ChatXiPT: The Real Story + China Invasion + Jensen + Talk to Your Elders
ChatXiPT: The Real Story + China Invasion + Jensen + Talk to Your Elders
Censors self-censoring!
You, yes you, could be advertising right here!
Respond to this email to connect.
ChatXiPT: CAC Gets a Taste of Its Own Medicine
Article by Bit Wise
Jordan wasn’t able to chat with Xi at last year’s APEC summit. But what if all he needed to do was download the latest CAC model?
The model that flew too close to the sun…
A government-affiliated Chinese research institute has trained a new large language model exclusively on Xi speeches and other official sources. Dubbed by foreign media as “ChatXiPT,” the model apparently fully grasps Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era. Sadly, it hasn’t been released to the public, and for now is used only for internal research purposes.
It was developed by the China Cyberspace Research Institute 中国网络空间研究院, a research institute under the Cyberspace Administration of China (CAC), the Chinese cyber watchdog and internet regulator responsible for ensuring compliance of other chatbots in China.
The release was initially published through the CAC’s magazine via WeChat. After a few days, however, the article was deleted, along with other related promotional posts on Weibo. All that remains today is a three-sentence announcement which does not discuss the model’s central corpus at all. The international media attention must have spooked the creators, leading the censorship authorities to censor their own shiny new model.
…and commercial models already do the job.
That said, international media have probably given “ChatXiPT” more attention than it deserves, because regular commercial Chinese chatbots already stick closely to the Party line, as documented by ChinaTalk and others. In fact, Chinese chatbots are required by law to represent “core socialist values” and go through a registration process through CAC before they are allowed to hit the market.
Chinese models are remarkably capable. They are 9th and 15th on the LMSYS leaderboard today, and some of the best models on the domestic leaderboard haven’t yet joined ChatbotArena. There is no chance that a model developed by a state research institute which relies on second-rate talent and limited compute can reason at the level of private Chinese players.
In fact, some AI companies are already offering tools specific to Party doctrine. For example, ChatGLM, a model developed by startup Zhipu AI 智普 (whose flagship model GLM4 is 15th in ChatbotArena), offers a People’s Daily Pearls of Wisdom 人民日报金句 tool: users to enter a keyword, and the AI will search through a database of People’s Daily articles to find relevant quotes.
For instance, I asked it for five quotes about AI:
The five selected quotes, all from People’s Daily:
“Artificial intelligence is a strategic technology leading this round of technological revolution and industrial transformation, possessing a strong ‘leading goose’ effect with significant spillover and drive.”
“The misuse and abuse of artificial intelligence can, at the least, harm human dignity and exacerbate discrimination and prejudice, and at worst, impact the existing legal systems, affecting government management, national defense construction, and even global governance profoundly.”
“The development cycle of large AI models is long, requiring substantial investment and carrying high risk. After years of continuous research and development, leading international models have amassed considerable resources and talent.”
“Artificial intelligence is an important driving force in the new round of technological revolution and industrial transformation. Accelerating the development of a new generation of artificial intelligence is a strategic issue concerning whether our country can seize the opportunities of this new round of technological and industrial changes.”
“Artificial intelligence technology concerns the common future of all countries. The ethical governance of artificial intelligence should be discussed and decided collectively by countries worldwide, focusing on the special needs of developing countries and striving to address the issues of uneven and insufficient development among nations.”
Does Chat XiPT have a role in training data compliance?
The most interesting part of the “Chat XiPT” announcement is that it may have been trained on a special corpus of “compliant” training data (网信研究数据库). Chinese AI developers talk openly about struggles to access high-quality Chinese language data. For instance, common Chinese datasets are polluted with Japanese porn (it’s a SFW link). In addition, Chinese regulations make it difficult to use foreign data without thorough filtering of content deemed politically sensitive in China.
The government is aware of the challenges this poses to enterprises. The Beijing municipal government sees a role for the state to step in, supporting the creation of high-quality datasets that comply with Chinese regulations, and ideally sharing them freely among Chinese developers.
