing its thing. There's not only is going to have a high quality product, it's going to be able to produce it at half the cost.

of, you know, any other competitor. That reduction of cost is not because of, you know, we're going to claim that it's unfair competitive practices, but it's actually, in fact, well, we're just smarter about how we build it.

On this podcast, we like to focus on what's possible with AI because we know it's the key to the next era of growth. A truth well understood by Stripe, makers of Stripe Billing, the go-to monetization solution for AI companies. Stripe knows that when launching a new product, your revenue model can be just as important as the product itself.

In fact, every single one of the Forbes top 50 AI companies that has a product on the market today uses Stripe to monetize it. See what Stripe can do for your business at Stripe.com.

Switching gears a little bit to what I'll call the cost of being polite to AI. So OpenAI CEO Sam Altman recently admitted that the addition of words and phrases like please and thank you in users' interactions with ChatGPT come at a real cost.

He tweeted that these pleasantries contribute to tens of millions of dollars in electricity costs for OpenAI each year. And so there's one thing about the dollar costs, but a lot of people are really concerned about the environmental implications in particular. Altman said that expenditures like this, tens of millions of dollars well spent. So question for you, do you think it matters like philosophically how polite we are when we communicate with AI?

Well, yes, but maybe not for the reasons that people might reflexively think, which is more about like when we're interacting with AI, it also involves us. It also involves how we behave with ourselves, with other people, not just with devices. It's actually one of the things I was always worried about, how the initial Alexa home applications were designed.

creating bad training for children or even adults who are not paying attention, like stop or stop, stop, you know, now already do this. It's like, what? That's not the way we should be interacting with each other. It's not the way we should be thinking about it. We should be kind of generally more civil. You know, politeness is actually, in fact, I think a good thing and that's worth it, let alone the question around, well, what outputs do we get? Because by the way, you know, people who are

deep studies of the prompting, you know, part of, you know, why I released the earlier book, Impromptu, people who are deep studies this, notice they actually get different responses from please and thank you and so forth because in part, you know,

This is generalized from a trillion plus words of human communication. And you're prompting when you say please or not, you're telling it a little bit about what kind of interaction you want from it, you're having from it, et cetera. And so it's actually a useful part of the prompt too. Now, I think part of what Sam's talking about here is if you go, well, it was such a great conversation. Thank you so much. I really appreciated it. And you're not actually building something else.

then that's good for you as per my earlier comments. Yeah. And I think it's a good pattern to be in. But on the other hand, you're not getting anything out of it and the electricity is being spent. It's like leaving the light on for an hour. Totally. Maybe you're going to walk into the room and, you know, so it's like, you know, you could decide to be a little bit less cautious there, but I would err on the

the side of politeness. I feel like AI might be the new waiter test. It's like it used to be when you went on a date with someone and they were rude to the waiter. That was like the ultimate red flag. And so the new red flag will be like, ah, I really like them, but they were so rude to their AI. I just, I couldn't get past it. So maybe that will be in modern dating. We'll see. But you mentioned something, how

saying please and thank you actually could be a good form of prompting. You're going to get something better. So sort of analogous to that, do you think that OpenAI, Anthropic, Pi, et cetera, should they be training people on how to prompt better themselves? It's like a big conversation, of course. How do you get the best prompts? But should the frontier models be doing that for their users? I mean, it's always helpful to do it.

I do think learning how to prompt well is really key. I do think it's really critical for people to be learning how to do this prompting in good ways, because part of the whole AI amplifying humanity, the amplification intelligence, is the theory that we can, by us bringing something positive,

fun and interesting and unique and perspective and creativity and, you know, kind of adapting to using these tools at the table that actually, in fact, we are much stronger together than being replaced in the work. And this is an area of active debate about, you know, within the both general work community, but also the tech community as well. Where over time will that line of transformation versus replacement be? And

I don't know, no one really knows. The claim that you know that for sure,

other than there will be some replacement, is foolish. Because we know there'll be some replacement, like, for example, customer service jobs with, you know, Sierra and others doing this. But, well, like, for example, a hot debate, software engineers. Will software engineers be amplified? Which I think is actually, in fact, myself more likely the case. Or will they be replaced? Because we're definitely getting...

you know, kind of higher quality with the kind of chain of thought models and all the rest to things that could lead to better coding. Everyone's working in coding assistance. And I think what we want is we want the maximum probability chance of all the jobs that people want to do or have any affinity for doing are transformation jobs, not replacement jobs. Now, part of that is how we're building the technology. I'm not actually an advocate of limiting the power and scope of the technology because you go, okay,

Yeah, fine. You know, make the cars really slow so humans can outrun them. It's like, no, that doesn't really work as a strategy.

