Thanks all to those who joined our first live session of the summer last week with Jasmine Sun. Jasmine has become one of my favorite people to read on all things AI and tech. She is a journalist by trade but an anthropologist at heart, and she collects observations about the tech and policy communities in San Francisco, Washington, and China. We spoke in detail about her recent NYT piece, “Silicon Valley is Bracing for a Permanent Underclass,” which went viral.
Our interview covered:
The discourse around “permanent underclass” in tech circles, what’s real and what’s tongue-in-cheek
Political socialization and nihilism among Gen Z
Emerging AI Populism in DC and whether there is a difference on AI policy on the American political left and right
The view from China and whether the AI race is an American construct
Advice on how to create and find your voice at a time when words and content are cheap.
You can watch the full episode in the embedded player above or anywhere you get your podcasts.
You can also browse the archive on the new website: www.thecivicforum.com
A lightly edited transcript is included below. Thanks to Karen Rose Tank for sponsoring the episode.
Upcoming Event: Thursday, June 18, 1PM ET
This week I’ll be speaking with Noam Lupu and Nick Carnes about their new book, Keeping Workers Off the Ballot: How Democracy Undermines Working-Class Representation.
We’ll learn about why working class representation is important for policy outcomes, the structural reasons why it’s so low in the U.S., and what can be done to get more workers on the ballot. The event will be on Thursday at 1PM ET, link here: https://open.substack.com/live-stream/220981.
Thanks for spreading the word and supporting my work, however you choose to do so.
Best,
Rory
A Permanent Underclass?
Rory Truex: So Jasmine, just to get started, you recently had this piece in The New York Times — I’m going to paraphrase the title, but basically there are people in Silicon Valley that are bracing for, quote, “a permanent underclass.” Can you tell us what is meant by permanent underclass and are people being sincere when they talk about it? What does that idea mean in the Bay Area right now?
Jasmine Sun: The permanent underclass is an idea that I’ve noticed getting picked up traction over the last year or so in the Bay Area, especially in the AI world and the startup world, where a lot of folks expect that as a result of super powerful AI — artificial general intelligence, AI that can do everything a human can do and more. We might end up in a scenario where you have a permanent underclass, a permanent lower class of people, where workers have basically lost their leverage, their ability to earn income via wages, because the wealthy will simply be able to rent super-intelligent machines for a much lower price and at much higher quality than humans can.
The idea is that our class positions would be frozen in time far off in the future, when AI is as good as people are. Because workers wouldn’t be able to sell themselves, rich people will use the AIs, and those who are existing owners of capital — those who already have a lot of wealth to spend on machine labor — will be able to become more and more productive and earn more and more, whereas everybody else might be rendered useless and forced to live on welfare at best, or doing extremely menial physical labor tasks at worst.
I will say that the permanent underclass is definitely, in a lot of ways, a meme and a hyperbole. There’s a lot of Twitter jokes about it. I don’t want to say this is a very common, sincere belief that the majority of people in Silicon Valley have.
But the thing that I do think is very real about it are two things. One is, I realized as I was in conversations with people in the AI industry — startup founders, venture capitalists, whatever — that the vast majority of people I knew in the AI industry expected that AI would by default increase inequality and decrease economic mobility. It’s not the sense of “Oh yeah, we’re going to get super abundance for everyone,” which is sometimes what you hear from the lab leaders and the industry CEOs. Most people I speak to believe there will be mass job disruption, at least in the short term, and that this is going to centralize power and wealth in the people who already have money and power today.
That’s frankly a very bleak picture of society — one where there is increasing inequality, one where there are winners and losers to automation. So that’s what made me quite worried. And then the other thing is that even though it’s a meme, I do think memes really drive behavior. It’s very common that you’ll hear among young people, like people in their twenties in Silicon Valley, saying, “Oh, the reason I gotta take this job is ’cause I gotta escape the permanent underclass.”
There’s this mass economic anxiety where for a long time software engineers were feeling on top of the world. They’re all making six-figure salaries at twenty-two. Now because of the layoffs and because of AI, there’s an increasing sense that there’s a very small window to still build wealth and get good jobs. If you don’t get a job at Google or Anthropic or OpenAI tomorrow, maybe you’re screwed in a permanent sense. So I do see that driving a lot of behavior from young tech workers.
