The Prof G Pod with Scott Galloway - China Decode: The World Now Views China More Favorably Than the U.S. (ft. Selina Xu)

Episode Date: July 21, 2026

Alice Han and James Kynge start with Trump's claim that China ran an influence operation using voter files from over 220 million Americans, and why Beijing is calling it fabricated. They dig into what the underlying intelligence actually shows, how it lands just two months before Trump's planned visit to China, and what it means that global public opinion is now tilting toward Xi over Trump. Then, Selina Xu, head of China research for former Google CEO Eric Schmidt, joins to talk about her recent trip to China with Schmidt, where the U.S. and China each have the edge in the AI race, how culture and adoption differ between the two countries, and whether Beijing's pushback against emotional dependence on AI is a sign of things to come. Subscribe to China Decode on Substack for weekly analysis, livestreams, and deep dives into the biggest story shaping the global economy: chinadecode.profgmedia.com. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Support for the show comes from MongoDB. AI-assisted and agentic coding is helping you build faster than ever. But if your data layer is a bottleneck, what's the point? Instead of wrestling with rigid schemas or translating data formats, MongoDB's native data model mirrors the language LLMs already speak. It ships at the speed of AI, is ACID compliant, and scales to handle massive Fortune 500 workloads. Ask any developer, it's a great database.

Start building Amogodib.com slash AI. I'm pretty confident talking into a mic. Hey, I'm doing it right now. But home projects, I second guess everything. Is that noise normal? Is that water damage? And who should I even call? That's where thumbtack comes in.

Upload a photo or voice note, and their AI-powered search helps diagnose the issue and match you with the right top-rated local pro. Instead of second-guessing or searching for, hours, you get clarity and can hire the Wright Pro with confidence. For your next home project, try Thumbtack. They know homes. Hire the Wright Pro today.

The Hulu original series Furious is coming to Disney Plus, starring Emmy Rossum. Furious follows FBI agent Alice Black on the hunt for a mysterious and calculating serial killer. Both walk their own paths toward justice, and as their lives start to intertwine, the line between right and wrong begins to blur. Don't miss the three-episode premiere of the Hulu original series Furious on July 27th, only on Hulu on Disney Plus. Welcome to China Decode. I'm Alice Han. And I'm James King. In today's episode of China Decode, we are discussing Trump's claims that China interfered in U.S. elections, and then we'll chat

with Selena Shrew on the latest on the AI race. That's all coming up, but first, Let's do a quick check-in with how the Chinese markets are starting off the week. On Monday, markets were mixed. The Shanghai composite rose 0.85%, while the tech-heavy Shenzhen component fell 0.71%. The CSI-300 was up 1.53%. Chinese banks performed well with ICBC up 2.25% and the Agricultural Bank of China up 3.44%. The People's Bank of China kept rates unchanged, which have been at record lows, for 14 consecutive months.

All right, let's get right into it. President Donald Trump's new claims that China may have interfered in the 2020 U.S. elections are causing backlash within China. China's foreign ministry called the claims quote-unquote entirely fabricated and aimed at vilifying China. It threatens to upset any advancements the world's powers have made with each other this year.

The news also comes as Pew Polling finds that China is now viewed more favorably around the world than the U.S. is, and that more people have faith in Shijingping than Donald Trump. James, before I get into the startling polling data, which is its own kind of headline story, let's do a bit of an update in the Beijing-Washington relationship. Things look to be pretty smooth sailing from Trump's visit to Beijing in May, but I think we're starting to see some hiccups along the way, largely with this now allegation from Washington that China has been involved in election meddling,

mainly political interference in the 2020 election. And Trump is saying that dozens of significant CIA and NSA records show that China was targeting voters during the election. And although this isn't seen as vote tampering, it's seen as vote influencing. Now, the Chinese foreign ministry has come out and said that this is basically a fabricate, and that China has not been involved in any kind of election interference. But I think the big question I'm still puzzling over is why is Trump picking up on this issue or relitigating an issue six years ago?

Obviously, this is a source topic for him because he lost that 2020 election. But I don't understand why he's bringing up this topic now two months before she is supposed to go to the US. Well, that's a really good question, Alice. I mean, to me, this is more of a U.S. story that concerns China than a China story per se. But I think, as you've already alluded to, the key point here for China is, will this allegation by Trump sink the superpower truce that has ensued after the summit between Trump and Chinese leader Xi Jinping in May this year? At that time, Xi and Trump agreed to build a, quote, constructive China-U.S. relationship of strategic stability. And that was really a key agreement because that meant that the two sides will have this as their guiding framework for the next three years and beyond

and move away from pretty much what was all-out suspicion and rivalry that characterized the relationship from probably about 2019 onwards. So that was a real moment, at least I regarded it as a as a, as a, real moment in May, and I think a lot of other people did too. So now Trump is coming up with this apparently very serious allegation of election meddling. And so it does raise the question, is this going to bring a premature end to the superpower truce that we saw agreed to in May? And my sense of this is, tentatively, of course, because you never really know. But my sense of this is that Trump's statement is probably more about domestic politics in the US than it is about trying to torpedo the slightly better relationship that China now has with the US.

