Latest / Barbarians at the Gate / AI Slop, AI Detectors, and Authentic Voice: A China-Tech Insider on Writing and Working With AI
Transcript
- How non-native English speakers use AI to compete with native writers Barbarians at the Gate Jeremiah Jenne: Hello, and welcome to another edition of Barbarians at the Gate. This is Jeremiah Jenne, broadcasting from the Sweet Retreat Farm in central Vermont, where I'm based for the next couple of weeks. And with me, as always — not based in central Vermont, but in central Beijing — is my co-host and comrade-in-arms, David Moser. David, how are you doing? David Moser: Very good. I'm getting better and better. My ankle is healing. I actually went out and had dinner outside the home, in a restaurant, the other night. It was shuizhu yu (boiled fish). So I'm a happy guy. The reason I'm in China is just the food. It's the food. And if I can't have the food, I might as well just die. I'll go back home anyway. Jeremiah: Okay, well, that is an endorsement for shuizhu yu like you've probably never heard before. Joining us again — speaking of somebody who loves Chinese food — is Yajun Zhang of Zhang Global Advisory. Yajun joined us earlier this year talking about technology in China and her trips to Shenzhen. She's back, and we're going to talk about all things AI in China. Yajun has been writing several articles on Substack and is also quite active on LinkedIn, sharing her views on AI, the development of AI, and policy changes underway in both the United States and China, and how they affect companies. Welcome to the show, Yajun. Yajun Zhang: Thank you very much, Jeremiah and David. It's such a great pleasure to be back. Jeremiah: So, Yajun, you recently published an article on Substack contrasting feelings about AI in China and the United States. I was wondering if you could talk to our listeners about some of the observations you made, both in America and in China, and how these two countries — currently engaged in a race for AI — think about it differently. Yajun: Yeah. The whole thing started from a recent interaction with my nephew in the U.S., who is sixteen. As you mentioned, I run an advisory firm focused on helping Chinese tech companies — mostly AI and robotics — go global. So I'm very keen to understand how people see technology. Talking to my nephew, who is super tech-savvy and can build a computer from scratch, I expected something like, "Technology is great, AI is really cool." But on the contrary, he wasn't a big fan. He saw it as a big capitalism scheme — using technology to widen the gap between rich and poor. That really shocked me. And more importantly, it's so different from what I've seen on the ground in China. I recently did a two-and-a-half-week tour through Hong Kong, Shenzhen, Beijing, Shanghai, and Tianjin. Everyone I talked to — from my mom, who is in her early 70s, to frontier tech people — everyone is so enthusiastic about what's happening on the ground, about how they can leverage new technology to make their business boom, or at least curious about what's happening. So listening to my nephew — a young person who really wasn't interested in AI — made me feel that, at least at the grassroots level, the attitude toward technology is so different. And that made me feel that, over the long term, because of attitude, people's tolerance and acceptance of technology, and of the consequences it brings, can be so widely different. David: Yeah, Yajun, I know you've read Dan Wang's book. He mentioned something I think you mentioned too, and it would be interesting to talk about. The U.S. is afraid of AI because it's the end of the world — we're going to lose all our jobs, there'll be no more creativity. Anthropic and these companies are trying to build general artificial intelligence; they want the most powerful computers in the world, which will then become our masters, and we'll be the slaves, right? China looks at it totally differently. They're saying, "Yeah, these are great. We'll incorporate them into online services and make life easier, and we'll also control it." If it gets out of hand — and they're already controlling AI like DeepSeek — there's no fear. This is just going to help make our lives better. Are people in China aware of this difference, and what do you think about it? Yajun: I totally agree with you, David, that the two countries take very different approaches to AI. In the U.S., the major tech firms chase AGI — artificial general intelligence. But you don't hear about that much in China. Everything is about how to deploy AI into people's day-to-day life and make life better — how to order a milk tea by leveraging the Alibaba ecosystem. That was a huge thing during Chinese New Year. So people see this technology very differently. I wrote a piece recently about why Chinese people do not fear AI. There are several reasons. One is that China has been benefiting from technology development since the 1980s and 1990s — the early stage of the internet era — and then the mobile era, and now the AI and robotics era. Over the last few decades, the only thing Chinese people have seen is that WeChat, Alipay, and e-commerce make their life easier. And people actually earned a lot of money from some of those technologies. They benefited from it. So naturally, people have a different attitude toward technology. Government support is another factor. From the latest 15th Five-Year Plan to the "AI Plus" policy issued, I think, in 2024, the government constantly talks about how important this technology is, and they really put money behind innovation in the field. So people feel this is a new economic driver for the country. And think about it: without this new driver, without new technology, people only hear about the real-estate bubble and slow consumption. But now you have this new toy that the