Rory Truex: When people talk about the permanent underclass, are they talking about white-collar displacement? We know that the CEO of Anthropic, Dario Amodei, has been on this press tour where he’s talked about — I forget the number, I think it’s 50% of entry-level white-collar jobs by 2030. I’m a professor; I think of my students trying to enter the labor force, and it’s very different from when I was young. The jobs aren’t there.
So we think about white-collar displacement, but then there’s also this talk of, down the line, humanoid robots empowered with AI or super-intelligence that can make your coffee or harvest crops or do whatever other jobs are out there. When the discourse is happening, is it primarily focused on this elite class of hyper-educated folks that are for the first time feeling the threat from automation, or is it both pieces? How do people talk about those two different aspects of it?
Jasmine Sun: I think it is both pieces, but the reason there’s so much anxiety right now about the permanent underclass is that it is these highly educated elite folks who never thought they would have to worry about being in any sort of underclass.
In fact, one of the most interesting things after publishing the New York Times article and seeing it get circulated on social media — it was the first one of my pieces to go Instagram viral, which I didn’t realize that long-form articles could go Instagram viral. There were two responses I got most on Instagram, which are very different from the kinds of responses I get on Substack or from people I know in Silicon Valley. One: America already has an underclass. Don’t you know that? We’re all in the underclass — so, like, surprise tech workers, you’re vulnerable too. And then number two: we should kill them, get the pitchforks out. And I was like, “Oh my God.”
So I do think that a lot of it is because AI is affecting cognitive labor and it’s inverting the assumptions about whose jobs are secure and whose jobs are not. It almost creates a sense of class solidarity or this realization for even highly paid tech employees that workers are workers. Your boss can fire you if they feel like it, and you don’t have that much leverage in that scenario. In a sense, you could say it’s the first time a lot of people are acquiring a kind of class consciousness in the Valley.
Rory Truex: Let’s return to that theme of AI populism, which you’ve also written about. It will be interesting that it’s going to create cross-class bonds, even cross-party bonds from the right and the left. There’s a lot of different ways people are talking about it, but it is affecting so many people.
Reporting Inside Tech
Rory Truex: You spoke a little bit about the public reaction. Part of your approach is to embed yourself in these communities and pick up things at events and social gatherings and reflect those back to the rest of us.
Did you receive a backlash within the tech community for what you wrote? Do people start to perceive you with some suspicion? You’ve become more prominent over time, which is excellent and well deserved. Do you find that that’s changing how you’re able to do the work you set out to do?
Jasmine Sun: The first thing I think of is — I went to this rationalist AI conference a couple weeks ago in Berkeley. It was an off-the-record sort of Chatham House scenario, which I knew and I was happy to do. But at the very beginning they did an orientation and they were like, “Be careful, there are journalists here.” And then they showed a picture of my specific badge. It was like, “Jasmine Sun, press.” And I was like, “Oh my God, did you really have to do me like that?” But everyone was actually very nice to me.
My career started out in the tech industry. I worked at Substack as a product manager mostly, and I was there for four years. So a lot of my friends and my social life is people who work in the tech industry and in AI. I think that helps because I get what it’s like to work at a company, to have goals to meet and to build technology products and have people yell at you about the technology products. I think that’s made me maybe a little bit more nuanced about the way I try to represent the tech industry. I have social trust with a lot of the people who I cover as well.
The other thing, frankly, is that with this issue in particular, there’s actually a situation where a lot of people in AI wish they could be more vocal about some of the risks they see. In my case I’m talking about the economic stuff, but I think people feel the same way about, say, frontier safety risks. They actually can’t do it because they are working at these companies. I will have these conversations with people working at VCs or AI labs where they’re like, “I am super freaking worried about how an average 22-year-old is going to fare in the future. I want you to be writing this story. I want you to be getting people on the record so you can tell the world and motivate policy action.” But they personally can’t do it because they work at companies that restrict what they can say.
I would literally have a conversation with someone who’s like, “Yeah, I’m talking to all these CEOs and they’re all telling me that they’re laying off people to replace them with AI, and they’re starting off by laying off contract workers overseas because that’s not going to create as much backlash at the beginning. But they’re not going to go on the record, Jasmine. You should know it’s happening because you should write about it, but they aren’t going to go on the record.”