And I reckon that I can't really give any overwhelming support. court of evidence to back that up, but I note that Xi Jinping is still invited to visit the US on September the 24th this year, and Trump is still apparently considering attending the APEC summit in Xinjiang in November. So those two facts tend to me to suggest that, you know, the relationship is not headed for the deep freeze. And the other thing is that the White House has not yet given any words on why or what implications this allegation of election meddling has for the US-China relationship. And my guess is that if the White House really had been intending to torpedo the relationship with China, it would have come right out and said this shows excellent.

Y, Z, and should have the following consequences for the relationship. But the fact that they were pretty much mum about the relationship itself suggests to me that Trump is focused more on domestic politics in this regard rather than China per se. I don't know, Alice, what do you think? It isn't that clear, I have to admit. Yeah, but I think you're right. I think that Trump is ultimately blaming the deep state, what he calls deep state, the intelligence community for failing to inform him, it seems. that the Chinese seem to have hacked 200 million U.S. voter records. And it's, I think, part of a calculated move. Maybe I'm reading too much into it ahead of the midterm saying, you know,

hey, there are these bad actors out there that are also tampering with information going to voters, largely through social media, I would say. And this is what the White House is also saying. And it's designed to, again, show that, you know, he feels that domestically, things aren't great. And I want to add to this domestic, I think, uncertainty, the fact that the latest polling data, I know we shouldn't read too much into it from Pew, is pretty astonishing. And it shows just the both domestic and global decline in faith and regard for Donald Trump, but also for the US compared to China. So now just the latest Pew polling shows that

46% of people worldwide have a more favorable view of China versus the U.S. But even when you break down into countries and your country is one of them, James, the UK, just the countries that have more confidence in Xi than Trump or have more confidence in China versus the U.S., there's a much larger pool of 36-polled countries than say that they have more favorable views of China. I remember, you know, back in 23 and hot off the heels of the COVID outbreak and China being blamed for it, that people were saying, you know, China's never going to come back from this. everyone's hating on China. The world is anti-China. But fast forward three, four years. What a world's, you know, what a difference those years have made? And it seems like through a mix of the U.S.'s own goals, but also, you know, China's pretty good soft power campaign,

we're at a stage now where China has more favorable views versus the U.S. Yeah, absolutely. I mean, I found this really quite shocking. And to your point, Alice, look at the difference between now and 2023. In 2023, the Pew data showed that 58% of people in countries, this is across 20 countries, polled, had a more favorable opinion of the US versus 32% having a more favorable opinion of China. And then going down to the list of nations now that have a more favorable view. of China, Germany, the UK, France. But let's just pause for a minute to think about the relative relationship between those several European countries and the US relative to China. I mean,

it's well known that the US is the UK's longest standing ally and friend, the amount of trade, people-to-people relations, investment relations, financial relations, political relations, relations, not to mention the fact that the UK depends on NATO, which is led by the US, entirely for its national defence. And yet, people in the UK still, according to this poll, have a more favourable opinion of China than they do towards the US. And you could say the same thing about Germany. But with Germany, there's an added ingredient here, which is that Chinese competition is wiping out a great big chunk of German industry right now. And yet Germany on average or German people on average have a more favorable opinion of China than they do the US. And I could go on.

So the real question to me is, you know, what is the key variable here? And I have to be blunt. I think it's Trump. I think that China is presenting a more positive picture to the world. But I think living here in Europe and traveling around Europe, I think people across this continent are deeply unsettled by many of the statements coming out of the White House, and they just don't know in what direction Trump is taking the United States with regard to its relationship with Europe. So I think the big swing factor is Trump, and then China's relative stability is playing to China's advantage, probably. But it's pretty shocking. You look to the U.S.'s closest neighbors, Canada and Mexico. So Mexico, there's a 19 percentage point swing in favor of China, where 59% people have a more

favorable opinion of China versus the U.S. In Canada, there's an 11 percentage point swing in favor of China. I mean, it seemed like only yesterday where we had scandals involving the two Michaels and Canada, you know, putting on sanctions on China, et cetera, et cetera. It's astounding to me that the two closest neighbors and most of Latin America have a more favorable view of China versus the U.S. And it speaks to potentially a failure in the Western Hemispheric approach that was codified, remember, the November defense report by Washington. How much of this is, yeah, I mean, at the heart of it is how much of this is Trump and the administration's undoing versus the Chinese playing a pretty cool, hand. Yeah, I mean, the other thing I think to say about this is that public opinion is fickle. It does really depend on the headlines of the day. But I think what you're referring to there,

