Can you (or DOGE) product manage the government?

Danny Crichton:

Christine, thank you so much for joining the podcast.

Christine Keung:

Thanks for having me.

Danny Crichton:

So Christine, you joined us a couple of weeks ago. You came to the launch event we hosted in New York City for our China EV game, Powering Up. And in that game, for those who haven't followed the last couple episodes where we've talked about this, we're modeling the local Chinese auto economy, specifically the transition from internal combustion engines over to electric vehicles. And we look at three companies, one that's sort of US driven, US General, one that's European, Autobauer, and a company called Shanghai Car Company, which is sort of the local indigenous Chinese company.

And then there are two characters who represent vice mayors of Chengdu and Shanghai and a foreign consultant who tries to connect all the dots. And you smashed, you were number one, you had 26 points, which was one of the highest scores we've ever seen, and you played US General. So with that sort of being said, what happened? What was your experience? You came into this kind of cold, you somehow beat everyone else at the table. What was sort of the strategy? What was going through your head?

Christine Keung:

And we should clarify. The roles were randomly selected.

Danny Crichton:

The roles are randomly selected.

Christine Keung:

So I didn't know I was going to play US General.

Laurence Pevsner:

I'll say they're semi random in that we do try to specifically give people roles that are not geared towards what they do day to day. So if you are normally a consultant in your day job, we didn't want to have you playing the foreign consultant for the risk game. But other than that, yeah, fairly random.

Christine Keung:

I can tell you going in. So when we drew our roles and we shared it with the rest of the team, my heart sank when I drew the US General Motors. I was like, "Oh my gosh, I'm basically guaranteed to lose here." So I was also surprised by the results. My very first war game was actually in grad school. I took Eric Rosenbach and Ash Carter's class at the Harvard Kennedy School. It was a class that was very much focused on defense tech, defense innovation, and every single one of our exams were war games that started at 6:00 AM because they wanted it to be fully immersive. So I actually haven't played a war game since I was finishing that class a few years ago.

Laurence Pevsner:

Well, ours was not hosted at 6:00 AM thankfully.

Christine Keung:

I did appreciate that. And ours was not out of wonderful Italian restaurant either.

Laurence Pevsner:

Yeah, exactly.

Danny Crichton:

Well, so what was the strategy? So when you think about it, you got this role, you were sort of surprised, you knew the story, because you covered this industry and have a lot of background in it. So you knew that you were playing a tough role, the role that in real life has lost, but you won, so you did something different than what happened in real life. So you got into this role, you got to the first scene. What were you thinking?

Christine Keung:

So I based a lot of my strategy on just some real-life experience having worked for the State Department over 10 years ago. So my first job out of college was actually a Fulbright Fellowship. So I was a Fulbright Scholar based out in Western China. Some part of China's industrial bases certainly concentrated. And I worked for Ambassador Max Baucus who was a very experienced and commercially oriented diplomat. So one of the things I saw living in China, now it's over a decade ago, was the nature of joint venture, right?

There was a time when every US entity that needed to enter Chinese market had do a joint venture and it made sense for them to do a joint venture. They were bringing a brand new product into a market with extremely tight regulations and it made sense for them to find a local Chinese partner to help them scale and grow and navigate those political nuances. And so that was true 10 years ago. And so I stepped in with that strategy and that posture and went in thinking like, okay, the very first thing I should do as a US company is set up a joint venture. And because I was trying to expand in China, I partnered with Shanghai Auto Company, I think quite successfully, and we were able to capture the market and create leverage very early on.

Laurence Pevsner:

I was the leader of your game. We had Danny's world on one half of the table and Laurence's world in the other half. And so as the game leader, one of the secrets that you didn't know that I was keenly observing because part of our scoring metric is do you give away information as part of that joint venture? And something you did differently than almost every other joint venture that we saw in the game was you didn't give away your information in that trade. You managed to somehow make a deal with Shanghai Car Company without giving up your IP as it were. What made you decide to take that tact and why do you think it was so effective?