This dataset could be understood as one of many government-sponsored efforts to create and share high-quality Chinese training corpora. Even so, it’s unclear whether this dataset is really even necessary. As we translated recently, leading figures in China’s AI scene like head of AI research at Bytedance Li Hang 李航 did not seem particularly worried about what language their data came in:
There are currently relatively many high-quality English data resources, while there are relatively few Chinese, especially on the Internet. But through machine translation or some other future technological tool, we can achieve interoperability between the data in the two languages. So I don’t think the gap in language data is going to be a particularly big obstacle.
AI-generated propaganda?
A bigger question is how much propaganda work the CCP even wants to outsource to AI.
The Party certainly acknowledges the potential. In a January 2024 article in the Party journal Study Times 学习时报 titled “Effectively use ChatGPT technology to enhance international information dissemination” 用好ChatGPT技术提升国际传播效能, the author argued,
In the era of smart communication, scientifically understanding and correctly applying ChatGPT technology is of great significance for China to form international discourse power commensurate with its comprehensive national strength and international status.
We should explore the powerful functions of ChatGPT in content production, dissemination, and distribution. The application of ChatGPT has changed the production process and model of international news content, improved the efficiency of international news dissemination, and brought revolutionary impact to international communication. Some foreign-media organizations are increasingly adept at using ChatGPT technology to lead and intervene in international public opinion, which requires our attention. … We should explore the user profiling function to promote precision in international communication.
Officials using AI to write propaganda material is likely already a reality — but not necessarily to the delight of the leadership. An article published in official journal of the Ministry of Human Resources and Social Security in April 2024 warns that AI-generated policy work is leading to formalism:
Many AI companies have targeted the ‘official document writing’ market. … With one click, various official documents and speeches can be automatically completed. … The popularity of AI official document writing tools reflects the problem of the resurgence of formalism. … Although the current AI official document writing is clear and structured, some of the content is too general, the language is relatively stiff, the understanding of the basic situation of the locality is incomplete and inaccurate, and it does not understand the new policies and new expressions of the locality, not conforming to actual work. If Party members and cadres want to make the content of official documents substantive and the policies they formulate effective, they must avoid writing content that is ‘applicable everywhere’, and use new technologies prudently.
The Chinese state is probably already using AI to generate propaganda material. But a new “Chat XiPT” model by a state-led think tank is not a game changer. Commercial models already provide decent capabilities, and do an excellent job of sticking to the Party line. The state may play a supporting role in the generation of datasets that comply with Chinese censorship directives — but even here, it is far from clear whether those datasets will be any better than what enterprises are already developing on their own. The bigger question is whether the Chinese state will be able to deploy generative AI at scale and actually reach audiences to consume the content.
Reasonable rates, unreasonable reach. Sponsor ChinaTalk!
Respond to this email to connect.
’s Highlights from Weibo
A compilation of responses to the question, “Has China ever invaded other countries in its history?”
“Wang Bo’s 王勃 grave is in Vietnam. Why? Because his dad used to be the governor of Vietnam Province.”
“We’ve always been a country of honor and ritual. You need a casus belli to start any war. You have no idea how many history books Zhu Di 朱棣 [Emperor 1402-1424 AD] dog-eared trying to find a legitimate reason to march on Mobei 漠北, and finally saw the Siege of Baima 白马之战 which Liu Bang 刘邦 fought [in 202-195 BC], and decided he had to get revenge for the Han Chinese people XD”
“I knew a Turkish guy before, and he said that we’ve been invading other countries since the start of history. I asked for more details, and he said it was ever since Genghis Khan.”
“Li Bai 李白, famous romantic poet in the Tang Dynasty, was born in Kyrgyzstan.”
“March on the east, war in the west, pacify in the south, police in the north … yeah, none of these are invasions, I’m pretty sure :P”
“I dunno. I just know that we built the Great Wall to defend against foreign invaders and then, somehow, it ended up deep in the heart of our own territory.”
“Based on the map, our ancestors have done everything they can to claim all the best land.”
“You ever heard of Lou Lan 楼兰? It’s currently the Xinjiang Uyghur Self Administration Zone, Ruoqiang County 若羌县. It used to be a country. There was a poem written back in the day with the line, ‘I shan’t return until I’ve destroyed Lou Lan’ 不破楼兰终不还.”
“Guess why I’m named Xinjiang?” [Literally, Xinjiang means “new territories” 新疆]
“In 2,000 years of Vietnamese history, if you cut out the 200 years they spent being invaded by western powers, the other 1,800 years, they were being invaded by eastern powers.”
