Reid Hoffman on what's real and what's hype in AI | Pioneers of AI

When you think about OpenAI and Anthropic, we tend to want to tell these stories as cage matches. Well, one of them's going to win and the other one's not. And you're like, well, actually in fact, there's a lot of room for both of them to win, incredibly. If you believe that AI has the impact of applying intelligence with the scale and price of electricity across everything, these are going to be two of the major providers of that.

Think of it as a little bit like internet valuations. Some of them will have turned out to be insane and go to zero, and some of them will have turned out to be way too low and have gone up a lot higher. The trick is, which ones?

My absolute pleasure.

Likewise.

Actually, I just said I wouldn't stand for re-election, which means I'm on it through the end of the year, but yes.

You know, when Ujjwal and I had been doing the work in the last couple months and we had been beginning to see some small molecule proposals from our AI drug discovery engine that our computational chemists are looking at and going, 'Oh my god, that's really interesting.' And that could work. And I said, 'Okay, I've really got to make sure that this thing is navigating and building these things really matters.' And so, I got to focus on this and talked to Satya and said, 'Look, it's been a long time. I'm always a friend and an ally. We'll continue to help even afterwards, but I'd really like to spend my time more on building, on being a founder than essentially being a governance person.' And I can still help when, as a matter of fact, Satya and I were just on the phone today talking about some piece of strategy.

It's been a couple things. One of the things is obviously the LinkedIn acquisition is one of the epic M&A of history in terms of growth and impact and revenue, having high operating margin, etc. Also continuing to be successful, right? LinkedIn is thriving still, and that's not to be taken for granted. And then, helping navigate the organization to buy GitHub, which was bought well over revenue in terms of comparable prices, but the strategic part of it and how that plays into the whole coding revolution now is really key. Facilitating conversations of trust between OpenAI and Microsoft. Those are some fun highlights.

There's a sense in which SpaceX talking about itself as an AI company is somewhat honest by the fact that we bought xAI. And what we're going to do is use our market cap to try to buy our way into being an AI company. It's essentially what it's trying to do. And so it's not surprising that, X days after the IPO, we're using our market cap to buy Cursor. And I wouldn't be surprised if there were others as well because SpaceX isn't an AI company. xAI is, as Elon himself has described, a complete train wreck for its building of foundational models and other kinds of things. All the founders have left. It's on its third restart. And if you look at the revenue in SpaceX, it's 'oh, we're going to be charging Anthropic a lot for compute.' Like, 'oh, you're a premium CoreWeave. I get it.' Which is not an AI company. The reason why they're saying they're an AI company is because they're going to go buy a bunch to try to become an AI company. I think what investors are buying or holding on in this was that they're going to go spend a lot of equity capital on buying a bunch of AI assets and seeing if they can be cobbled together. You could almost think of it as the IAC of AI in terms of how it's playing. So, I'd say it's TBD, watch this space, but use the market cap to buy AI companies and try to buy your way into relevance.

I'm sure the two companies do, right? There's definitely a lot of rivalry between the companies. And there's the theory of who gets out first and do you absorb the capital? I don't think it matters. I actually think there's a lot of room for both. I think they naturally have categories that they're very strong in. With everything that's going on with code and its applications from Anthropic, and Anthropic's clearly testing the waters on expanding in design and legal, and OpenAI with ChatGPT is clearly the front-end search Google equivalent. And it's also testing the waters on other things and is coming up strong on Codex, which is actually insufficiently talked about as a really compelling coding product. In fact, it's stunning how much cloud code and Codex are. One of the questions in the SpaceX acquisition of Cursor is that Cursor seems to have had its bright star some number of months ago and seems to be fading over the horizon. So what the theory is that it doesn't completely fade over the horizon is an interesting one. We tend to want to tell these stories as cage matches. Well, one of them's going to win and the other one's not. And you're like, well actually in fact there's a lot of room for both of them to win, incredibly.

Think of it as a little bit like internet valuations. Some of them will have turned out to be insane and go to zero and some of them will have turned out to be way too low and have gone up a lot higher. And this is what happens when you have high growth, hot, fast moving tech companies, of which AI is most of it. Look at what Google became from being an internet company, and then other things bombed for being too early. And so I think there'll be a number of these valuations that'll just be like, 'oh yeah, you pay a lot for that and that turned into zero.' But I think there will be others. For example, when you think about OpenAI and therapeutics, why should it be valued in the same trillion-dollar company category as these others? The answer is, if you believe that AI has the impact of applying intelligence with the scale and price of electricity across everything, and these are going to be two of the major providers of that. And what's more, you can already see some of the revenue really well. And frequently the revenue comes later. Like in the OpenAI category, advertising. For example, Google's early theory of revenue was we're going to sell enterprise servers. And that didn't work. And then AdWords, the best business model invented in human history. If you say all the valuations are crazy, you're wrong. If you said some of the valuations are crazy, you're right. And the trick is which ones.