Christine Keung:

I just felt like the information was all I really had because all of us also started with different numbers of chips. I noticed that my role, I actually had the least amount of social capital, social or political capital versus I had a lot of information. So when we started playing, I was also I guess counting everybody's chips.

They were all color coded and figuring out who has what I might need. And Shanghai Car Company actually had way more political capital. So intuitively, I knew that in order for the joint venture to work out for me, I wanted to sort of leverage her political capital to grow. And intuitively, I knew what made us, and this is probably true, even outside of the war game, what made US strong was the product that we were going to bring to market. And so I was careful from the beginning to just protect that information.

Laurence Pevsner:

Yeah, that's exactly the thing that we were trying to simulate. You nailed it. We were, the whole point was the US comes in strong, we have the IP, we have the technology. We are more advanced than EVs at the start, but if you transfer that IP, then of course China starts to get an advantage because now they have the same tools that you do. Whereas obviously Shanghai Car Company or their equivalent in real life's advantages, they have the social capital, they're going to have the connections and the wherewithal and the political advantages that simply a US car company will never have. So I think that you read the situation correctly, whereas we would say in real life, unfortunately, we feel that the Fords and General Motors of the world didn't assess it the way that you did.

Danny Crichton:

Well, and for some reason, your world actually was very smart. So I'm looking at the results, but 36 billion cumulative revenue for the European US automakers, just 6 billion for Shanghai Corporation in your world. It was basically the complete opposite in mind. Shanghai actually made 22 billion and US General and Autobauer were in the single digit billions. And so the key thing here, which I think you recognize right away, even though you had never played the game, we didn't really give you rules, was you only have one resource. You have this thing you can leverage, and as soon as you sort of give it up, that's all the leverage you're ever going to get.

And so you have to stretch it as far as it can go. Otherwise, you have nothing else to really offer at the table. Let's move on. So you hosted this game, but more broadly you've been in wargaming, you actually, my understanding is you've been looking at the wargaming space from the venture capital lens, which is very intriguing over here on my side considering that we are the wargaming department of a venture capital firm. We don't actually invest in a wargaming platform or anything like this. But I'm curious what attracts you into political simulation's idea of play as a way to experience complex scenarios and why you're sort of interested over there.

Christine Keung:

Yeah. So we did invest in a company that I would say is risk game and war game adjacent, but we didn't make that investment because we wanted to just invest in wargaming technology. What's interesting about war games is it's probably one of the oldest processes in the world, as long as there's been competition, opposition, and wars, simulation is how you come to a set of scenarios that could happen. And historically, a lot of that has been paper-based and very similar to how you guys ran the risk game, which is it was in real life, it was tabletop and it was basically run concurrently to basically try and capture every possible outcome. What I got really interested in, and I think this is something that software is able to do better than their human counterparts, is capturing some type of telemetry throughout the decision-making because so much gets lost in the thought process of making decisions and executing on strategy.

And that's actually something that software, especially game development tools is very, very good at doing. So the company that we invested in that is supporting the Air Force with some of these initiatives is a company called Gamebeast. We co-invested with Andreessen Horowitz's Gaming Fund, and Zander actually comes from a military family. So this was a founder who is a prodigy game developer. He actually built his first multi-million dollar game when he was 16, wrote a best-selling book on Roblox user-generated development. And we invested in his company when he was 21, and I think he's like the youngest person on Forbes 30 under their 30 for the gaming category. So he's somebody who is both extremely embedded and well-versed in game development, but also has the intuition and appreciation for why what he does in his day job has broader implications. And one of the things that he was able to build was a live operations platform.

So basically a tool that game developers can use to monitor and track and measure play in real time and basically push out changes that optimize play in performance in real time. And that's something that I noticed you guys did, right? You guys did it manually, but what I really liked was Danny and Laurence, when you guys were running your worlds, you also had Ian kind of walking around. You also had other, and I think Michael too, folks were walking around trying to collect some of that metadata. You guys were actually listening in on our conversations. We actually had the option to basically pull the end in court, stop, ask for clarification.

