The Geopolitical Stakes for LLMs: Josh Wolfe & Teresa Carlson with Zoe Weinberg

Josh Wolfe, Teresa. Let's see, I don't know if this is on. Okay, great to have you guys here. We're in a moment of a lot of geopolitical uncertainty. Tensions are rising with China. We have conflicts spilling out across the Middle East and Europe. More people than ever before are going to go to the polls in 2024, including in our country. It's always struck me that when it comes to investing with a geopolitical lens, there's two approaches. One is that you take advantage of the dislocation. So more conflict, oil's going to go up or something like that. But the second is about investing to achieve a certain outcome. We want to see the US prevail on the global— in the global stage. Therefore, we're going to invest in the technology to make that possible. It's much more like normative style. I've been under the impression that both General Catalyst and Lux subscribe to that latter approach, but would love to hear, is that right? And how do you think about investing with a geopolitical backdrop? Josh, do you want to start?

I would, first of all, say it's never not been uncertain. There's never been a time where we said like 2024, my God, this is the year of certainty. And so there's lots of exogenous shocks and surprises, whether that's at a company or country level. We are in the business of making cash on cash returns, and so we're going to go wherever we think that we can do that. But we definitely have a geopolitical lens where we are American patriots, and we have identified over the past decade some foreign actors that we think are disrupting systems. And we think that there's in turn demand for technologies and solutions that can help to counter those disruptions. Some of those are physical and kinetic on the military side, and some of them are AI and detecting efforts to infiltrate and corrupt democracy. And people can also claim that some of our own domestic companies have done that as well. But yeah, there's definitely a lens of how do we make money, and then how do we also do things with US and allies that we think are pro-social and advancing democracy.

By the way, thanks. It's great to be here. And I want to go back to go forward for just a minute if I can. In 2010, I've been at Catalyst for a little over a year, and before that I was I ran Microsoft's federal business, and then I went to AWS in 2010 and worked for Andy Jassy for 11 years and built a worldwide public sector and regulated industry. So I want to talk about that for a minute to say why I think investing is important to think about it from a local, regional, geopolitical lens. And General Catalyst absolutely has that view starting at the top with Hamut, who's been thinking about this already. But in 2010, if you think about when cloud was happening, we were going around the world and talking to different countries and opening up business. And there was no way that we could get in anywhere without a lens of what was happening in that region or that country. And I love where we are right now with investing because back then, believe it or not, we couldn't get VCs to come in to places like the Middle East or Asia or, you know, even Europe, we were seeing great, great, great startups. And because the opportunity in the US was so amazing, you know, most VCs would say, look, I don't really— I don't think I need to go anywhere outside the US. But fast forward today with AI and what's happened on the geopolitical stage, we've had a massive shift even since then with what's happening with the Ukraine. And Russia war, what's happening in Israel. You're seeing election shift right now. We just had Macron announce that, uh, a snap election. We're seeing that in London, in the UK, and you can see complete shifts of, um, who's in charge and how they're reacting to technology. And we're seeing much more of a, of a mine, you know, this is— I need to do this here. And because of that, you really have to think about how you how you invest globally and how you create a model that's really impactful for these founders. And when you look at companies like Menstrual and others that are popping up in Europe, they are getting that DNA of how you create something. And when you look at who's building those large language models, they're not all in the US. So we need to be able to take an approach globally. And we've done that. You probably saw about— it's been about 8 or 9 months ago now, we merged with a group called La Familia. That has expertise in Europe, and we just announced a merger with Venture Highway, which is in India, and we think those are two really important parts of the world.

Theresa, is that part of the reason— or I'll let you tell me— that GC has had the strategy of global resilience? And it seems like that name was chosen very deliberately, whereas a lot of equivalent or maybe somewhat similar strategies are around American resilience or American competitiveness, like Why global resilience?