Alice, is a much longer-term structural issue here. You know, as we've discussed many times on China decode, China is becoming the world's leading technological power. In this episode, we're talking a lot about China and AI and the success that China's experienced in that regard. And throughout much of the developing world, China has been bankrolling, much of their infrastructural investment and construction. And so China is beginning to get a reputation as the country that can get things done. I think that's really quite important. So even though I do recognize the fickle nature of public opinion and polls like this. I also think that there are bigger structural changes underway.

Okay, we'll be back with more. After a quick break, stay with us. Support for the show comes from Odu. If your business runs on five different apps, 12 browser tabs, and one spreadsheet that everyone's afraid to touch, it's probably time for Odu. Odo is an all-in-one business management software that connects every part of your business into one powerful, easy-to-use platform. So, instead of wasting time switching between disconnected systems, your entire business works together in real time.

Your team moves faster, your data stays accurate, and you can actually focus on growing your business. Let one unified system run your entire business. From the first opportunity to the final payment, everything works together in one place. Whether you're a small business or managing a large operation, O-Doo gives you one. flexible platform built to grow with you. Try today for free at odu.com slash prop G. That's ODOO.O.com slash prop G. Support for the show comes from LinkedIn. If you're a small business owner, you don't need me to tell you how much hiring great people matters. But the time and resources you have to spend to get it right are precious commodities.

Sourcing, connecting with, and screening candidates can quickly eat into time, better spend on your customers. That's where LinkedIn hiring pro comes in. It's designed to be your hiring partner, helping you source the right candidates faster. That way, you can hire with confidence without making it feel like a full-time job. LinkedIn Hiring Pro simplifies the entire process, all the way from writing your job post to shortlisting candidates and running AI-powered initial interviews. Plus, it does it all through a conversational interface where you can just describe what you're looking for in plain language. LinkedIn says nearly 60% of hires find someone to interview within a week. With Hiring Pro, you spend less time searching and more time connecting with the right

talent, so instead of sifting through piles of resumes, you get a tight, high-quality shortlist that actually moves things forward. Join the 2.7 million small businesses using LinkedIn to hire. Get started by posting your job for free at LinkedIn.com slash pro. Terms and conditions apply. Support for the show comes from Pendulum Metabolic Daily. There's a gut bacterium that PhD researchers from Harvard, Berkeley, and Stanford have spent years studying. They were specifically trying to figure out what it does for the metabolism. It's called Akramansia. And it's the key ingredient in Pendulum Metabolic Daily. Akramansia is the gut bacterium that helps signal your body to produce more of its natural

GLP1, and you already may be familiar with GLP1, which is the hormone that helps regulate appetite and feeling full. Pendulum Metabolic Daily is backed by rigorous clinical research. It's a probiotic design to help you feel full or longer, experience fewer cravings, and make it easier to stay on track with your health goals. It's just one capsule a day taken with food. Some people report noticing fewer cravings and more energy in as little as a week, though individual results can vary. Right now, our listeners can get an exclusive 20% off their first order

at pendulumlife.com slash provchee. That's pendulumlife.com slash provchee for an extra 20% off your first order. Again, go to pendulumlife.com slash profi and get 20% off today. Welcome back. We're joined now by an expert in all things AI, Selena Xu, a writer and head of China research for former Google CEO, Eric Schmidt. Selena, we're so excited to have you on the the show. Thank you for having me. I'm so excited to be here. You know, you and I have been

talking a little bit, and I know you've been in and out of China, so I really want to get your fresh take on what's happening. Obviously, we've seen the recent announcement from Kimmy in terms of their capabilities catching up to some parts of the anthropic suite. It seems like every other week we hear this, James and I talk about this quite a lot, these new announcements, these Deep Seek 2.0s that are coming out of China. How should we be reading these announcements? And And what is the broad takeaway from them when it comes to U.S. versus Chinese capabilities? Yeah, that's a great question. I actually visited Munchat and also Z.A.I. in Beijing, end of May. So I think it's really interesting just how fast things are moving. If you look at the pace of releases, right, like think about it.

Deep Seek v4 came out in April, and then we had the whole fable mythos, you know, saga. and then Fable came out after the bends were lifted. And then in June, we had essentially a GLM 5.2 from JCPOZ.A.I. And then now July, Kimi. So the frontier of intelligence is moving so fast that any lead you have is essentially appersible. It's just you might hold on to it for a max a month. And then it's the next model.