For the most part, most people I spoke to for the story felt like it was an accurate representation of at least one of the major subcurrents going on in AI culture right now, and a lot of the worries people have. There were definitely a few people — Marc Andreessen did some tweets where he was like, “The permanent underclass is a crazy idea. Jobs doomerism is bad.” There was a little wave of anti-jobs-apocalypse pieces after the Times piece came out where people were saying, “AI’s going to create more jobs than ever. Everyone who believes in a permanent underclass is stupid.” But I’m mostly characterizing the belief. I’m not saying that I believe in a permanent underclass, so I think it’s relatively fine.
Separating Hype from Reality
Rory Truex: I live in Philadelphia, and I love my city, but I wouldn’t say we’re at the forefront of technological innovation. We have great healthcare, we have other things going for us. So when you’re that degree removed from the debate, one of the things that’s difficult is it’s hard to separate the hype from reality. A lot of these AI companies, the CEOs, people like Andreessen and others, are trying to shape the narrative, but they have such a clear conflict of interest. It’s clear as day they want to make this technology seem like it’s godlike, super-intelligence, AGI, all this discussion, and hype up the job creation aspects of it. But that’s partly because they want the valuations and other things.
In terms of the AI jobs apocalypse, the permanent underclass — you wrote about this in your New York Times piece — most economists to date, from what I can tell, believe there will be disruption, but we’re not headed towards 40% permanent unemployment, that kind of thing. So for you, as someone who operates in this space, how do you separate the hype and the posturing from the facts? Who are you listening to? Do you find yourself turning to outside experts, academics, other journalists? For me, just the conflicts of interest are so challenging to discern.
Jasmine Sun: I think it’s interesting. For sure, I spent a lot of time for this piece talking to folks in Silicon Valley about their beliefs, but they’re not economists, and as you said they also have conflicts of interest. So I do try to make sure I’m talking to academic experts — I talked to a bunch of economists for this, outside policy experts — and try to apply my own sense of scrutiny to it.
My sense, for what it’s worth, is that Dario Amodei and a lot of the folks at Anthropic, for example, genuinely believe a major labor disruption is coming. Dario going on a media tour and talking about the white-collar bloodbath is not making anyone like Anthropic more. As the IPOs come, both Sam Altman and Dario have pulled back on this language. That’s when you know I don’t think they were doing it for marketing. I think they actually believed that was true, because now that they need public investors to like them and they’re seeing rising waves of AI populism and they’re becoming much more worried about political backlash blocking their data centers and whatever, they’ve actually moderated a lot of their language about the economic disruption. That tells me they do believe the economic disruption was coming. That’s what I heard in private conversations.
What I will say is that the insularity of the AI industry means that even while they may believe in these slightly apocalyptic scenarios, they may be wrong about those beliefs even while they earnestly believe them. The obvious thing here is that a lot of people’s assumptions about AI progress are based on coding agent progress, and coding agents are in fact really good. They can do a lot of things that junior software engineers used to do. It’s true that lots of people don’t write lines of code anymore, and it’s completely transformed the profession.
But the reasons that coding is easy to automate relatively don’t really apply to most jobs. If the only job you’ve ever had is a software job, you don’t really realize the way in which the gains on coding don’t actually transfer to other domains. In coding, you have tons of training data in open-source GitHub repositories. All of the context lives inside your code base inside the computer. There’s not a bunch of tacit knowledge in people’s heads. It’s something you can do remotely. It’s not a physical-world task. It’s not that social. It’s not like you’re constantly having to go places in the world and see people and persuade them. And it’s very verifiable. You can tell if the code runs or not. You can tell if something’s more efficient or not.
I do think that most people in the AI industry are overestimating how easy other jobs will be to automate. That’s coming from the insularity, but it’s not necessarily coming just from marketing hype, I would say.
Gen Z Attitudes
Rory Truex: I wanted to touch quickly on Gen Z. I don’t mean to make you into the voice of your generation… you’re a voice who happens to be of that generation. How is Gen Z coping with this moment? One hears stories of a cynicism. In one of your early pieces, I forget the phrase, so I might butcher it — but a culture of “I kind of gotta get mine. I gotta get it now because God knows what the future’s going to look like. The system’s a bit rigged. The country’s kind of a sham, and this trickles down from DC and Trump and all these things.” How would you characterize the mood, or the approach to work and life, from people who are entering into this really strange economic and political environment?