government and tech companies can point to and say, "We have something new, it's shiny, it can help us be more efficient." So it's hard to see people not embracing it. David: Can I just give one example of integrating it? There's the 12345 hotline in Beijing and across China — a hotline where you can get any of your problems solved. You make a complaint and it instantly solves the problem. I saw their operation, and it's amazing — it's science fiction. There are maybe 800 or a thousand people on the phones every day, answering many thousands of calls. The whole thing is integrated, and it's all AI — the types of complaints, how many people, in what areas, and so forth. But one of the issues is the people answering get calls from, say, elderly people who are lonely and just want to call up and talk to someone. And I thought, yes, this is a great integration, because you have to have a human being take the call — you get all kinds of wacky things. But the data management and analysis is all AI. I thought it was a perfect example. Yajun: Yep. Actually, I had the pleasure of talking to the person who built and managed that whole system in a tier-one city. This cadre was a good friend, and he showed me the data room. It's huge screens with real-time data — how many people are calling in. They track each case, the whole process: which department should get the report or the complaint, who is following up, what percentage of cases has been solved or is pending, which district gets the most complaints, and what the major issues are. They have all of that data. And I asked him about AI deployment in the process. He said they have about a thousand staff to take calls every day. So I asked, why not just use AI to answer the phone calls, exactly as you described? He said they still use humans to take the calls, for the reason you mentioned, David. A lot of people call because they want to talk more, to explain the problem more. You need somebody who is very patient. Second, a lot of them have an accent. Even though China's technology for identifying accents is really good compared with other countries, AI can still get it wrong. And people feel AI doesn't have the kind of sympathy a human has. So he said all the data sorting and categorizing is done by AI, but the phone calls still require a lot of human empathy, emotion, and sensitivity — that's done by human beings, and they don't plan to change that. Jeremiah: Moving from AI as a big concept — solving citywide problems and possibly becoming our new overlords, in David's terms — there are a lot of applications of AI on a more micro level. I just saw an advertisement during the FIFA World Cup for Genspark AI, a kind of digital assistant, and I noticed that their mascots are a team of small crayfish. Obviously that joke may be lost on people watching the ad in English, but people familiar with "open claw" in China would probably get the humor of having crayfish solve all your problems. One challenge AI helps solve for a lot of people is writing. There's been a lot of angst online over AI slop — generated posts and essays, even books being pulled out of production over allegations of AI. But one area that hasn't been looked at as much is the use of AI in the creative process, in writing, and the way it can bring new voices in. Yajun, you recently wrote — in fact, just over the weekend — an article about being a non-native writer of English and using AI to help you write. I was wondering if you could talk about your creative process, why AI can help non-native writers of English, and what some of the traps are that native speakers aren't always aware exist for the non-native writer using AI. Yajun: Thank you for that question. I have a love-hate relationship with AI. I use the technology a lot. For me, as somebody bridging China and the rest of the world, the crucial thing is to get my ideas out — particularly my observations on the ground, my insights, and the details to explain what I've seen and thought about China that I feel the rest of the world should understand. So the ideas and insights are crucial. However, as you said, I'm not a native speaker. So when I write, I want to make sure of a few important things. First, the idea should be clear. Second, there shouldn't be any typos or grammar mistakes that make people stop reading. So I definitely use AI to correct my writing, to make it smooth and make sure there are no grammar mistakes or typos. I also use AI to help source data. For example, if I want to know the population of mobile users in China in a certain year, I have AI help me source that — and of course, afterward I verify whether the information is accurate. It does significantly increase efficiency and reduce the time it takes to write articles. But at the same time, I feel some dilemma when I use it. I write in a certain way, and even though the majority of my articles are written by me — I only ask AI to correct typos and so on — a lot of the time, when I put it into an AI detector, it still says what I wrote was written by AI, which is not true. That really confused me. It took me a while to understand why. Basically, as a non-native English speaker, I was trained in a certain way. In elementary school, we had grammar classes that teach you, when you structure something or explain an argument, to write in a template: first, second, third. And you should always wrap up an article with "in conclusion, this is like this." That's how we are trained. And apparently the large language model is also trained that way. So we have a mutual understanding — that's why whatever I wrote, when you put it in the detector, it says it's written by AI, which is not true. I think personal experience makes it worse, too, because I work in consultancy. In a consultancy, you need to write very precisely and