So I think this is kind of the pace here. And if you're thinking about the vibes on the ground and how people are talking about it. I felt the biggest takeaway in terms of vibes is that compared to last year, Chinese AI companies now are so much more motivated and they have a very clear goal in mind. I think last year it was a bit of soul searching after DeepSeek R1, right? People are like, can we replicate that? What is the next paradigm? Do we have to figure out something after the transformer? And now it's like, okay, anthropic has shown us the way. Coding agents is the way to go. We are going to all do this and we'll do it better. So I think that's kind of the very gung-ho energy that I sensed in China

and amongst talking to these AI founders. And I think once they have this clarity, there are a lot of advantages that China has on their end. So I think one is that they're all planning to stick to open source for as long as it makes sense. I know there was a bit of a debate, you know, the past few weeks about whether China was going to restrict open source. And I think we've seen with the latest releases

and with what Chinese president's Xi Jinping said at World AI Conference, that they will continue to. to do open source. So I think one, that's open source will continue from Chinese AI companies. And then the second one is, you know, there's been a lot of efficiency techniques that these startups have done, which essentially means that they will try to extract the maximum amount, you know, of training efficiency from every unit of compute. So because of the chip constraints that they have, they've been forced to do this. And as a result, their models are so much more capital efficient and compute efficient compared to, you know, open AI and anthropic.

So I think they will continue to do this. And then the turning point, I think, with Kimi K3 is that it is a huge model. It's the largest open source model. It has 2.8 trillion parameters. So this is huge. And they're not really trying to, you know, compare themselves to deep stick V4, which is super cheap. Instead, I think they're trying to show that, hey, open source doesn't just have to be cheap and efficient.

We can also be at the very cutting edge. So they're trying to like toss that away. And I think their president, Zhang Yutong, had this post on social media and she's saying like open source doesn't have to just go the cheap route. We're trying to do something else. So I think this is kind of a turning point for Chinese open source. Terrific, Selena. It's so great to have you on. Maybe I could drill a little bit deeper into the structural advantages that China has because we hear about an inexhaustible supply of intellectual talent. We hear about very low electricity costs. some of the things you've already mentioned with regard to the efficiency of Chinese models. If you throw this race forward, the race for AI between the U.S. and China, do you think that these structural advantages will begin to tell in China's favor? Yeah, I think that's kind of the big question when people are thinking about the U.S.-China race. Like, what are the chips that each of them have on their side?

And I think for sure, China has huge structural advantages that you've mentioned. So one, energy. So energy is one of the main constraints on the U.S. side. We've seen all the protests against data centers getting built and, you know, proposals of moratoriums on that. So I think the timeline of trying to build a data center and get it erected and have the right chips inside for it to come online is just way longer in the U.S. compared to China. And China, the main thing is what they call Dong Shui. Si-Suan, which is essentially like east data, west compute. So you have, you know, the major cities on the coast,

essentially using a lot of AI. And then you have the compute built out in, you know, far-fetched areas in the desert, you know, in our Mongolia or in Guizhou or other more far-flung places where actually a lot of renewable built-out has taken place. So that's kind of the strategy that China has been going with. I would say that, you know, beyond energy, I think one of the big things, actually, that people have taken less into account in this race is what you would call, like, domestic sentiment and societal transition. So I think this is one thing that nobody really anticipated, because if you look at past technological cycles, you know, initially there is a bit of resistance, for instance, during the Industrial Revolution and you had the Luddite protests of people

smashing, you know, the automated looms. But afterwards, when the benefits start showing, you have people adopted, and it generally lifts, you know, everybody's boats and society adopts it and innovation is good. I think this time round with AI, it is a lot trickier because the whole pace and scope of automation will impact much larger swaths of people who will have a lot more political power compared to previous cycles of technological transition. And we see this playing out very acutely right now in the US when you see a huge anti-AI sentiment across the country. There was a poll in March when people disliked AI on net more than, you know, ICE and more than, you know, Iran.

So I think this is something that's very hated, especially by younger people who tend to usually be more pro-tech. And then in China, there is that similar worry because so much rests on, you know, social and regime stability, but China really needs AI to drive its engine forward. So I think the common challenge in this race also beyond, you know, energy, talent, compute, and everything else, it's really which country can handle this. transition better and, you know, balance both having a lot of AI growth for their economy and then being able to control it so that, you know, the whole system doesn't really collapse. Really interesting. On the subject of transitions, I have to alert everyone to your amazing

op-ed with Eric Schmidt and the New York Times in which you talk about your takeaways from your month basically in China over the summer. And what struck me is interesting in that piece is the fact that both the government in China and the entrepreneurs have this kind of slow-burn approach to AI because they are mindful historically, as you've alluded to, to instances in the past where there has been, say, a Luddite backlash to technological innovation. So in a way, this explains why the government has been, I think, pretty alerted to the fact that there is a labor issue to too much AI transformation, that the gig economy of 320,000, 44% of the workforce is at risk. You see that in terms of them slow rolling autonomous vehicle licenses in China.