Well, we're definitely subsidizing it for growing. A modern exemplar, speaking of trillion-dollar companies, is Amazon, which was unprofitable and barely profitable for a very long time. And yet has an amazing strategic position and continues to be that. So, it doesn't matter if they're profitable at IPO. It does matter that companies become profitable and then eventually become seriously profitable. But part of valuations, which most people don't appreciate, is future expected terminal value. So, you go, well, this thing is so strategically important, it's going to exist for a very long time and continue to grow and have an important market position even if its revenues are modest or its profitability is wobbly. That still makes a very valuable company.

Part of the thing is, is there some possibility? Because if you said there's a 3% possibility it becomes an AdWords, that's huge.

The change of defensibility and the change of what kinds of things lead to sustaining value. People tend to be overly dramatic and say it all goes out the window. That is certainly not the case. But on the other hand, a bunch of things that used to be strong defensive modes don't. Part of the reason why people were talking about the SaaS apocalypse is because they went, well, the previous dynamic with SaaS companies was it cost a billion dollars to build the baseline tech that everyone would use. You chip away and eventually get an ecosystem and customer base, and then challengers, it's very expensive to come because you're continuing to grow your customer base while they have to spend a billion dollars and then get the first couple customers and move on. And then you go, well, if all of a sudden it doesn't cost a billion dollars to make one of these things, the product feature set of what's useful to them may change radically because individual companies may want things deeply personalized in ways that they can do it with the coding agents themselves. So the economic model, which was have a high operating margin because switching is very expensive, it's like, oh, I don't want to pay you that operating margin anymore and competitors can come in. Hence, SaaS apocalypse. Now, the reason why the SaaS apocalypse is overly stated and it plays the defensive moat question that you have. The real question is not short all SaaS, the real question is short any SaaS that's not aggressive and driven and committed to becoming AI native, and buy the SaaS that is, as a direction. That then gives you some code also what moats might be. There's a bunch of standard defenses that I think will continue in the current age. Now, the interesting question is which things will add in. And I wouldn't be surprised if there were new kinds of moats. People have speculated, will there be data moats? Maybe, especially on things that are real-time oriented or you can only serve the product if you have this particular data versus training data moats, which I tend to think I'll be a little bit more skeptical of. How do network effects play in? I think there's a set of different questions there. There's the kind of in some senses brand and trust matters a whole lot more. So, if you've established a brand and trust and gotten through the noise, how does that matter? So, I think there's a stack of things that matter there. I do think one of the things that you and I love about both being founders and investors is that hey, when you're throwing up all the cards in the air, it takes taking intelligent probability risk-adjusted bets, working with speed, making it happen, and that's fun.

Well, I think you would ask a couple questions. Because if it's just like your company is a thin wrapper on a model, it's like you're basically just waiting around until the model company decides to go first party. And they may still allow you to run and all the rest and maybe you'll switch to 'oh I'll be using the Chinese open source.' Kimi is a way of doing this which is kind of what Cursor ended up doing off cloud code. But it's not as good. And by the way, part of the reason why these startups are so valuable whether it's coding or legal is because they're dealing in very high economic value. And so the difference is you say well, Kimi's almost as good. It's 85% as good. In high economic value things, 85% kind of rounds to zero. So it really has to be something that is not a non-obvious problem to be solved. It has to be something that it isn't just that the model company can literally almost tell its model go close the loop and learn everything about this and then just offer the product. It could be that you're integrating data sources that are fundamentally not really available to anybody including the model companies. For example, even if Anthropic decided that I'm going to start doing Airbnb, it couldn't do that. What Sid, Ujwal, and I are doing with Monas is like, 'No, you can't just go, now Claude, tell us about biological models, biological molecules that might actually detect and prevent or cure leukemia.' It doesn't work that way.