And those are all things that nobody tends to pay attention to or talk about in a risk game scenario, but I actually think is critical for capturing every possible outcome and selecting the best outcome. It's the behavioral metadata. So he's doing it on the gaming side and without being too on the nose here, intuitively, we also understand why it's important in a military context, so much gets lost between the lines. So much gets lost in the negotiation decision process that if you can systematically capture all the data that might actually lead to less intuitive but more probable outcomes.

Laurence Pevsner:

Right. We have lots of data on the actual decisions that got made, but none on the micro decisions that led to that conclusion. And so your point is if we could actually see all the path and exactly where they followed it, we can both understand why good decisions were made, why bad decisions were made, and maybe they're actually just on the level of those micro decisions rather than just, "Oh, I just decided, I just went for it." There was actually a whole thought process that I went into it to begin with.

Christine Keung:

One thing I want to underscore with that is the use of telemetry has been commonplace in software development. When I worked at Dropbox, that was actually very common where software development teams would basically use software to basically monitor their network activities. That's very commonplace in software development. That's commonplace in athletics where another J2 investment is the Oura Ring where consumer wearables are very commonplace for training high performance athletes, even non-high performance. I'm not a high performance athlete. I use the Oura just to get data on myself, to get continuous data on myself. And that's true obviously in gaming. Those are the folks that Zander is working with. So why not bring some of these common practices that we've already used to optimize athletics, optimize our sleep, optimize our software development? Why not bring that to even more critical decisions related to war?

Danny Crichton:

I think one of the crazy things when you look at human performance, particularly in sports teams, the technology that we have available, the way that we do strides, the way that coaching is done for very elite athletes is insane, right? The platforms, we actually do 3D animation and modeling of their bodies, we're actually looking and being like, "Look, you're not hitting your toe exactly in the right spot. You need to do this versus that." I just try to get on a treadmill 20 minutes a day and I fail at this and then people are doing these micro optimizations.

But to me, when we're directing this back to decision making, I think of the qualitative and the quantitative. One thing that's definitely missing is the quantitative when for one reason is that we often don't make the same decision over and over and over again. And that's where the gaming piece becomes interesting. The other side of this on the qualitative side, and this is just from observing games in a kind of Tom Wolfe Bonfire of the Vanities sense is you watch how people interact with each other in these scenarios and it actually matters who speaks first.

It matters when did the idea show up in the room versus later like in the way that this China EV game works, you have quite a bit of time pressure in this game that we normally don't have. And so it actually matters if someone makes that proposal in the first minute versus the last minute. So good ideas can just fail. It was badly timed. They talked to the wrong person, the first person they talked to said no, but if they talked to someone else, maybe that would've cascaded and actually taken place. And I find the qualitative piece really fascinating because it's really hard to capture. I think of our own industry of venture, a partner meeting, we have a discussion, it's like how much contingency based on each of the individual factors in that room actually influences the decision is sort of mind-boggling to me.

Laurence Pevsner:

Yeah, I mean I remember when we were setting up this game and we toured it all across the country and around the world, one thing that I was surprised mattered quite a bit was the seating arrangement. Just which players would we put us general next to one of the cities, we put them next to Shanghai, would we put them next to their competitor in the EU? And I believe in your game, you were sitting next to Shanghai Car Company.

Christine Keung:

I was.

Laurence Pevsner:

And so it was very easy for you to just turn to them and say, "Hey, let's get this deal done," versus even if you were just sitting two chairs away from them, it might've created a different experience.

Christine Keung:

That's true because it would've created at least like an increased barrier to engagement.

Danny Crichton:

And look, there's actually research. I mean, there's empirical research from like MIT Media Lab and others around social physics. But the idea of if you're the next office over versus down the hall versus different floor versus different building, different city, different part of different continent, your ability to collaborate with people changes dramatically and exponentially decreases with every foot, if you will, of distance. And so you see the same thing with decision-making. We see the remote work debate, we see the in-person office debate.