And also more recently in a piece you may have seen in terms of AI companions trying to set regulations in terms of making sure that they're not too personalized or anthropomorphized so that people don't have too much of an emotional or romantic connection to these AI bots or agents. Can you speak to us more about this relationship that governments and entrepreneurs in China have to AI and how that's quite distinct from what we see in the US? because it seems to me that at the heart of it, these two countries see AI in very, very different ways. Yeah, it's definitely a different relationship between the companies and entrepreneurs with the government, as we all know in China versus the U.S. I think in China, the interesting dynamic is,

and you've talked about this before, Alice, like the entrepreneurs and the tech industry were basically reined in during COVID under a sweeping crackdown. And then now, after, after the after the after COVID ended the government realized actually we really need the private industry and we need tech and AI specifically as this new engine of growth so there is this you know almost reluctant but pragmatic like concession like we need you and they and this is mutual but I so I think having that way on the minds of the entrepreneurs it really really I think shapes how they see their

relationship with the government it's not just like let us run complete deregulation, we got it. It's more of like, okay, we want to negotiate some amount of room for growth and we will have this constant feedback with the government. But at the same time, we know if there is any hint of instability or things getting out of control in a place like China, this will be quite, I think it would be a direct trigger for the government to step in and to basically re-regulate and crack down on the industry again, something that I think nobody in the tech industry wants in China. So I think that's very alive in their heads. And the interesting thing about AI companions is that this is like a global issue. We see a lot of debates about this in the US as well

with, you know, very high-profile cases of teenagers committing suicide. In China, I think the worry is that, but also romantic relationships being formed between especially young women and AI companions and the refusal to then get married have babies and that impacts China's fertility rate. So I think it is a very a complicated issue at the heart of, you know, China's whole debate on what do we do with AI. And I think the best thing to take away from this example of how the guidelines and then subsequent regulation got formed was they released draft guidelines on restricting AI companions in the end of December last year. And then they opened it to feedback from society and industry. And actually, the current regulation that we've seen take an effect as of last week

is very much watered down and a lot more narrow than the December guidelines, which had pretty strict restrictions on, for example, you have mandatory human takeover in case of any triggering words. The platform has to step in and bring in a human. That's like a pretty, you know, high burden for a tech company to have like 24-7 staff being, you know, constantly monitoring. So that probably got industry pushback. And then now it's a, it's not mandatory,

but you have to give like an emergency contact or like a guardian, and it's only mandatory for specific groups of people. And then the other thing that got watered down mainly was training data. So the initial draft guideline said that all training data from these kinds of anthropomorphize interactions needed user consent before a company can train on it. But in the new current regulation that's taken effect, this essentially only pertains to, you know,

specific personally sensitive information and other things are good. for a, like, it's good for a company to train on. So I think that's an example of how the government and the industry are basically negotiating with each other on what kinds of regulations will make sense to protect some of the government's priorities while allowing industry to continue to drive AI development forward. Support for the show comes from MongoDB. AI-assisted and agentic coding can help you build faster than ever. But if your data layer is a bottleneck, what's the point? MongoDB actually gets out of your way.

MongoDB is a unified AI-ready data platform that empowers you to build scalable generative AI applications. It eliminates the need for separate, specialized vector databases by combining a flexible document model natively with semantic vector search, full-text search, and real-time operational data. Instead of wrestling with rigid schemas or translating data formats, MongoDB's native data model mirrors the language LLMs already speak. Plus, MongoD's D.B gives you the flexibility to ship at the speed of AI. The ACID compliance lets you sleep soundly at night and scales to handle massive Fortune 500 workloads. Developers have a word for that kind of reliability. Actually, four words. It's a great database. Start building at MongoDB.com

slash AI. Support for the show comes from SOFi. Whether you're buying your first home or just looking for ways to make your current mortgage work better for you, the process can feel overwhelming. news, you have more options than you might think, and the right home loan can really make a difference helping you reach your financial goals. That's where SOFI can help. Whether you're buying, refinancing, or planning your next move, SOFI offers a tech-forward human-backed mortgage experience designed to make the process simpler. You can apply online in minutes while still having access to a dedicated mortgage loan officer to answer your questions and guide you through every step of the way. So-no matter where you are in your homeownership journey, SO-Fi has solutions designed

to help you move forward with confidence. Head to SOFI.com slash PropG loan to explore your options today. That's SOFI.com slash PropG loan. Mortgages originated by SOFI Bank, NA member FDIC, NMLOS, 696891. Terms and conditions apply, equal housing lender, verified pre-approval letter, terms apply, see conditional pre-approval letter for details, not a final loan approval and not a commitment to lend. Support for the show comes from CHIM.