So, I think the good thing about the physical model space is that there's a whole bunch of stuff that's pretty unique that's outside of the data sets that OpenAI and Anthropic are doing. Including getting your own data for the physical AI world, right? Exactly. The data for the physical world, what kind of compute fabric you need for that, what compute fabric works in the right kind of time frame. LLMs are still pretty bad at real time, right? They try to hack it in various ways. So, there may be a whole bunch of stuff there that creates something. But there's a reason why a lot of smart people are doing it. It's a good bet because it's like, 'Well, here's a new area that's different than LLMs that could be really interesting. It's a risk-adjusted bet.'

I don't really know what fully the ripple effect is. It does tend to be that when people make a pile of money, they tend to go both into some philanthropy, which is a whole bunch of things, and some investing. One of the things that I think has been not told as loudly as it should about how the ecosystem of Silicon Valley got both so broad and deep is because when you had successful people, they stayed there and then they turned into angel investors and advisors the next thing, and then eventually it compounds. When my PayPal colleagues and I came out, we were talking to each other, investing in things, starting things, etc. That will certainly happen because it's not just the founders of each of these companies, but what happens is all the executives and everyone else tends to make a bunch of money, too. And the really interesting question is, what happens when you're getting pitched ideas? Let's just be a little bit fanciful on this. An AI is pitching you for their business idea. And they're going to go execute this with a whole bunch of different compute fabric. And either you or you plus an AI or an AI is making the capital allocation decisions because compute allocations will be capital allocation decisions. My guess is on that universe, there's some deeply utility things about there being firms because the firms are running competent specialized AI with data and all the rest. This is all similar to the startup landscape becoming highly variable. We now of course have the investing landscape changing shape, too.

Part of what has made capitalism so essential and so successful to humanity, to all societies, has been that the allocation of resource decision tends to go to the things that drive quality up, price down, etc. And so the answer of how does work get allocated is, where does the work get done at a certain amount of quality at a low price? Now, that being said, humans will always be, I think, central participants in this. The question will be, how do we feel that we are in a collective market game that benefits enough of us in this game, call it 80% plus, that we will all hold ourselves to this game. The notion that human beings will say, well, we're just less effective at everything, AIs are more effective at everything, they should do everything, is like, but we still have to have a role and have to participate. And so we'll be looking for those. But I think we will create the markets that enable that. Now, you mentioned legal before. I think one of the funny things is, and I don't think they should, but I think they will, I think lawyers will go, 'Well, if it's law without a lawyer, it's not law, just like license.' And so we have to be there.

Exactly. And by the way, that's what I think should happen. But I think the notion that we, a little bit like the Pope's great encyclical, it should be like, 'Hey, be focused on what humanity at the center means.' It just doesn't mean historic humanity at the center. It means a good role at the center and how the future is evolving.

Or at least chaotic since they seem to love volatility.

I think sovereign wealth funds are a good idea. I think they've been executed in good ways in a number of places in the world and it's one of the things the US is behind on. You can look at the Singapore fund, you can look at the Norwegian fund. There's tons of these that are good. Then I think it gets down to how do you get the sovereign wealth fund started? How does it go? What's in it? And I think a little bit of the question comes down to, I'll put this in classic American Wild West libertarian. We're going to go steal some property from some people in order to put it in. Right? Like we're just going to go appropriate X percentage. That I think is very alarming and is destructive to the entire system. Now, if you said something in between the two and you said, okay, this is going to have a huge amount of economic growth, we're going to have civil disruption because of questions of how jobs and industry changes. So, we're going to do something intermediate and say, hey, companies X, Y, and Z, you have to accept investment from us at the same price that you had been priced by the market most recently and we're going to invest a bunch and own a bunch and then you can use the capital to grow your business and that's part of the benefit of being here and you had some relatively equitable way. I'd prefer it wasn't forced, but whatever, to incentive to happen. Then that's more like, okay, we'll have some economic basis of this. Now, part of it is I think there's been a bunch of investment in Anthropic and OpenAI from a number of funds that oversee pensions, 401ks, all the rest. There's already a certain amount of public participation in that. So, those are all some early reflections, but I do think sovereign wealth funds are a good thing. I'm not sure anywhere in the world we've made state-owned enterprises kind of equivalent functional entities. Right? I don't know if I've ever seen that. I've seen sovereign wealth funds work amazingly well. Kind of as venture firms, but they, you know, like, 'Oh, it's a state-owned enterprise.' I mean, for example, utilities companies just about anywhere, kind of a problem.