Jasmine Sun: I’m really interested in a thing I characterize as Zoomer nihilism, basically. Which is, of course, a sweeping characterization of my Gen Z peers. But there’s a bunch of broad cultural trends and polls and stats you can look at together. I think of everything from trust in institutions has fallen off a lot. It’s across all of American society, but I think especially Gen Z is very low trust in corporations, very low trust in the government.
If you were 22 today, Trump has been running for president or president since you were, I think, like 10 years old, which means that your story of how to become the most powerful man in the world is to be a grift and a fraud and whatever, right? I actually feel grateful that I kind of remember Obama, and I kind of remember W — who, for all his flaws, was a normal president or something. But I think the kids who are five, 10 years younger than me do not remember anything else except a world where the pathway to be the most powerful person in the world is to be extremely corrupted. So I think that affects people.
I look at the rise of a lot of populist candidates on both the right and the left, in the sense that, again, the game’s rigged, so let’s support someone who’s going to break down the system. You look at the rise of sports gambling, these “get my bag” economy things. Like, it doesn’t matter about this hard-work thing. That’s not going to work out, so I might as well bet on crypto. I might as well bet on sports gambling. Just try to get as much as I can right now, because who knows when stuff’s going to go under again.
I do think the mood in a lot of Gen Z is this kind of precarious, cynical, not-very-values-based sense of: we just gotta save ourselves because the system is not going to save us and no one is coming to save us. I sympathize with that, because the system is broken in a bunch of ways. But I do think this influences some of the backlash against AI billionaires and whatever.
Rory Truex: Just to indulge my inner nerd a little bit… your point about political socialization is spot on. That window of time from like 10 to 25 is so critical for how we view the world. W and Obama were kind of my formative socialization years, and that gave me some reservoir of hope and a little trust that’s being drawn down on now. But your point about how people are being raised in an environment where all bets are off and bad behavior is being rewarded is insightful. I don’t blame people for feeling the way they do and trying to cope with it.
Does AI Have a PR Problem?
Rory Truex: We’ve danced around this idea of backlash, and you’ve also written a fair amount about how AI has a bit of a PR problem. Let me summarize my sense of it, and please correct me and elaborate.
These folks are creating a technology that has the potential for amazing gains across a lot of domains and research, but it also has the capacity to displace a large number of workers and potentially even cause an extinction-level threat — different estimates: 10, 20, 25%. They’re asking the American public to support this, go along with this, almost entirely on the grounds that if we don’t do it, somebody else will — one of the bad guys, Russia, most likely some company in China will do it. It’s a fear-based narrative that we’re more virtuous, and the cat’s out of the bag. It can’t be undone, it can’t be slowed. It’s not really a compelling pitch, and I think we’re seeing that AI populism emerge.
You’ve written about Bernie’s campaign against data centers. Josh Hawley on the other side is now being very active on AI. You recently visited DC. Can you tell us a little bit about that trip and what you learned about the debate at that time?
Jasmine Sun: I mostly don’t cover the industry more so than I cover the policy world, but I try to go to DC roughly once a quarter just to talk to people about AI policy, learn what’s going on out there.
One big shift I noticed, probably starting in 2026, is that the policy and political conversation has broadened a lot. Until 2025, every time I talked to AI policy people in DC, it was a fairly narrow technocratic community, mostly concerned with the national security story — the China competition stuff, frontier safety, chip controls, that whole side of the story. But the majority of political interest groups in DC were not engaging very much with AI, didn’t really care about it, didn’t really have an AI strategy.
The thing I noticed starting this year, when I went in February and March, was that many more political groups and entities were starting to get engaged in AI, mostly in a very anti-AI way. Whether you were a states’ rights organization angry at the preemption fight, a family-first social conservative scared about how AI companions are ruining kids’ mental health, the unions worried about automation, or a neo-Brandeisian worried about the concentration of corporate power — all these other interest groups in DC have realized that AI is a big enough force in the economy and politics right now that they all gotta have an AI strategy. You’re also starting to see these very strange-bedfellows coalitions emerge across these groups, sometimes even across party lines, to push back against AI — whether through pursuing things like a data center moratorium, or through the preemption fight to regulate AI really aggressively.