You, you hit the nail on the head. It is really about that, and, and I love it. It's part of the reason Paul Kwon and Hema, or they're talking to me, I was like, I love it because I'm, I'm, my background is in, uh, public sector and regulated industry, national security defense. And when you think about global resilience, you can't, you really do have to have a very geopolitical lens and say How is this group going to think? And if you think about markets like defense, intelligence, financial services, energy, these systems have to be resilient in their local economies. If these things start to fail, it's a problem for America as well. Why do you think we, you know, at the governmental side, we supported the Ukraine war? If these things start to fail, there's a domino effect. So when it comes to investing, we need to have a different lens. And I love America. I think America— I'm always so amazed at what we create here. You know, you go around the world and as we teach these founders how you bring on what we have here in terms of grit and resiliency, I mean, it's such a great lesson for them to learn from US investors and founders. But they have to do it in their own way. And I'll just give one little example. If you start talking about capitalism and economic growth in Europe, that only goes over so far because they see, for a variety of reasons— they're a bunch of small little countries. And if you talk about too much economic growth, they see it— as an example, it could hurt the climate. If you don't do it in a very thoughtful way. We don't tend to think about that too much. We're like, how do we build more data centers and how do we get AI out here in North America? So I mean, we do have to have the lens of how they think about it, even though in the United States we may not always agree with it. But if we wanna invest thoughtfully, we can make really good choices and make money, I think.

As conversations about regulation and AI, you know, continue to become front and center. A lot of VCs have now opened shop in DC, presumably for some investing purposes, but also policy influence. What is the right way for VCs to be engaging in that conversation, and is there anything that you think is out of bounds? Josh, do you want to start?

At a high level, I always say that ignorance of the macro and regulatory environment is no virtue. You know, if you think about what we do, it really is almost security selection, entrepreneur selection at a micro level. And you could pick the best entrepreneur, and I equate it to picking the best dish in a restaurant, having selected that restaurant from a neighborhood, having selected that neighborhood from a city, et cetera. And then all of a sudden, you're about to take this delicious bite, and Godzilla comes and steps on you. So you need to understand what these external exogenous forces are that could trump what you're doing at a micro level. Regulatory, where you stand on the issue, particularly in the context of AI, and I think the main debate at the moment is open source versus closed. Depends on where you sit in the cap table. And so Vinod and I were actually in the White House 4 weeks ago, and we stood on different sides of the issue. Vinod sits in the cap table of OpenAI. We sit in the cap table of Hugging Face, both earliest and I think largest investors in both companies. And he would make the argument that we need to have a central US closed model because we need to beat China that way. And I would make the exact counterargument. And the logic would be, most open source really is almost in this Karl Popper sense of science about conjecture and hypothesis and code and then criticism. And you need open systems and free speech for criticism to be able to correct things. Good luck in a closed system of, for example, if it was China-controlled, of being able to question the Uyghurs or Xinjiang province or search for Tiananmen Square. And so those things will always approach a really far asymptote of truth, whereas you can do much better in open source systems for these correctives. So I think in the spirit of regulation, the government would love to be able to regulate, to be able to control, particularly for military use, for being able to protect for election integrity. But I think that open source is just far more diffuse, and the history of it is that it will win. But there will be a lot of money spent on top of the hundreds of millions that are being spent now lobbying to try to keep those things captive.

Yeah, I want to build on that a little bit and bring you in, Theresa. Zooming out a little bit on the subject of regulation AI, it seems to me like the national security argument for not regulating goes something like this. It's like, we overregulate AI, it's going to slow down development, and China is going to win and the US is going to lose. Makes sense. But it feels to me like we never really finish that sentence, which is like, what is the end state that we're actually driving towards? Presumably, the reason we want the US to win the AI race is that we want a form of AI that is deeply embedded with American democratic values, probably things like preserving privacy, respect for property rights, in this case maybe intellectual property, a certain amount of commercial openness and accessibility. Is it possible to achieve that kind of democratic form of AI without guardrails by just saying, like, this is a free-for-all when it comes to development? I don't know, Theresa, if you want to start there.