“It’s not an invasion — we’re educating the barbarians 教化四夷. I’m one of those barbarians.”
“You don’t really think ‘expanding our territory’ 开疆拓土 meant only territory that nobody was using before?”
#2024 college entrance exam takers are so chill
The whole world is cheering them on, but they’re probably thinking about whether they want to dye their hair blonde or red afterward.
A compilation of comments:
“It’s at crucial moments that your phone is the most fun to play with.”
“I honestly don’t even feel like I’m about to take my exams.”
“Lol. I’m 10 days away from my exams, and I’m still scrolling through videos on here.”
“I can have fun for another month before I can have fun for three more months!”
“19 days until I can have fun for the rest of my life.”
“Honestly, every graduating class is like this. It’s only outsiders who think we’re super stressed.”
“You’re wrong. People with good grades aren’t worried. People with shit grades aren’t worried. People with middling grades are having fun while being worried.”
“True anxiety only shows up a few days before your college end of term exams.”
Talk to your elders!
By Arrian Ebrahimi of Chip Capitols
On my flight from Beijing back home to San Francisco last week, I was reading Yu Hua’s 余华 book Chronicle of a Blood Merchant 许三观卖血记 (1995). Reading Chinese books on planes and trains serves dual purposes for me: it’s fun way to deepen my understanding of China, and it also helps strike up conversations with Chinese folks around me who otherwise wouldn’t expect me to speak Mandarin.
At the start of Chapter 6, I came across a term, 粮票 liáng piào. I could guess its meaning (literally, “food ticket,” ie. Mao-era rations), but the context of the passage left me feeling as though I was still missing something: the main character’s wife, Xu Yulan 许玉兰, asked her husband, Xu Sanguan 许三观,
There is no rice at home — only enough for one meal tonight. Here is a food coupon, here is money, and here is a rice bag. Go to the food store and buy some rice.
家里没有米了，只够晚上吃一顿，这是粮票，这是钱，这是米袋，你去粮店把米买回来。
If Xu Sanguan had a ration, why would he also need “money”?
I asked an elderly couple (two retired teachers from Shandong going to the US for the first time to visit their daughter) what a 粮票 was. The two teachers painted me a vivid picture:
Mr. Li: In the Mao era, everyone got monthly ration tickets 粮票 to buy foodstuffs, like bread, rice, the like. The amount of rations you got depended on your province as well as your job/status 取决于地方和单位.
Mrs. Kang: A 粮票 is about this big [folded a napkin to the size of a carnival prize ticket]. There were red ones, green ones … each color denoted a different allowance.
Mr. Li: But remember: a ration ticket wasn’t enough — you also needed money. If you wanted to buy a steamed bun, you had to give the guy the ticket to show you were allowed to buy it, and then you gave money to actually buy it.
I certainly never learned that in a history book! Some things, I think, can’t be learned any other way than by talking to people.
The last thing Mr. Li said reminded me about China’s unique dynamism: “Even your Chinese friends who are the same age as you wouldn’t know what a 粮票 is. They’re too young.”
Moral of the story: talking to seniors anywhere in the world is enlightening. The insights shared by this couple — both of whom were my age during the Cultural Revolution of the 1960s — made Yu Hua’s description of an alien time and place really come to life for me.
Some Tweets
If you don’t know, now you know.
Jordan: This was an excellent interview. He’s doing Ray Dalio Bridgewater stuff, but actually out of love. Here’s my favorite bit:
PATRICK COLLISON: So if we fast-forward just a little bit to the NVIDIA of today. How large is your leadership team?
JENSEN HUANG: NVIDIA’s leadership team is 60 people.
PATRICK COLLISON: And they all report to you?
JENSEN HUANG: Yeah, they all report to me.
PATRICK COLLISON: Sixty direct reports.
JENSEN HUANG: Sixty direct reports, yup.
PATRICK COLLISON: Which is not conventionally considered a best practice… (Audience laughs.) I agree that the best practice kind of...
(Audience laughs.)
JENSEN HUANG: I’m certain that’s the best practice. It’s not conventional, but I am certain it’s the best practice. (Audience applauds.) By the end of this, I’m going to convince all of you to have 60 people on your direct reports.