The other thing I was thinking about when I came up with the term “AI populism” to try to put my finger on what was going on was, I was trying to understand what the gap was between AI enthusiasts and AI critics, because it always felt like people were talking past each other.
You look at something like Waymo. Waymo’s proponents — for what it’s worth, I love Waymo, I’m a Waymo proponent — will cite all the safety stats. Waymo is much, much safer than a human driver. It’s pleasant. People love it. That stuff is broadly true. And yet Waymo faces a ton of opposition in a lot of the cities it’s trying to expand into. Those people aren’t really disagreeing with the substance of the argument that Waymo drivers might be safer than human drivers. They’re often making an argument about labor impacts. They’re making an argument about corporate power.
My sense is that a lot of the folks who have recently become engaged in the AI debate, who oppose AI, are not necessarily talking about the specifics of the technology. The frontier safety people are very concerned with exactly how many flops is this model. But the labor folks or the kid-safety folks or the neo-Brandeisians, they’re actually worried that a small number of corporations in Silicon Valley and a small number of AI billionaires are able to have this scale of influence over our society and economy. It’s almost an outgrowth of anti-elite sentiment, or anti-billionaire sentiment, or anti-corporate sentiment, rather than an argument about the characteristics of the technology.
It’s a nitpicky definition, but that to me explained a lot of why my friends out here in Silicon Valley would be like, “But the Waymos are so safe, and ChatGPT is so useful.” And yet other people could still be like, “Yeah, we know ChatGPT is a better Google, but we still oppose it because we don’t like the way it’s transforming our society.”
Rory Truex: In the populist parts of our political spectrum, you have left populists and right populists. They have a similar diagnosis about the issues, but different prescriptions. On the right, it’s focused on immigration and closing off borders. On the left, it’s focused more on redistribution. You mentioned strange bedfellows — do you see a left-wing version of AI populism and a right-wing version of AI populism, or is everybody just against data centers and against the consolidation of economic power? Does that question make sense?
Jasmine Sun: I do think you see both. I think the political lines are still ironing themselves out and I wouldn’t say they feel very set right now. The data center stuff is very bipartisan — if you look at the polling in Michigan around data centers, you almost see the exact same level of opposition between Democrats and Republicans. So there are certain issues — mostly the stop-AI issues — that Democrats and Republicans can really agree on when it comes to opposing AI.
I do think there are other parts of the debate that are going to have more left and right flavors. When it comes to raising taxes on corporations and capital gains — Elizabeth Warren has put out an op-ed in Time recently saying we have to tax the AI companies because of these worries. That’s a very left-flavored version of the thing. The left is probably going to talk more about civil rights type issues, though those have decreased in concern relative to the story three or four years ago.
The right tends to be somewhat more focused on the kids-safety story because social conservatives are a big part of the coalition. A lot of the groups that were opposing Facebook and big tech for harm to teen mental health have now switched their focus over, or are expanding their focus, to AI and mental health as well. I expect kids-safety legislation to be a bigger focus of the right.
But I think it is in some ways more bipartisan than one might expect. One of the things that surprised me most was literally being in meeting rooms where you have a very strong progressive and a Steve Bannon guy at the table next to each other saying, “Yeah, we disagree on a lot, but we’re going to work together.” I remember one of the lines I heard that I thought was very funny: “I wouldn’t send my kids over to a playdate at the Polychaele, but I’m willing to build coalitions.”
Rory Truex: That’s that ethnographer’s instinct you have that gets us those juicy quotes.
The US-China AI Race
We’ve moved from San Francisco to DC. I wanted to take a quick pit stop in China, which is another place you visited in the last year. There’s a parallel debate going on in Chinese society. The competition around AI is quite vigorous, and a lot of the folks that really are pushing acceleration largely are doing it on the grounds of US-China competition. They view it almost like a race for the atomic bomb. What did you observe? Tell us a little bit about that trip. Is there a cultural difference about how people are thinking about AI? Sometimes China’s labeled as a very pro-AI country. Did you sense that when you were there?
Jasmine Sun: I’m no China expert. I’ve been about once a year for the past three years, and I usually try to talk to some tech and AI people there, as well as I have family there, so I’ll talk to family to get another barometer.