I don't think we've actually developed high quality decision-making tools in a remote context where people are not in the same space. And so that'll be something that's interesting going forward. But I do want to draw our attention back to you because obviously you've had a very, very fascinating career. You were in Western China, you were at Dropbox, you were back in government under the SBA, the Small Business Administration, during a very time for small businesses. And then you have an amazing early backstory as well. So I had love to hear the personal, how did you get into this industry? How did you get fascinated by some of these topics?

Christine Keung:

Well, I can tell you I definitely did not set out to be a venture capitalist. It really wasn't something I thought about as a viable career path for me. So I would actually say my path into venture really started with my time in government service. So maybe I'll share my professional background too and then maybe share a bit more about what got me interested in China and my family's immigration journey. So my background has always been a mix of government and private sector, but I was somebody who actually started off in the private sector working at high-growth tech companies like Dropbox, Figma, Fountain, typically on either technical or business operations teams when the companies were hitting hypergrowth. So what's really amazing about these learning opportunities is you get to learn how to make decisions and lead teams when the company is just going through unprecedented growth, the type of growth that you actually can't really simulate.

Elad Gil just published a book called the High Growth Playbook where he interviews some of them are my former colleagues. He interviews a lot of folks who have had similar experience working in leading when companies are hitting that unprecedented J-curve. And the reason why he chose to do it in an interview format instead of maybe taking it more academic or statistical way is because it's hard to study this unless you've actually been there leading in high growth settings. It's not an intellectual exercise. It's hard to simulate. You sort of have to be there in order to learn how to do it. And so that was actually most of my career experience in tech was just working for these fantastic companies. And what I really appreciate about each of those teams were how lean we were when we were meeting that unprecedented demand. I think at Dropbox, I think we were maybe like a thousand employees max worldwide, a thousand full-time employees max when we were serving 500 million users around the world.

And if you actually think about it, a thousand people serving 500 million users, that is incredible, incredible leverage and scale that you really don't see right now in the US government. So I did all of that up until 2018, went to business school and graduated into the pandemic and decided to actually take what I thought would be a short break, but turned into two and a half years from the private sector to go and work in government. I became a Trump political appointee in March of 2020. I joined the US Small Business Administration to help them launch the Paycheck Protection Program. I like to argue that the Paycheck Protection Program, the PPP, which ended up deploying like $800 billion of capital into the US economy, was probably one of the first high growth programs that the government ever had to administer.

Danny Crichton:

Interesting. Yeah.

Christine Keung:

Just to give you some data around it, the US Small Business Administration gave out more loans in the first 30 days of the pandemic than the last 30 years. It was a time when a federal agency that I would argue many folks might've not heard of or known about, became super important during the pandemic. It was everybody wanted a PPP loan program. It was an agency that historically was B2B, business to business. It was an agency that was developing loans to be distributed by banks. They went direct to consumer for the first time. They were administering loans directly to end customers. So it was an agency that was actually going through their own kind of high growth startup experience. And I unknowingly kind of went in really as a data scientist to help them figure out where the PPP loans are going in real time.

But it was one of those moments where I realized, wow, my startup experience is actually really relevant in government during the pandemic. And I think that realizing that I could make a difference, realizing that what I have done in the past was actually crucial during the pandemic also compelled me to stay. So I served for the next two tranches of the Paycheck Protection Program basically until the end of that administration. And then I went back home to California to serve the city of San Jose as chief data officer. And a lot of what I was doing in San Jose was enabling local government's ability to deliver services in a remote first setting. Danny, you were talking about how corporations aren't unsure how to drive efficiency in a remote setting.

When I was in San Jose, we didn't have enough laptops for every single employee and intuitively that makes sense because the city hall local government model has historically relied on in-person customer delivery. And so when the pandemic hit, schools were shut down, and government buildings were shut down, right? Everything from DMV to the registrar's office was shut down just because you shut down the initial point of customer contact doesn't change the demand. If anything, demand only increased. And so a lot of what we were doing at the Mayor's Office of Technology Innovation, MOTI, was building out the capacity to still deliver those services in digital first way. And we did that with a combination of bring really, really smart and talented folks into government. So we procured talent very, very aggressively. We use data in new and novel ways, but we also redefine what dual use technologies can mean. I worked on a team that was really thoughtful in the types of technologies that could actually help scale those government services, and we gave out government pilots and government contracts to companies that you wouldn't think were dual use.