But the nudges to say, hey, A and B are both performance systems, but A allows a much better partnering with human capability to get a much better output. That's good for the individual, good for society, et cetera. And that gets all the way back to your question on, well, people have got to be learning to use the devices better. You've got to be learning to prompt better. And it's one of the reasons why I love Ethan Mullock's work.

It's part of the reason why, you know, I myself, like, you know, when I was on Dax Shepard's podcast, you know, Armchair Expert, I was like, okay, let me pull out my phone and let me show you how to prompt the phone to start doing this in ways that are useful to you because that's actually, in fact, part of how we all shape the future more collectively. Cool.

What would you say, though, just last week, the news from Dario, the CEO of Anthropic, was that he said that they'd be rolling out AI coworkers as soon as 2026. You think that's exaggeration? You think they're going to be doing parts of roles? Or do you think that's real, that we're going to have AI coworkers coming as soon as, you know, a year from now? Well, it depends. Coworker can mean a lot of different things. So,

In a sense, with a fuzzy definition of coworker, I think it could be a co-pilot. Yeah. Yes, exactly. It's like we have co-pilots today. Right. So now what he means, of course, is that and not a surprise, co-pilots will continue to improve. And what I think he also means is done like the, hey, I have I'm helping guide each step. I might be able because just like chain of thought thinking, you know, with the co-pilot.

one models and others is I send it out on a task, like a set of work, and it comes back with all the work done.

Right. Like it kind of it guided through it, changed his plan some, et cetera, et cetera. And that's obviously what we see being developed already now. But again, you say, well, that's a co-worker that's more robust than the current co-pilots. But it's also not a co-worker in that the, you know, hey, Ari, I've got this really cool project. Hey, I'll talk to you in two or three weeks about it. You know, and you're off assembling the resources, doing all this stuff. So it's like this is a whole continuum. So.

Will we be advancing the continuum? Guaranteed. Absolutely. Now, one of the things this gets to is there's a lot of dispute around, like part of what AGI kind of means intuitively is, is it a capable independent agent in the way that a human being is a capable independent agent? It has context awareness, can change its goal sets, can remake plans and triage based on new data, can defend itself. It's like, well, no, I...

I asked you to work on this book thing and you came back with this really interesting art project. Like, why is that? It's like, oh, no, no. But this is the reason we were talking about like what we were trying to accomplish. This suddenly turned into the really interesting thing. I changed my mind. I didn't listen to you. I like based on my own goals and what I know about going on. Right. You have some agency. Yeah. Yes. And so the question is, is where are we heading towards that?

And the answer is more, but how much more? And I tend to think that this is kind of two theories of kind of what is the next, call it five, 10, 20 years of agents look like. And 20 is like, you know, impossibly long and these things. Most people are trying to talk two years. But it's like, well, is it a progressing set of savants where it does amazing, amazing things? Right.

But part of the reason why you stay close to it is because occasionally it fucks up in like stunning ways that literally like if a human did it, you'd be like, what were you thinking? Right. Like what happened? I mean, as you often say, like predicting the future, even a year or two out right now with AI is pretty impossible. Which, by the way, just for everyone else, a lot of people's natural response is to go, oh, shit, you can't predict a year out? That's terrifying. Right.

Hence, all the super agency, you know, doomer, gloomer. But actually, in fact, it's super interesting and we can help shape it. That's what's great. Like, you know, navigate risk concern, but it's exciting as well. Absolutely. Possible is produced by Wonder Media Network. It's hosted by Ari Finger and me, Reid Hoffman. Our showrunner is Sean Young.

Possible is produced by Katie Sanders, Edie Allard, Sarah Schleid, Vanessa Handy, Aaliyah Yates, Paloma Moreno-Jimenez, and Malia Agudelo. Jenny Kaplan is our executive producer and editor. Special thanks to Surya Yalamanchili, Sayida Sepiyeva, Thanasi Dilos, Ian Ellis, Greg Viato, Parth Patil, and Ben Rellis.