In China there’s a stronger separation between the central government and the party versus what industry or civil society actors might think or be interested in. I’m not spending any time with party officials, so I’m mostly talking to AI researchers or startup founders or people in the tech community in China.
Among those folks, among the Chinese AI industry and tech industry, I would say the sense of a race against the US is much less salient than the US’ sense of a race against China. The race in many ways feels a bit more like an American invention than a Chinese invention. I do think there is a sense of solidarity among a lot of the open-source labs in China to work together in a more collaborative way.
They both respect, for example, companies like Anthropic. Everyone loved Claude Code — which they’re not supposed to be using, but every single lab we talked to (we talked to most of the Chinese AI labs) were like, “We love Claude Code, it’s the best product ever.” But at the same time, they would say, “It’s a little frustrating that Dario Amodei is so arrogant about China,” because he sort of tried to cut off the Chinese labs from accessing Claude Code and other Anthropic services.
I think of American AI as having almost three eras. There’s an academic and research era during the 2010s mostly. There’s the commercial era of AI that kicks off with ChatGPT in 2022. And I think we’re just starting to enter the political, geopolitical era of AI — that Mythos, the Pentagon drama, all of this stuff is really ushering us into.
When I talk to Chinese AI people, it still feels like they’re in the academic era — there’s a lot of collaboration between labs. There are some efforts at commercialization, but not that many. It still feels very much like a research culture.
On the public side — this was the thing I was thinking about during this trip, because I was finishing up my New York Times piece when I was in China — like you said, there’s sometimes an impression that China is a more techno-optimistic society. People will look at polls where Chinese people seem more positive on AI’s benefits.
I think this is somewhat true, but I would frame it more as a kind of techno-determinist pragmatism. The sentiment I heard when I talked to people about AI — and like random people, not just people in the industry — was not necessarily, “Oh, I’m so excited, I love AI.” It was much more that there was no sense of why you would resist AI. Technology and its progress is seen as more of a given. It’s seen as part of this inevitable process of modernization. And also there’s not a culture of resistance in China. There’s not a culture of protest very much.
I’m not saying there are no efforts to resist new technologies, but the idea of talking to your boss and being like, “Oh yeah, I don’t want to use AI” is not seen as a common thing. First of all, there are a million people behind you in line who will take your job if you don’t want it. And in recent history, technology has mostly brought economic benefits to Chinese people. They see that, they feel that very clearly. There’s an assumption that if we’re embracing the AI thing now, presumably at some point this is going to bring economic benefits to us too.
And then the individual thing — the focus is much more on: how do I use AI to improve my own economic prospects at work, to be the best worker I can be — because, again, if I don’t compete, somebody else using AI is going to out-compete me in the job force — rather than attempting to resist it in any form.
Rory Truex: Thank you for that answer. I thought this piece you got at, about the salience of the competition and the arms-race nature of the competition maybe being higher in the US than in China — I thought that was quite striking. I’m still trying to wrap my head around this. I’m not an AI person.
I’m a domestic Chinese politics person, and my sense at this point is that AI has great opportunities and the potential is amazing, but the risks are quite real, and at a minimum it’s going to have to be very heavily regulated domestically and then eventually internationally. If it is an extinction-level threat, if misaligned AI is what people say it is, at some point US and Chinese researchers are really going to have to collaborate quite closely on setting standards — in the same way that nukes were developed and then highly regulated and we’ve avoided nuclear war.
Did that idea come up in your visit? Is there an appetite for that collaboration in terms of safety and alignment? Is that entering the discourse in China from what you can tell?
Jasmine Sun: On the level of people in the tech industry, I would say no. Among people who are at companies, AI researchers, engineers, startup founders, we did ask questions about safety. Most people were fairly dismissive, or they would say it’s important, but clearly in kind of a “Oh, of course it’s important” way. These labs are very compute-constrained. Given they have orders of magnitude less compute than American AI labs do, they’re not spending nearly as much of it, of course, on doing safety and alignment research.
They’re also behind the frontier, which means safety and alignment research is probably seen as a lower priority compared to if you’re building frontier models. My impression is that the main way Chinese AI companies think about safety is as a matter of compliance — compliance with government policy, whether that’s speech policy, whether that’s “turn off the LLMs during Gaokao season” or whatever it is.