Danny Crichton:

When you think about the future of government, I mean this is super fascinating because obviously there's a huge debate with the Trump administration coming back in. Do you bring everyone back to the office? There's a big return to office initiative going on. The federal government has hundreds of millions of square feet of office space it's not using right now. I saw a statistic unverified, we'll just call it completely made up, of something like only seven to 10% of federal office workers actually work in an office in a federal government building. And so a lot of people are remote. It's kind of a mess. We never really brought it back together. And at the same time, I've written for places like City Journal where AI technologies are still in their infancy in government were not really using them.

We're in some cases not even up to the point where we have databases. So you have to start with the data before you can get to the AI systems of record to systems of engagement to systems of intelligence or whatever the case may be. But we're still at the systems of record stage, particularly as you get into the municipal level. Where's the action today from your perspective? You just came out of the city and you're at a venture capital firm now with J2, but is there potential for this to change radically in the next couple of years? Is it still a slog? Do you think There's a couple of lighthouse cities or government agencies that are like, "Look, we just did it. Look at how great this is. Come follow us on this path"?

Christine Keung:

I appreciate the spirit of what Elon is trying to accomplish with DOGE, but what underpins his assertions is that the drivers of government efficiency is people not technology, a lot of the broader... And whether or not that's actually his intention, if you read a lot of the broader rhetoric around the DOGE program, people are talking about, oh my gosh, are folks going to lose their jobs? Is he going to furlough a third of the US government? And I actually think it's the opposite problem. And that's actually backed by the statistics where we're living in a state of the world where there's over a million job vacancies in state and local governments. And state and local governments represent the largest employment category in the United States. It's 24 million jobs in the US that folks can't actually fill right now. That's leading to state and local governments to spend over 11 billion annually in overtime pay.

So you're actually seeing this negative feedback loop where folks are actually burning now. Existing employees cannot meet the growing demand because they are on understaffed teams. They might not have the technology tools to provide them the leverage they need, and it's also driving up costs, right? City budgets have never been higher because of the cost of overtime pays. And so we actually see the opposite problems where I'm actually definitely, definitely afraid of people not wanting to work for the US government, and I'm worried that the existing workforce just don't have the leverage to deliver all the programs and services that the American people need.

Laurence Pevsner:

On the Riskgaming newsletter a couple of weeks ago, I think it was, Danny, I recommended a Substack that was essentially making exactly the same argument that if you want efficiency in government, you actually want more bureaucrats, not fewer bureaucrats that the hold up is the fact that we don't have enough people to do these jobs. I certainly felt this when I was at the State Department too. You can just tell everyone is overworked. And also if you're not a political appointee, the way to get these jobs is absolutely bonkers. You have to take the job description and basically copy and paste it the whole thing into your cover letter.

And then the longer your resume that you've just stuffed with other things, the better because there's literally just a checkbox where the people reviewing it have to by law so that they're not "discriminating," have to make sure that you meet every little requirement that precedes the job description, and it just makes no sense. And so they have an incredibly difficult time hiring even people they know are really qualified and would be really useful. And instead, then we get this crunch and then some people saying, "Hey, the government is not delivering the way we want to see them deliver." It's like actually the problem is because there's no one there to deliver the services to you.

Christine Keung:

And there's also issues with internal mobility too, because when we think about government efficiency let's just say from purely a talent perspective, it's not just about bringing great people there, it's retaining them, it's growing them, it's helping them evolve and adapt with the organization's needs. I led technical teams when I was in government. So oftentimes when I was at the city of San Jose, I was their very first chief data officer. I wasn't the very first person who worked with data. And so when I joined San Jose and I was tasked to really lead a data team to actually create insight and telemetry on our service operations, I started by actually identifying who was already doing data work within the city. And a lot of them were actually GIS analysts. And from there, it's realizing that, okay, a lot of these GIS analysts are technical, right? They're very, very good.

Laurence Pevsner:

Can you share with our audience what GIS stands for?