Maybe it's mixed news. I think it's broadly bad news, but there's probably some good elements to it. The bad news part of it is it doesn't look like there's anything that's particularly principled, here's the way that we're navigating through things, apply rule of law and predictability. It's more like, 'Hey, we got triggered and we had some hostile interaction or some contentious interactions with this company anyway, so we're going to hit them with a stick and we're not justifying why we're hitting them with a stick, but not OpenAI or others.' There's ways of doing this. They say, 'Well, those models aren't the same.' It's like, 'Okay.' It's not incredible relative to the cybersecurity particular issues, although I think that Anthropic was putting real energy into making sure even in their general release the cyber thing didn't create any problems for key industries. And so, it's a little unclear. And a kind of autocratic willy-nilly way of applying it is very suboptimal. Now, the good part of it is, look, I think we should be paying attention to major threats. Cybersecurity, bioterrorism, etc. And should intervene when we need to on that. And at least that's potentially a positive case of that.

Asking for a friend. Well, as an investor in your fund, I'd say you know, don't share all the secrets on here. Yes, exactly. So, look, it's obvious that I think the biopharma stuff is pretty important. I don't think there's going to be room for only one with Monas, but one of the funny things about the Monas pitch deck is we have 'we are an AI drug discovery factory for creating monopolies' and you're allowed to say monopolies in your pitch deck because that's essentially what the IP is for drugs. Right? So, we're not making any antitrust violations or anything else and it's perfectly fine for it to be discoverable. Yes, yes, this was our pitch deck. Right? Because of that. And so, I think there's a whole bunch of different stuff on that. I think one of the maybe more uncomfortable areas but what will be huge is a bunch of stuff going on in defense tech because I think we are in a time of more war and conflict. So, I think that area will be another one. I think the questions about where the AI can be embedded in something else that has a really big moat is interesting. I'll say one thing that's kind of not per se economic. So, it's kind of easy as like how does AI mean what does AI mean for the reinvention of universities? It's been one of the things that I've been paying attention to. And I don't think it's an economic thing, although I do know of one for-profit university in the UK that's going heavily in that path. Anyway, so those are some random comments.

I think for sure it goes all the way back to John Naisbitt's Megatrends from the '70s, which is high-tech, high-touch. I think for sure as we get high-tech, we want more human on a variety of things. The question really is, what does it mean for investing and so forth? For example, just a couple weeks ago I was in Japan and the focus on tea ceremonies and that kind of thing. Human experiences. I think there'll be a lot of desire for those kinds of things. Which of those things are interesting economic things at small scale or at large scale, I think are TBD. But I think the intense demand for the human will be very high. And so it has like where does the buy-side demand match up with the supply-side pricing? But that's part of the reason why I think, when people are trying to think about what happens when we get to an AGI universe, and they say, well, I think live entertainment and hospitality and so forth tends to be areas. You say, okay, well, what's the particular way to bet on that in a way that gets amplified? I don't know. But those kinds of things, I think it's one of the reasons why a lot of people say, 'Ah, sports teams, no one's ever going to want to watch robots playing basketball.'

Work in progress.

Well, as you know, this is part of the reason I wrote Super Agency, which is agency is some degree a mindset. And so, I think what's really important is you embrace AI as part of your agency. And you say, this is part of how I learn, how I work, how I live, how I do things, and so forth. And it isn't like I sit back and I do whatever the AI tells me to do. It's the AI is my tool, companion, car, etc. as I navigate things. And I think it's really important to start iterating and doing that. And it's a little bit like one of the things I've been thinking about, writing an essay on and doing some stuff, has been the kind of mistakes that are made by college graduates booing or otherwise dissing AI. And you're like, look, you guys have the opportunity to be generation AI where you come into the workforce saying, I know this a lot better than all of you. You should be hiring me in order to help you become AI native organizations and so forth. And it should be an opportunity, not a threat. They say, 'Well, but entry-level jobs market's way down.' It's like, yeah, at the moment it's actually not really because of AI. There may be a do this, that, and the other that is. But it's actually because of global turbulence and businesses being unable to figure out how to invest and plan. It's because of over hiring in the pandemic and the 'hey, maybe remote work really does work. Oh, right. Remote work's pretty hard to make work.' Lots of AI washing. And so it's AI washing of all that stuff. But it's embrace it for your agency. AI can do a whole bunch of amazing things itself, but is not complete. And humans can add in a lot of significant and important things.

Yes.

Always awesome.