So I would say Chinese researchers and engineers and founders don’t seem very engaged in safety. Of course, the policy conversation, the academic conversation is a little bit more open, and I know there are track-two dialogues going on.
One of my senses — and you would know more about this than me — is that there is a somewhat different conception of what safety is in China. Safety is a very broad term. In America, we might be thinking primarily about rogue AI and misalignment. But the little that I know about Chinese AI policy domestically is that it seems relatively less focused on the misalignment stuff and more focused on, whether it’s the speech stuff — don’t talk about these issues. There is some increasing concern, it seems like, about labor displacement, since the government cares a lot about social stability and keeping people employed. And it does look like there’s attention to things like CBRN risks — collaboration on making sure that the AIs don’t allow for the creation of biological weapons. I think there’s a lot of room for collaboration on CBRN risks internationally.
Are you hearing from the policy and academic community a lot of it focused on the extinction-level risks?
Rory Truex: It’s interesting. Again, this is not my field, but I’m realizing I need to begin to learn more about it. I say that because in the last probably month I’ve gotten several emails from people working on this issue — who maybe are in the school of what people call the decelerationists, or the pause school of thought. There’s this question of: would China cooperate? Is cooperation possible? How can we cooperate when the relationship is so fraught? Trump says Xi Jinping is his good friend. Well, Xi Jinping doesn’t really have friends. How can we cooperate when cooperation is at an all-time low, and is there an appetite for it? That’s why I asked you the question, because I’m trying to learn.
Jasmine Sun: So I don’t have a great point of view. I think you should engage in it because, frankly, one of my perspectives and one of my little bugbears is that I think the AI safety community has a very poor understanding of Chinese politics, and just has a very poor understanding of China in general, and makes a lot of naive assumptions.
This is one of my annoyances I’ve had for several years. For many years, China was the boogeyman: “We hate China, and so everything we do on AI has to be about China’s going to blow up the world with their AIs, so we gotta be really careful about everything.” And then I have noticed in the last six months a turnaround where a lot of people became much more open to the idea of an international pause. All of a sudden, the same people who were huge China hawks on AI in particular, without having any China background, suddenly were like, “Oh yeah, China loves to cooperate. China’s very concerned about rogue AI.” And if you would look at the policy things they were citing, yes, it’d be some Chinese policy document talking about AI safety, but maybe they’re talking about speech regulations or labor issues or something, and it would be characterized as: “Don’t you see that they actually care about loss of control?”
My sense is there’s actually something going on where there are folks in the AI safety policy community who are trying to manufacture a consensus that may or may not actually exist, by reading the tea leaves too much or too hard in China to be like, “Look, the Chinese government really cares about this,” when it’s, “Oh, there’s one mention of data control on page 50 of a 70-page document.”
I’m a little bit concerned about the safety community’s honest ability to understand Chinese policy. I don’t know anything about it either, but you should help them out.
Rory Truex: There is a tendency, which you’re getting at, in the China-watching community to impute the intentions of the Chinese government based on small things. It’s such an opaque system that people can basically cherry-pick whatever quote they want to make the argument they want. We often don’t have enough humility, honestly, about what we don’t know about the system.
I love Kyle Chan — he’s a great writer. Graham Webster has a Substack. So there are people who are thinking about it. But I agree, I think some infusion of China expertise in the debate would be good.
Jasmine’s Advice On Writing and Creating
Rory Truex: I wanted to tee you up for kind of a fun question. We’ve traveled the world. I want to come back home. You give out good advice. You recently had a piece to the class of 2026. You’ve also recently done a session about what it means to have a voice in this moment, to be a creator — what type of voice and insight will be valued at a time where words are cheap.
Do you have any advice to people? We’ll start with creators. Can you tell us a little bit about how you approach creating, and your advice to people who are trying to emulate you or do something related to what you’ve done? You’ve built this amazing platform in such a short period of time, and it is remarkable and well deserved. I really enjoy your work. So give us some tips, Jasmine.
Jasmine Sun: Oh man. Thank you. I was about to be like, “I don’t have any tips,” and then I was like, “No, I did have tips recently.”
One thing I think a lot about is the balance between creation and distribution. I have a lot of writer friends who hate marketing and hate posting on social media and hate all of that stuff. And I also know people who I think spend too much time on growth hacks and marketing — how do I phrase the tweet the exact right way, how do I max out my Substack recommendations — and they’re just not focusing enough on doing good work. Finding the right balance for oneself between those two things is really important.