Christine Keung:

Yes. Geospatial information systems. So they're very, very good at using geospatial mapping software, which is a very specialized skill, but they might not have the skills that a data scientist who had worked at a Meta or a DoorDash would have, but it doesn't mean they can't grow into it. And so a lot has been written about some of the work that my team did in bringing new talent and especially new talent from the private sector into government. But I would say some of the proudest things that I was able to push when I was in city government was actually a new HR hiring processes, new HR code, new compensation categories, because it's not sustainable to only bring talent from the inside. It's really important to actually identify folks who all have that same potential or who already have that talent and help them grow into those roles.

Danny Crichton:

Let me ask you, I mean, one of the big challenges that I see is if you come from a high growth startup, there's not a mismatch between the product management team, the engineering team, the growth marketing team, the hiring team at those high growth moments, and you've been at a couple of them and I've been at one or two, you can feel the magic in the room. Everything is just flowing together really, really fast. Every hire is additive, everything is multiplying together. Government fundamentally constitutionally is designed to be competitive. It's designed to have different branches of government fighting each other. And so unlike the kind of simplicity you can develop in the private sector because you're like, "Look, let's edit the rules of our application, let's pick which features are highest priority," etc. You might say, "Look, the court has determined that every application needs to be reviewed by four people," and that is like a court case.

And now we need four times the amount of workforce in order to process applications than we need to be before. And then your city council or your state legislature or congress comes up and it develops a set of rules that look great on paper legally. But from an implementation technology data perspective, it's like, well, this doesn't line up with any ability to store in a database, because that wasn't what it was designed. It wasn't designed to say, "Okay, how would we store data in a really effective, efficient way?" I mean, we always talk about tech going into this and fixing it, but how does tech influence the policy world to say, "Hey, there's ways of doing, I don't know, clean environmental policy that actually could be win-win." And I'm annoying Laurence, he was just telling me I'm not allowed to use the term win-win at lunch today.

Laurence Pevsner:

It was the term that we tried to strip out of UN resolutions. It was the term that China would always put in there that secretly meant actually, this is just winning for China.

Danny Crichton:

Exactly. So we'll call it win-lose or something, that's America's mentality. But nonetheless, win-win. But is there a way for tech to sort of inform and say how do you design these systems to work better, architect them to be smarter so that you can reach your objectives, but oftentimes with more efficiency or performance?

Christine Keung:

So this is very much my personal take and not that of my employers. I actually really been the independence actually between the two. I see them as separate things. I think some of the perceived friction that can come from policy, and I actually don't consider myself a policy person. I am much more of a strategy data operations person. And even in government, I had a chance, I think I got to inform policy, but I'm not like Laurence, for example, where Laurence, I'd love to have this debate with you because your training is in policy versus my training is very much focused around business operations and the execution of policy. I think technology can actually help enact and realize the full potential of policies or admit whether the policies are good or bad. I think great technology, great processes have the power to actually just fully implement policies and perhaps decrease the cycle time of us knowing if it's working or not.

Perhaps great technology and great process and great people, most importantly, can help iterate and make policies better. But I don't know if I really want to live in a state of world where technology actually is informing those policy. I have to believe that or I choose to believe that when the courts say, "Hey, these before we do X, Y, and Z, it should be reviewed and approved by these four people that they have their reasons for wanting to put that friction in." Perhaps it's the type of friction that could lead to important outcomes that reduce bias, that ensure fairness. So the way I would think about if policies like that is the technology that could help realize the full spirit of that decision. And Laurence, I would love to get your take, especially since you have experience actually making policy.

Laurence Pevsner:

Yeah, I completely agree with you that technology is facilitative and not determinative that. And I think it's complementary to Danny's point that, okay, yes, you have the friction of government, but there's a difference between intentional friction and unintentional friction. Places where there was a policy that was put in place to create justice or fairness or equality or promote better outcomes. And then there's the actual practice of trying to enact that and technology is best served in the latter part.