For a lot of you, picking a bank was probably something you did at young age and never look back. But banking has evolved since then. And like most things in life, it's gotten better over time. Enter chime. Chime is not like a traditional bank. Chime offers the most rewarding, fee-free banking with the power of the word no. No overdraft fees, no monthly fees, and no minimum balance fees. You can get 5% cash back with chime cards on things like gas or groceries, all while building credit through regular everyday spending. You can also grow your money faster with a savings rate that is nine times the national average. And if you're ever in a pinch,

SpotMe lets you overdraft up to $200 fee-free. Join the millions who are already banking fee-free with America's number one choice for banking. Head to chime.com slash profjee. That's chime.com slash prop-G. Sign up now. It only takes a few minutes. Chime is a fintech, not a bank. Banking services and chime card provided by Chime's bank partners.

Qualifying direct deposits required terms and limits apply. Go to chime.com slash disclosures for deal. details. And to follow up on this, Selena, you know, we've seen some interesting data from Stanford more recently about Chinese versus American users' attitudes towards AI. 84%. This is a shocking number. 84% of Chinese surveyed respondents said that they were more excited than worried about AI versus the US 10% said that they were more excited rather than worried, meaning that a higher number potentially could be worried in the U.S. about AI, but it speaks to how there is quite a dichotomy but in sentiment between Americans and Chinese. Can you explain that? And to what extent is that

related to what you've just alluded to and talked about the differing approaches on the one hand, this kind of fast-paced, unregulated, I would say almost hair-like in the tortoise and hair analogy approach to AI versus this kind of slower, slightly more regulated tortoise approach to to AI in China? I think this statistic took a lot of people by surprise. There were a few other surveys by like Edelman and others that people started paying attention to, but I think as negativity in the U.S. towards AI grew to quite a feverish pitch this year, people are suddenly like, oh my God, why is China so much more positive? And obviously, you know, the surveys are not fully representative of sentiment on the ground, which is often not monolithic. But being in China, a few,

few times this past few months. I think the main thing, the main attitude difference that I sense between, you know, Chinese and Americans is that in China, people are very pragmatic. They don't see AI as this, you know, in general, this machine god. Talk of AGI or artificial general intelligence is pretty much absent from most convoes, other than if you're talking to a very specific minority of westernized, you know, policy elites. And for the average Chinese person, AI is, either a tool where maybe they have to learn because they are mandated to buy their bosses and their jobs. It is like an entertainment thing similar to, you know, Dohing and Red Note before. Now they're like, hey, I get to play these like virtual script games with my AI boyfriend who is very hot and looks like an anime character. And this is basically a continuation of all these like love and deep space automate games where you have romantic storylines, right? It's just like a natural continuation of that. And then for some people, it's like if you're elderly, you're like, oh, great, AI allows me to, like, Photoshop or, like, ask for medical advice. So there's, like, very pragmatic tool, like, ways and people don't imbue with it.

All these very doomsday, what people now call doom trolling in the U.S., of, like, AI will make us extinct. AI machine gods going to come. It would be, like, the Terminator. It's terrible. There's none of that, even with all the humanoid craze in China, which actually is another structural advantage on hardware side for the U.S.-Chinaina. race for China. The humanoids in China, a lot of them are, you know, build now as AI companions. So UBTAC released a bunch of these anime, gorgeous-looking humanoids that are supposed to be like, you're a friend or your helper. And then there are these very cute, smaller humanoid robots.

And also AI-enabled toys, they're fluffy and soft toys that are all about, you know, how do you integrate AI into your everyday entertainment? So I think the way people interact with AI in China just feels very much more grounded, a lot more for entertainment and pleasure, and then also much more pragmatic compared to the US. And I think that goes back to, you know, the broader question that you brought up about the differences in broader approaches. I think China by necessity because of chip export controls and then also because of just the structural advantages it has in hardware has allowed AI to be diffused in a way that feels a lot more palpable, a lot more. spread out across industries, and people are not, you know, are seeing many different visions of AI emerging compared to in the US, where it's very much dominated by one Silicon Valley narrative that's all been about AGI and now has only recently realized, oh, people don't like this, so we're going to change our narratives. So, Selena, I was recently in China, well, a few months ago, and you mentioned humanoid there. I became absolutely fascinating.

by the humanoids. What's the situation these days? Are we seeing diffusion of humanoid robots into the workplace much in China? I mean, some of the people I said to say, well, they could be in like nursing homes, helping to care for old people or helping to care for patients in hospital or or helping with other kinds of tasks, you know, everyday tasks. Are we seeing that? Are we seeing that? or is it just a little bit too early for that? And then, I suppose, an allied question is people talk about the chat GPT moment for humanoid robots, which is supposedly the moment where a humanoid kind of becomes almost as good as a person.