For me, I think of the core of my work as: I need to write good essays. If I don’t write good essays and I don’t write good articles, and I’m not saying something new and I’m not saying something interesting, literally nothing else matters. So the starting point is, am I producing good work?
But given I’ve produced a piece of work I’m proud of, I do want to shout it from the rooftops, and I want to shout it from the rooftops in any medium that is available to me. I spent two and a half months reporting out that permanent underclass story. It took probably cumulatively 250 hours. I have 350 pages of notes. I went through so many drafts. That was because I wanted to produce something rigorous and nuanced and really as well-researched as I could.
But once I put that out there, given I’ve made this gigantic time investment, I basically blocked out the whole next month of: I am going to go on podcasts, I’m going to go to conferences, I’m going to take meetings. I really want to make sure I’m not a purist that’s like, “You have to consume my reporting in long-form New York Times form.” I know that people get their information through a lot of sources these days — whether it’s tweets or podcasts or in-person meetings or whatever. Now that I’ve done the work, I try to invest a lot of time in making sure it goes out on all of the mediums and channels into different types of communities that I can reach.
The other thing is trying to have — it’s always hard to tell someone how to develop a voice and a style — but to whatever extent people can figure out the voice and the style that is theirs, I think that’s really important, because LLMs have basically created a world in which polish and standardization is no longer a proxy for quality. There was an argument maybe before that if an article was extremely grammatically polished and every sentence was perfect, it meant someone spent a lot of time on it, and you could trust that maybe it was also really well-researched. But these days you can do that in a second, literally.
So really the thing that proves there’s a life behind the voice — that there’s a human, there’s a very unique point of view — is actually being a little bit more irreverent, being more weird, doing whatever format you want to do. Remembering the point of being a creator is: people want to know there’s a specific individual life behind the voice. It’s not just about presenting the information, but show the proof of work, show how you did the reporting, show what your personality is. Bring the whole process into it. Show that you’re not just something an LLM could produce.
Rory Truex: Something I’ve learned from you, I guess two things. One is the human piece of it is so important, and I think about that with my teaching as well. LLMs and all these things can summarize information quickly, and so if that’s what you’re doing, you’re going to blend into the background. But if you’re bringing your own experience to whatever you’re doing — whether it’s your writing, your teaching, whatever — I think that’s become really critical, and I picked that up from you.
The other thing I appreciate about what you just said is so much of the advice about social media is about volume. You go on YouTube, it’s people with hundreds and hundreds of videos just rolling the algorithmic dice. But just slowing down and producing something rich and good as opposed to cranking stuff out, I think is a really valuable strategy in a time where content is so superfluous. Making sure that when your name is on something, it means something — it means it’s high quality and well-researched. I really appreciate that about what you do.
Jasmine Sun: When I worked at Substack, we would look at the data about how people — because people are always like, “How much do I post? What day of the week do I post?” — and it is true that if you look at the data, posting three or four times a week is optimal for growth. But I don’t think people should look at those aggregates.
I thought about it. First of all, I am a slow writer. I literally cannot produce three essays a week even if I really wanted to. I even tried to get myself to produce one essay a week for a period of time. But I was just really unhappy with everything I produced, because I didn’t feel like they were as well-researched and nuanced as I wanted them to be — because that’s just how much time I take to think about things.
So I actually write much less than most people who are full-time writers do. I write maybe once every two to three weeks. But I think I justify it, one, by spending a lot of effort on distribution and talking about the thing in different formats to make sure it creates a lot of reach.
It was also because I looked at my stats after my first six months last year. There was a time when I was producing once a week, and then over the summer I think I wrote one essay a month for a few months. Writing one big hit, one really viral piece, would bring 10 times as many subscribers and new growth and new audience as writing one kind of mediocre, “I just jotted it down in a couple days” type piece.
It made me realize that not only do I prefer spending longer on pieces, but assuming the pieces I do produce are really high quality and really in-depth, and they are hits that are going to get broader reach, then it’s actually worth it for me to slow down a little bit and produce higher-quality work. That gave me a lot of reassurance that even though it wasn’t actually the growth advice that might be most data-backed, assuming the work’s really high quality, I think you can slow down.