So when I think about the comment I was making earlier about hiring processes and having to put way too many unnecessary things in the application, you could see how that's probably not what the policy goal was. It's the result of what they feel like they have to do with the tools available to them and as the freedoms that they feel like they have. Now, there's a two-pronged approach to that. One is on the policy side, clarifying that that's not what we mean or not what we want, but then there's also giving them tools so that they don't have to sort through every single resume and check all the boxes in that same way.

Christine Keung:

No, I just really appreciate that point and we will just stick to that example of the bingo card that is government hiring. I think that's just one interpretation of the policy, right? I am pretty sure the policy doesn't actually say the resume needs to be 10 pages long and have 80% of the keywords, but I see that as more whoever operate was trying to implement that policy unintentionally made that the status quo to which everybody who hired suddenly realized, "Oh, if I interpret this policy this way and I do X, Y, and Z, this is how I'm going to get a job at the State Department." And so what I like about technology is I also think it allows people who have to make these operational decisions like alternatives, whether it's accelerating reviews or providing alternative ways to underwrite their own qualifications. That's where I really see the true potential of technology. Yeah, I love the idea of it actually enforcing or shedding light on the quality of those policies faster.

Danny Crichton:

Let me pivot the conversation. So we've talked about local government, federal government a little bit here at technology, but then we get to international realm and back into US, China in the last week we've been having this huge debate around DeepSeek and who's ahead? Who's behind? We've got this whole arms race debate, et cetera, et cetera, et cetera. But you were part of our China electric vehicle game. Obviously it's not just about electric vehicles, it's not just about AI, it's not just about biotech. There is just a larger conversation around how does the US interact in a competitive environment with China, and that's one that I believe you have quite a few different views on. So I'm curious, things stand today, has anything changed with DeepSeek just dominating the headlines in every possible media publication for a couple days or that we're down to your sort of fundamental beliefs that have been existed for a while?

Christine Keung:

I wasn't very surprised by the quality of DeepSeek. And maybe this is where, I mean, Laurence and Danny, you guys know a bit about my family history and my personal background, so I'll just share. I am Chinese American, Italian, I am Italian by nationality. I am a daughter of Chinese immigrants and a naturalized US citizen. And so I think just a lot of the perspectives I bring now as a venture investor, as an investor of dual use technologies is informed by also just I guess my own identity as an immigrant and as somebody who is bicultural and multilingual, I am fluent in Chinese. I am a consumer of Chinese media, and I have been a user of Chinese tech. So I am somebody who have... And I've lived in China when I was working for the State Department.

I've been very aware of advances in the Chinese technology landscape for decades now at this point and have known how sophisticated their capabilities are. I actually remember when I was at the State Department, one of the more formative things that we got to do was actually we had a chance to see the DJI factories in Shenzhen. And this was like 10 years ago before, sorry, DJI was considered a dual use technologies. And at the time, I was just blown away by the quality of those consumer drones. I've been very aware of how good the tech is and actually how a lot of US tech companies have historically borrowed from the Chinese consumer feature set. Meta is a great example of that where when Meta launched some of their live video applications where you could do social listening, a lot of that was actually borrowed and taken from a consumer app called Bilibili.

So I've seen these trends over the year as someone who kind of straddles both worlds. And so what I want to say about DeepSeek is I think some folks are arguing this is our Sputnik moment in the Cold War with China, but we have to remember that Chinese developers are some of the most active contributors to open source code in the world, and that hasn't changed, right? DeepSeek, I'm not surprised, for example, that DeepSeek is an open source model that they've published their training methods.

Because if you go back even just years from now, if you just go on GitHub for example, and you look at who's actually contributing, Chinese developers are some of the most active contributors of GitHub code. I think Lux Capital is a investor in Hugging face. So if you go on Hugging Face and you look at the top 10 LLMs that are actually downloaded on Hugging face, a lot of them are actually open source Chinese models. And so China has always been a paradox in the sense that it's a closed technological ecosystem where in general, Chinese... Actually, no, I was about to say Chinese apps aren't really used by Americans. That is actually not true anymore. But in general, American apps are not really used by the Chinese population.