Well, kind of two points. I would say one of the things with government that you really need to do is you do need to radically collaborate with governments. Trying to always use a hammer with them doesn't really work in today's world. It may not work either way, because nobody's agreeing right now. But in some ways, sometimes for standards and policy, that is OK for folks like us if we're investing. Because if you look at what's actually happening in regulation right now, Nothing. I mean, it's really a lot of talk. You have, you know, Colorado that's done something at the state level. I think a high risk may be that you have states that you have to pay really close attention to, could get out there in areas that may not make as much sense. So we need to be looking at state legislative actions as well as federal actions. And at the federal level, let's say that Biden stays in, we already understand what he's going to do. There's a path where we understand kind of how to approach the administration and the congressional leaders, um, you know, who's kind of on point in making those decisions. In the Trump administration, they'll totally change it. I mean, I don't know what it'll look like, but they will change it. It might be good, it might not be good, but they totally will not do the same kind of things. And in the national security space, uh, what you have to think about at the federal level is you have— you are going to have some natural guardrails already without even having mandated legislature. Like you have something called FedRAMP, which I know a lot of you all are familiar with in the federal level. They already have an AI model for that. So companies can get so far, but then after that they will have to pass these guardrails. But one other thing I'll just share in this room, congressional leaders are so excited to talk to venture capitalists and startups. They really do want to hear from them. They may not always— and I say this with respect— they may not always understand exactly what the technology is doing. So I think they're very hungry, especially in an AI world, to see it and kind of get their hands on it. And we've been asked for a lot more. Show me, explain it to me. And we are in a totally different world. So I would just encourage us all to like engage if you can and be open with your legislators to actually talk to them because there is a big leap on education on what they understand or don't because it's a new world order out there for them.

Anybody that has capital or is perceived to have capital can certainly get an audience with legislators who are very keen not only to learn from you but to take your money and donations. And that's, you know, in every industry. But I think that in particular, they have identified that there's a lot of capital in venture and private equity and technology that can be— put in their coffers. I think that there are two things that matter regulatory. One is domestic, and it's really thinking about the M&A environment. And Microsoft, I think, has done extraordinary regulatory arbitrage in circumventing the DOJ, FTC, and figuring out how they can effectively have captive control through terms and structure in OpenAI, in Inflection, and in others. Instead of doing an acquisition, doing a license. There's a lot of things that Brad and crew have done that I really applaud in a cleverness to— route around regulations and in the sort of regulatory arbitrage. And that will continue sort of with them unfettered getting access to what they need. The second thing that matters in the regulatory environment is really protecting the American public and our companies from unfair competition and influence. So there's a handful of us on the venture community and hundreds others that were very actively involved in what ultimately, in the end of this year, will either result in divestiture or banning of TikTok, which I would argue is really the first AI-driven algorithmic assault on the American people. And it was very insidious, and invidious, and clever, and also admirable in how it was done. That, to me, has been sort of the greatest threat of AI. And I think that that is a role for the government to do in protecting and identifying. And it doesn't stop with that one app. There's going to be about 2 dozen others.

What's like the good outcome for TikTok? Is it like— in your kind of like end state, what do you— What do you see?

To me, it's akin to other social media that we have in the United States, which is you can still say that Twitter, which I love, is toxic, or it's a cesspool, or whatever. But it's a cesspool that doesn't really have the foreign adversary with its own sovereign interests tipping the scales of the content that people can see. And so that, to me, and being able to influence— there was no way in the 1950s we would have given, or 1940s or '30s, given Nazi Germany its own dedicated channel. Soviet USSR in the 1980s its own dedicated channel. And not only have we given the CCP, the Chinese Communist Party, its own dedicated channel, but every single person that has a phone between 18 and 24— 18 and 32, 30% of their news is coming from TikTok. And that news is, of course, misinformation, disinformation, and the ability to enrage and engage people and skew the things that they support, which in turn then translate into what they demand of their legislators, which in turn affects where capital goes and to which allies it goes or doesn't, where it's withheld. That to me is the democracy algorithm that has to be regulated.

Theresa, anything on TikTok that you think?