In other words, you can just ask it to do something and the humanoid will work out how to do it. Did you pick up much of that? Is China getting close to this potential moment, do you think? So I think China and the U.S. are both trying to figure out what are the use cases for human rights. So in China, the closest thing I've actually seen to actual deployment instead of just, oh, display for hype, is having them do some very routine task in a controlled setting like a factory. So mostly some EV companies have been doing this on their factory floors. And most of the EV companies in China have highly automated factories that already have very few humans who are on the grounds. So plucking a humanoid inside to do one very tiny segment of the overall automation.

For instance, like checking, doing checking this thing and moving it across like 10 meters away into a box. So those kinds of very simple mundane tasks. So that has been, I think, the most deployment I've seen in a factory setting. And this, I think, is a bit more for data collection. It's not really for the factory to be better, because most of the factories have really great industrial robots, like arms or like AMRs that are transporting materials. So it's really about data collection,

which goes to your second question about, when is the CHETGBT moment coming for robots? And I think the biggest kind of bottleneck here is in data. So you can have a humanoid be in a very controlled setting, doing a regular task. But the moment you set them into the real world right now, say the angle of the cup is different, or the height changes slightly, or something is different in everybody's household, the humanoid doesn't know what to do. And they are, the hardware currently is there.

And when you see these gorgeous, like, impressive unitary robot performances of them doing, like, kung fu or, like, parkour, those are mostly local motion, and it shows that, and you can program them. And it just shows how good the hardware is in charge. But the main problem of having a robot brain being able to autonomously figure out what to do in a very wild, irregular setting that's unpredictable, currently doesn't exist. And that brain question is kind of the big question that you need for a chat GPT moment to happen. And there are many different approaches here that people have argued for. So some people are saying, hey, you can have this as an extension of multimodality. So if you continue on the LLM route of large language models and you just make multimodal LLMs, they're so much better, they can allow us to get to that autonomous robot brain.

Some people are like, no, we need a totally new paradigm that's not LLMs. We need world models. So there's what people like Faye-Fei Lee or Yan Lacoon are talking about. They think you need basically the robot to understand how, you know, the physical rules of the world before it can start to navigate that. And then there are obviously some other approaches that are emerging. But there's no really a clear consensus. And I think so much of it really depends on people now trying to do massive data collection so that the robots can get smarter by being trained as as many contexts as possible.

And in China, this has manifested in these data factories for robots. And I think Sao Mi's founder, Lejeun, recently posted one such example of a humanoid doing something like this in his EV factory, where you have humanoids being tested out in different settings. And then there are also people who are going down the simulation. route, which is to simulate a lot of data, but the quality of that is, you know, up for debate. That's absolutely fascinating. I mean, it's just so tantalizing to think of a world in which humanoids are roaming around doing work and apparently sort of being able to respond to real-time cues and instructions. But I guess from what you're saying, it sounds like that's a few

years away. Would you say that? Yeah, I would say that. There are people in the industry like Wang Sing, who is the founder of Unitri, who basically thinks general purpose human rights are three to five years away. But obviously he's a believer. Some people think the timeline is a lot longer. All right, Selena, this might be an unfair question, but I see this a lot in my feed as to whether or not China is just cheating the exam by distilling U.S. models. You know, it's in the vested interests of Dari Amadei and Open AI to be saying this. But based on your best understanding, because you know, you're very much plugged into what's... happening in the U.S. and China on the LLM front. How true is this and how true will this be when we

have a conversation about this in a years or two years' time? Yeah, I mean, distillation definitely happens. And I've seen, you know, what Anthropic has published. Some of those are like very small number of instances. Some are huge, like a more egregious like mini-max. But I think my personal take is distillation is a very widespread industry practice. Someone like Elon has said that as well, XAI distills from Open AI. Like, everyone does it. And it's generally a widespread academic practice for people who don't have the right resources to be able to still be able to catch up and contribute breakthroughs, right? So having that demonized right now in the U.S.-China geopolitical framework, I think more of it is about, you know, the commodity of intelligence. Like

open source models from China very obviously undermines U.S. labs who have close models and they're charging a lot of a lot for those. So I think there is a clear conflict of interest. So I do see why they are against that. And it is it is anthropics on, you know, prerogative to say, hey, we want to clamp down on these practices to protect our profit margins. Right. I think that's very valid. But on, you know, the broader argument that seems more ethical here on like, oh my God, our Chinese labs cheating, they have been copying us. I think it's like, it kind of flies against the whole state of the landscape, which is that these Chinese companies are pushing out amazing open source models and openly publicly publishing their techniques.