Laurence Pevsner:

Well, they're not allowed to be used.

Christine Keung:

They're not allowed to be used. So China has a closed technological ecosystem, yet they are the most active contributor of open source code and open source frameworks in the world. So it's just something that I've always been aware of. And so when DeepSeek came out, my first thought was actually that I was like, God, I actually felt like this was a product of US export controls. We created DeepSeek.

Laurence Pevsner:

So we're talking today, it's the day of the Lunar New Year.

Christine Keung:

Yes.

Laurence Pevsner:

Obviously very important in China, do you think-

Christine Keung:

Yes, I'm wearing red right now.

Laurence Pevsner:

Do you think there's no coincidence here that it happened to come out then? Do you think it's just all by chance in that sense? And I asked that in the context of a larger question of, okay, great, this is open source. Here at Lux, we're very supportive of open source. But I wonder how you think this fits into a broader narrative question of is it a strategic competition with China? Are we in a Cold War with China? How do you see that relationship evolving over time? What's your take? I'm wondering whether this is all like, "Oh, this just happens to be that they did this right now and that's just the way things are," or if this is part of a larger issue and debate we're having with China.

Christine Keung:

So we're definitely in a strategic competition with China, but I would argue that we've probably been in a strategic competition with China for the last 20 years, but this is sort of the first time where I actually think the big tech, at least the players in the US, were starting to worry. I think we've been in a race for a long time, but I think for a long time we just thought we were significantly ahead. I think the difference is just our perception of potential outcomes. I'm worried that we might lose. That is actually what keeps me up at night and I think what really motivates me as a dual use technology investor. What I'll say, I think what DeepSeek has really proven was their ability to create a high-quality model at a fraction of the cost.

It's basically a counterpoint to the Western companies with closed models that say that the only way we could improve our large language models is through scaling up. The US is sort of pushing a narrative that we need Project Stargate, we need billions of dollars into an AI infrastructure. We got to build a bunch of these nuclear coal plants next to these data centers. The US is pushing a very strong industrialization agenda, and I don't know if it has anything to do with the year of the snake, but I think what China has done is prove a counterpoint that you could actually create comparable quality of large language models at significantly less costs. And that that might be the path for democratizing access to that technology. And that's kind of scary, right? It's always scary to be proven wrong, but I also don't think it's bad from a societal perspective.

Danny Crichton:

I think the most important thing for me is how fast can you learn from the situation. So one of the topics we've had on the podcast and in the newsletter has been around radical learning going all the way back to Paul Collier, who runs a lot of global economics over at Oxford. A recent post we're doing on the German economy, the reverse issue of the German miracle turning into the German nightmare, and formerly a massive export economy now falling apart. And the big piece of that is actually there were all these signs that things weren't working, but then you decided to just go another decade in the same direction and then found yourself in the wilderness. To me, the question is, do we respond? Either at the industry level, at the political level, there's plenty of tools to be done here, it's not like we've lost.

The battle has not been won or lost by anyone, but do you have the competitive instincts? Are you ready to make relatively fast decisions? And in some cases, those decisions will not be aligned with the richest companies in the country, right? So if you are Sam Altman and your whole model is sort of this capital moat, and that's what he's sort of saying, someone on the internet actually quoted him when he was in India a couple months ago, and someone asked him, "What would happen if a model came out of nowhere that costs $10 million and beat you?"

And he was like, "If you don't have billions of dollars, you are irrelevant." And that's his attitude. That's what he's talking about in D.C. That's what he's talking about with corporate leaders. And to me, we've just disproven that. So the question is, do those folks revert back and say, "Well, that was wrong. What else are we getting wrong? Let's fix these policies or do we keep doubling down on the same old tactics?" And that to me is a very open question. That's a live discussion, both the federal level of China, it's at the same city level with you, whether it's San Jose or any city, do you keep going towards the past or do you look towards the future? A lot of places are going through that right now. But with that, Christine, thank you so much for joining us. And of course, Laurence joining us all the time.

Laurence Pevsner:

Thanks, Danny.

Christine Keung:

Thanks, Danny.

‍