Well, you know, that's such a hot potato topic now, you know, in Washington, D.C., on the Hill. It's crazy. I— they are just— everything is anti-China. I mean, there's no person you talk to right now. They're looking at everything, every aspect. If you come to them with good investing opportunities in anything in the supply chain, shipbuilding, weapons systems. I mean, they are, and they're just anti-China altogether. I think there will be some outcome, but I think whatever that outcome's gonna be, it's gotta be pretty much purely US, and there'll have to be oversight. They will wanna look at who's on that board, who's running those operations, where do those servers sit. I mean, there'll be a lot more oversight. And for a lot of us, those aren't things we like here in the US. If you think about Europe is usually the ones out in front, good or bad, on regulation, and we're way out in front on innovation. And we kind of move so fast in the US that the legislators are going, what the heck? I mean, they're like— I remember when we were doing drones at Amazon. We were 5 generations ahead, and they were like, what's a drone, kind of thing. And if you— I mean, look, drones is a good example. If you look at policing agencies here in the US, 89% of those still use Chinese-made drones. And we only have 2 real factories here in the US. Many are not American-made. And how are the wars being fought? Who of us here in this room would have thought 5 years ago we would have been talking about weaponized drones and how you know, defense and intelligence was gathered. So a whole new opportunity, right? If you're thinking about what gets made here. But we got to figure out that manufacturing process in the US. We got to rebuild American manufacturing, and we've got to do it in the right ways. And I think that attention will grow. So lots of opportunity, even from basic welding to wire harnessing. We need to get right here again.

Two quick points on this. One, we are making foreign investments in sovereign models. So my partner Grace and Brandon led an amazing investment in a Japanese company called Sakana that is really their national darling. And they're a US ally. And when you see the people that are involved, it's very much focused not only on the sovereignty in Japan and the models and the talent, but also on where the US and AUKUS in particular can ally. And it is really against, in specific, the CCP as opposed to China writ large. But how do you defend against that? On the physical aspect of this, I think it's really important because when you think about great power competition, you think about the projection of power geopolitically. World War II, it was about how do you make stuff. It was manufacturing. Can we build ships, and planes, and boats, and so far in the gears of building the arsenal of American democracy. Cold War, it was about the atom. It's about nuclear and the ability to both control and project that power, and then ultimately our use of it that made it credible as a force of— mutually assured destruction. Today, a lot of people are like, well, AI is the new nuclear. And I think we've talked about this. And I disagree. In part, it will become that because policymakers and military leaders think that AI is so important. So there's just this positive feedback effect that it will become important. But it is far more diffuse than the controllable local atomic matter. On this idea of the physicality of it, you think about Taiwan and the geopolitical implications of TSMC and making the GPUs, which are powering the AI revolution. This still comes down to mass. It is going to be less about AI. You think about the ships that are needed to defend Taiwan Straits. You can't use AI to change 37 megajoules per kilogram for the diesel fuel that Navy ships run for. You look at the battlefield space that military commanders are looking at. You've got 40 amphibious ships. You've got 50 subs. You've got 200 surface combatants, 2,000 jets, 3,000 UAS, 750 satellites. That's all physical mass and matter. And so as excited as we are about AI for a warfighter to be able to control multiple entities and have a better sense of the battlefield, it still is physical mass and manufacturing and being able to have these attritable things to disrupt.

And can I just say one— because I really agree with that— but one quick thing I want to say on that. Remember, we need three things to make all this work anyway. We talk about this. We need energy. We need compute power. And we need data. The only way— and we talk about these, and AI is not there yet in a lot of these areas. The data and the way it's connected is not even set up properly to use yet. So a lot of this stuff is so new as they build the models and they train the models. It's early days. And I think the concern is if someone, to your point, physically is not overseeing that, it could get a bit out of control. There was a really great article in The Economist if y'all haven't read it, on how the AI warfare is changing AI. It's actually really well written. But this is the thing, like we're early days and 80% of the budgets now for governments for AI are spent on training the models. So they're early days, which is kind of interesting. You think like you need good tools to even just train the models for them.

We're effectively out of time, but I do want to ask a couple really rapid-fire questions, and you have to answer really with just one word. Kai-Fu Lee recently said that China is 6 to 9 months behind the US when it comes to development of LLMs. Would you say you're over or under on that?

Under.

What do you think?

Yeah, under.

OK. Do you think that AI is going to come up in the presidential debate tonight?

Yes.

No.

OK. And do you think Biden and Trump themselves have actually used AI tools?

Yes.

Yeah, you think Biden's like a big Midjourney guy?

Oh, he's on TikTok. He's— yeah.

What about you, Theresa?

Their teams have, for sure. I'll say that.

OK, fair enough. OK, we'll leave it at that. Thank you, guys.

Thank you.

Thanks.