professionally. Basically we write like a robot, in order to make people feel it's professional. You don't need to be creative or write like a journalist; you just need to be concise and condense a world of information for a C-level reader into one page. After that, I worked for the World Economic Forum, which is famous for big jargon and long sentences. A lot of the time, from the outside, people consider it very vague and abstract. So my whole English training doesn't help me write more like a native speaker. It's professional, it's templated — writing to a certain template. So it's not as natural as a native English speaker. But at the same time, I keep trying to think about what is considered an authentic voice. If the idea comes from me and I have AI help me correct typos and grammar, can I consider it authentic writing, an authentic voice? I'm still trying to understand it. Like you guys said, everyone is trying to understand what counts as AI writing and what counts as authentic. So I'd love to hear from you two as well. David: Yeah, this is a really good question. A good friend of mine, Douglas Hofstadter, wrote a book called Gödel, Escher, Bach. He was one of the first authors to write about AI and its possibilities. He said something I'd like to get your thoughts on. He knows people are using AI — especially non-native speakers, for sure. So when he's looking at a report, he realizes that whoever wrote it put it through an AI editing process. There was a specific event that happened: he got an email from a Chinese woman asking him to give a talk. The email was perfect — totally native. He found this disturbing and strange, so he asked her: "I assume you're not a native speaker, but your email was perfect. Is this how you would have written it if you hadn't gone through AI? Because I feel like I don't know you. I don't get a sense of who you are. It's fake. I would rather see your errors and the way you really talk. I don't care about the mistakes — I just want to get a sense of you as a human." This went back and forth, and she agreed with him and said, "I'm sorry, I won't do it anymore." So what do you think? There must be a place where the things you write have to be perfect, structured, and printable. But do you think you'd draw the line and say, when I'm writing an email, I should do it in my own voice — correcting typos and so on — but I want the person on the other end to have my voice in the email? What do you think of that? Yajun: If I were that lady, I wouldn't apologize for it. To be honest, as a non-native speaker, I've gotten a lot of crap — seriously — for my accent, for my writing. And I'm not alone. I have a lot of colleagues, whether they're Indian or from Southeast Asia, and we talk about this. We're not native speakers, and we try to write as professionally and correctly as possible. But still, we're not native, right? I'm so glad your friend is so tolerant and considerate, accepting an imperfect email. But most people are not. A lot of people wouldn't even consider talking to certain people if they make a typo. I have a lot of clients like that, trust me. And it's the same with speaking. I know people who wouldn't consider non-native speakers for speaking gigs or roles even though they're experts in the field, just because their English isn't perfect. There are a lot of people like that, unfortunately. So it's really hard to say what the preferred way is. At the end of the day, we just have to improve our writing. But I want to say to the people who are really picky about us: you try writing in your second language and publishing articles two or three times every week. That's why, for me, authentic voice is about your ideas. If you can get the idea right, and if this tool makes people more inclusive and helps them get their ideas out — so the global view of the world is more inclusive and comprehensive — I would vote for that. Jeremiah: I also wonder — and I'm not speaking about your friend specifically, David — whether there's something a little insidious here: the idea that authentic Asian voices must somehow be flawed in their English delivery. I think about the movie The Last Emperor, which was dubbed into English. The conceit was that people in the 1980s wouldn't sit through a whole movie in Chinese. But when the characters are speaking English to each other — and we're supposed to pretend they're speaking Chinese — they're all speaking broken English. Like, "Emperor sick today. What happened to Emperor?" And I'm thinking, what's going on here? If they're supposed to be speaking Chinese and we're hearing English, they're probably not speaking broken Chinese to each other. But somewhere in the 1980s, this was the ideal — it won't sound authentic if the Chinese characters are speaking flawless English. I'm not saying that's what's going on here, but Yajun has a point that non-native speakers are caught in a trap: if they don't produce, they get judged on their presentation of English; and at the same time, if they present something flawless, they must have somehow cheated. I wanted to ask you this too, Dave, because you're on the front lines of education, and unlike me, you work in both worlds — with Western students and a lot of Chinese students. In this age of AI, in this age of encouraging students, as Yajun said, to get their ideas out — ideas that might have been locked inside but for this technology — how do your students use it? And how do you, as an educator, draw the line between encouraging students to express themselves this way and not letting it become a crutch or a way to generate or fabricate ideas, if that makes sense? David: Yeah, it makes total sense. I'm old enough that I went through two different eras — when there was no such thing as these devices, and now when they seem so ubiquitous. By the way, it's