So if you think about the recent, so Thinking Machine, which Miramurati started, recently released their own open source model. And they publicly said we used techniques and used and trained it on models like Kimi K2.5. So when a Chinese lab open sources, their models, it actually gets adapted by U.S. companies who can use it as a base layer. and it actually honestly boosts American innovation in some way. So I'm very pro-open source, and I think there is this extensive co-opting of each other's knowledge that's happening. That's not one way, especially now with, you know, Chinese AI companies dominating open source.

I think it's very simplistic and almost too reductive to be like, hey, it's Chinese companies copying American innovation. Right now it's obviously two-way. And some kind of an argue that, you know, most of the American startups nowadays in AI are actually building on top of Chinese open source models. So there is, I think there's a total reversal of what we used to see in the internet era where we're like, oh, Chinese tech industry just copying Silicon Valley. I think now it's a lot more of mutual co-opting. Is it safe to say that open source has really revived? I mean, two years ago, it was all about

closed source. It was about the frontier labs excelling and accelerating. And now, Nvidia is doing it. Merrimurati you've said is doing it, that more players in the U.S. are going into open source. Is it fair to say that when we talk in a year's time, we'll have a stronger open source ecosystem than even today? Yeah, I hope so. And I think a lot of American AI companies are trying to build, you know, the American open source model like Nemotron from Nvidia, thinking machine. There's a company called it Reflection AI.

So a lot of people are trying to do this. And I think it's better for the ecosystem if you have, more stronger American open source alternatives to Chinese ones, right? Like when people complain about Chinese open source models, I'm like, just build your own then if like if you're so unhappy with, you know, the ones coming out of from China. So I think my prediction is that in a year's time, we'll see a lot more very competitive American open source models, but whether they can beat China and, you know, not just at home, but also in the global market really just depends on a lot more things like cost, like efficiency,

and also, you know, whether they're able to bring in the users who are now very entrenched in a very different ecosystem. Amazing. Thank you so much, Selina. I've learned so much from this conversation. Really appreciate it. Thank you for having me. All right, James. You know what time it is. It's prediction time. As you look into the future this week, what do you see? I'm going to talk about an economic issue for my prediction. we saw in the first half of the year, retail sales in China climb a meager 0.3%. And of course, consumer spending and retail sales in China are an absolutely crucial determinant of broader economic growth. And so my prediction is that we will see in the second half of the year a slight improvement in overall retail sales. But I'm a going to say that for the full year, China's retail sales will grow at less than 3%.

And I think that this is rather significant because it means that China will not be able to resuscitate consumer demand very much. They will do a little bit, but not very much over the course of the year. So we'll see overall on the GDP register, sub-fusk growth continuing for. the rest of the year. Very interesting. So mine is in the economics meets technology space, and it's really about chips. I don't know if anyone's seen the figures, but in H1, the first half of this year, basically semiconductor exports from China to the rest of the world by value doubled relative

to last year. I think when we look at the second half of the year and the year as a whole, we're going to see record exports and a growing amount in the next two quarters, I think, of Chinese chips, especially in the DRAM and memory spaces where China is injecting more supply at cheaper costs. Look at CXMT for instance. It's competing with Samsung and SK-Hydex. I think that's going to be a huge story

when we think about China's export machine for the whole of the years is the big chip and AI hardware export story. That's so fascinating. I'm really, really looking forward to seeing how that one pans out, Alice. This is such a big shift for China to be exporting so many chips in, as you say, record numbers.

Yeah, who would have thought? All right, that's all for this episode. Thank you for listening to China Decode. This is a production of Prof G Media. Make sure to follow us wherever you get your podcast so you don't miss an episode and talk to you again next week. If you want a $3,000 a month payday for life,

what would you feel free to do? Maybe take a long week. Weekend, every weekend? Or try a bunch of new hobbies? Would you feel free to upgrade and listen ad-free? Don't worry, we get it. Every $20 ticket could win you $3,000 a month for life and supports life-saving cancer research at the Princess Margaret.

Feel free to buy your Payday for Life ticket today. Raffle number 155-2194. Please play responsibly.