the end of the semester, and my Chinese students had to write a short paper. Two of them, two women, approached me and said, "Can I get extra credit because I didn't use AI?" And I thought, we're in a world where they're supposed to be heroes if they didn't take advantage of AI? They still have a sense that it's cheating. But you know what I said to them? I said I have to draw the line here, because if everything in there was your ideas — which is what you said, Yajun, and I totally agree — then who you are is the ideas in the paper. I said, if everything in there is your ideas, I don't care if your English is perfect or not, because I'm judging you on the ideas. That was especially true before AI, because of course there was no way they could write like a native speaker. So there are two things: either it was all your ideas, but you got them across in imperfect English; or you plagiarized something, which is a real big no-no; or, even worse, you kind of wrote it but also let AI come up with some points you wouldn't have thought of on your own. I can sort of see there that it's not just you — it came partly from the AI. So that was my point. I said, no matter how good it gets, it's going to be harder and harder, unfortunately, to deduce where the creativity is coming from. It already is getting harder. But I strongly feel — I never counted off on my Chinese students' essays for bad grammar or anything like that. I did count off if they gave up and wrote a one-page thing filled with nothing because they couldn't put it down. I'd say, "You could work harder on that." But that's how I feel about it. And yet everyone is cheating — the teachers are cheating now too. You could come up with a closed test, A, B, C, or D. When ChatGPT first came out, I said, let's see how I can do this. I asked for a closed multiple-choice test — choose option A, B, C, or D — for the exam Chinese people have to take to become an English teacher. I said, "You come up with the topic, format it, everything." Within three seconds it started printing out, perfectly formatted, and the topic was electric vehicles, which is a great topic. Everything was perfect. But then I noticed the correct answer to number one was A. The correct answer to number two was A. Number three was A. The correct answer all the way through was A. And I thought, aha, this can't work. But then I tried it again recently, and that went away. So they learn really fast. Yajun: Yeah, they learn really fast. David: Beyond that, I'm flummoxed. I really don't know what to do about this. One of the proposed solutions is that final tests or essays should be succinct and done in the classroom by hand — or maybe typed out — because in the future, if students go home, they'll come back with everything perfect. So I think that might be a way. It's kind of a kludge, as they say — not a perfect solution — but at least you can be sure that what you're getting on the paper is theirs. I don't know. What do you think, Jeremiah? Jeremiah: I was wondering whether you use any of the available tools, like GPTZero. I noticed that one of them — maybe Google Docs, maybe GPTZero, maybe Grammarly — had a function that records the typing. The person types into the document and turns that in along with the document, because the typing rhythm kind of gives it away. If it's cut and paste, that's one thing. If you just sit there and type straight through without any stopping or thought, then you're probably copying it from another document. But none of these solutions are foolproof. It becomes an arms race — you create tech to detect tech, and then there's new tech that can humanize a paper or intentionally add typos and spelling errors to flummox the detectors. It's an ongoing thing. Whether it was the internet 25 or 35 years ago, then Wikipedia, and now AI, teachers and students are always going to have to deal with this. And again, it will affect what kind of skills students bring with them when they join the workforce, as Yajun pointed out. I didn't used to mark off a lot, but I would mark students off for what I'd call technical errors. When they'd come to me and say, "Why'd I lose three points for typos?" I'd say, "Because someday, twenty years from now, you're going to write a report for your boss, and after three typos he's going to throw it right back at you. And I guarantee you, he's going to be a lot harsher than I was." Yajun: Exactly. I totally agree with that. That's why, when I listen to David — normally teachers are enormously patient with their students, because you're trying to train them to think independently, to develop critical thinking, to learn all the skills. But if we change the scenario — if your students are your consultants and you are their client — do you have any patience for a typo or any error at all? You wouldn't. You'd look at something and say, "Why is there a wrong comma or period in that memo?" and you'd just stop hiring that person. So it really depends on which scenario we're talking about. David: Yeah. I used to work for a law firm for about two years at Guomao. All the staff were Chinese, typing up legal documents. They sometimes had existing samples to copy from, but they also made mistakes. My job was to do the "AI" — I was the H.I., the human intelligence — saying, "Yeah, I know what you're saying, but that's not how we say it, and this is a typo," and so on. I felt this is how it should be: it should be vetted by a human. But I also had so much sympathy with all of them, including my students, because I've had to write lots of things in Chinese, and it sucks in exactly the same way — blatantly wrong characters that happen to be homonyms of something. I've done it and gotten back red ink all over the place: "We don't say that." That will always be the case. So if I were in your shoes — if I had to write a document to the United Nations about something — I'm going to put it through Claude, or whichever is the best one, and make it perfect, with no apologies, because I can never write something the way you can write in Chinese. So I totally agree. But it's a fine line. There are areas where you should expect perfection, and areas where we just have to give up and say, look, we have this tool, let's use it. Yajun: Yeah. In this era — once again, everything is happening so quickly. One or two years ago, people just talked about it and didn't feel we needed this kind of intensive discussion about how we use AI to write on a daily basis. But only two years later, here we are, debating how non-native speakers use AI. So in a couple of years, or even a few months, a new model will come out, new technology will come out, and we'll face new issues. But human beings are extremely adaptable, capable, and flexible, and we adapt to the situation. For me, like you said, I would never write like a native speaker, but selfishly I embrace this technology, because it gives me space, opportunity, and help I wouldn't otherwise have. It makes it possible for me to write several articles in English each week while doing my client work and presenting that work as well as possible to my clients and my readers. From that perspective, this technology is really helpful. I've also noticed more and more Chinese non-native speakers writing on platforms like Substack and LinkedIn — thanks to the technology, I'm very sure all of them use AI. And not only Chinese: Japanese, Korean, and people from Southeast Asia and other emerging markets now have this tool to put them on equal footing with native speakers. Before, it was so hard for them to get heard just because their grammar wasn't good. But now, with these tools, they can find a platform to show what's happening and share their voice. From that perspective, it's really a blessing. Jeremiah: I'll say, too, that I see your point, Yajun. And I sympathize with people who post on LinkedIn complaining about all the AI-generated slop. There are a lot of those posts, they're annoying to read, and it's disappointing when you see native speakers — particularly ones who think of themselves as content creators — and you take a look and it's all created by AI. So I sympathize with people who feel that way. However, I also think there's a subset of people, maybe not even doing it consciously, who are gatekeeping a little bit against this new wave of voices coming in. Frankly, the white guy who explains China might not need to explain it so much if now you have all these platforms — once dominated by native speakers explaining other parts of the world because those people couldn't explain it themselves — and now you have those people explaining it. Is it always perfect? Does it sometimes come across as AI-generated? Possibly. But as Yajun pointed out, if you're dictating your ideas, speaking into the machine, and the machine is helping make those ideas accessible to the greatest number of people, that's a good thing. I can also see why it might be threatening to a few people — or more than a few. Yajun, before we go, the last time we spoke you were in the planning stages of your new firm. I wanted to give you a chance to talk a little about Zhang Global Advisory — the new company, what you're up to, and how people can find you. Yajun: Yeah, thank you. Like you mentioned, last time we chatted we were still in the preparation phase. About three months ago, I launched the firm, Zhang Global Advisory. It's a boutique consultancy, mainly bridging China's technology with the rest of the world. On one hand, we help Chinese tech companies — particularly in AI, physical AI, and robotics — explain their stories to overseas audiences. Meanwhile, we help overseas stakeholders, including investors, companies, business leaders, and political decision-makers, understand what's happening in China's technology field. So far we have a few Chinese clients who are keen to expand their global footprint and share their technology on the global stage. The latest progress we've been making is helping Chinese companies appear on the international stage — for example, at CES, the leading global consumer technology event, hosted in Las Vegas every January. We're also helping Chinese tech founders build their international profile as individuals. That holds a special place in my heart, because in my seven years at the World Economic Forum, I noticed a lot of discussion at the global level about China — particularly, these days, about China's technology — but very few Chinese speakers at that level. So my wish is to help Chinese tech founders build their international profile and bring them to the international stage they deserve. On the inbound side, we've been trying to understand the different markets, from Europe to America — what's their interest in China, how investors can find good companies or industries to invest in, or at least collaborate with, and what the role of local government is. People may not even know that in technology development, local government plays a very important role in supporting companies and industries, and can be a partner for overseas stakeholders. So hopefully, through our work, we can build that bridge to help stakeholders on both sides understand each other better and build some business collaboration. Jeremiah: That's great. Well, thank you so much for joining us, and thank you for your insights. David, it's good to see you sitting up — I hope you're back on your feet and doing wind sprints down the Second Ring Road very, very soon. And thank you, all of you, for joining us today on Barbarians at the Gate. We'll be back in two weeks. And even in the woods of Vermont, I hear the ominous sound of drums. David: Thank you.