PATRICK COLLISON: The floor is yours.
JENSEN HUANG: First of all, the reason is because the layer of hierarchy in your company really matters. Information really matters. I believe that your contribution to the work should not be based on the privileged access to information. I don’t do one-on-ones and my staff is quite large. Almost everything that I say, I say to everybody all at the same time.
The reason for that is because I don’t really believe there’s any information that I operate on that somehow only one or two people should hear about. “These are the challenges of the company,” or “This is the problem I’m trying to solve” or “This is the direction we’re trying to go into. These are the new endeavors.” “This isn’t working. That’s working well.” And so all of this type of information, everybody should be able to hear.
I love that everybody’s working off of the same song sheet. I love that there is no privileged access to information. I love that we’re able to all contribute to solving a problem. And when you have 60 people in a room and oftentimes, my staff meetings are once every other week, it’s all based on issues, whatever issues we have. Everybody’s there working on it at the same time. Everybody heard the reasoning of the problem. Everybody heard the reasoning of the solution. Everybody heard everything.
And so that empowers people. I believe that when you give everybody equal access to information, it empowers people. And so, that’s number one, empowering. Number two if the CEO’s direct staff is 60 people, the number of layers you’ve removed in a company is probably something like 7. Depending on how it is.
PATRICK COLLISON: Is it 60 at every layer? As in, if I’m one of the fortunate 60, do I also have 60 direct reports?
JENSEN HUANG: No.
PATRICK COLLISON: Okay.
JENSEN HUANG: I don’t think that that’s scalable downward. And the reason for that is because you need more and more supervision depending on certain levels. And at the E-staff level, if you’re so unfortunate to be serving on NVIDIA’s E-staff, it’s very unlikely you need a lot of managerial.
PATRICK COLLISON: So I rarely find myself having to... stand up for conventional wisdom. But if I were to steel man the other side, I’d say, “Well, one-on-ones are where you provide coaching, where you maybe talk through goals together, personal goals, career advancement, what have you. Where maybe you give feedback on something that you see somebody systematically not doing so well and so forth.” And there is all these things that one is, again conventionally supposed to do in the one-on-one. Do you not do those things, or do you do them in a different way?
JENSEN HUANG: Really good question. I do it right there.
PATRICK COLLISON: Right there in...
JENSEN HUANG: I give you feedback right there in front of everybody. In fact, this is really a big deal. First of all, feedback is learning. Feedback is learning. For what reason are you the only person who should learn this? Now, you created the conditions because of some mistake that you made or silliness that you brought upon yourself. We should all learn from that opportunity. So you created the conditions, but we should all learn from it. Does it make sense?
And so, for me to explain to you why that doesn’t make sense or how I differ from it, half the time, I’m not right. But, for me to reason through it in front of everybody helps everybody learn how to reason through it. So the problem I have with one-on-ones and taking feedback aside is you deprive a whole bunch of people that same learning. Learning from mistakes, other people’s mistakes, is the best way to learn. Why learn from your own mistakes? You know, why learn from your own embarrassment? You got to learn from other people’s embarrassment. That’s why we have case studies, and isn’t that right? We’re trying to read from other people’s disasters, other people’s tragedies. Nothing makes us happier than that.
PATRICK COLLISON: Have you succeeded in getting other leaders at NVIDIA to adopt this practice? Or is that difficult?
JENSEN HUANG: I give people the opportunity to decide for themselves but I really discourage one-on-ones. I really discourage one-on-ones. Nothing is worse than the idea that somebody says “Oh, Jensen wants us to do this.” Why does that have to be said to anybody? Everybody should know. Or, “E-staff said that.” Nothing drives me nuttier than that.
PATRICK COLLISON: You once told me that you really didn’t like firing people and very seldom did it. Can you elaborate on that?
JENSEN HUANG: Well, I’d rather improve you than give up on you. When you fire somebody, you’re kind of saying... Well, a lot of people say “Well, it wasn’t your fault,” or “I made the wrong choice.” There are very few jobs. Look, I used to clean bathrooms, and now I’m the CEO of a company. I think you could learn it. I’m pretty certain you can learn this.
There are a lot of things in life that I believe you can learn and you just have to be given the opportunity to learn it. I had the benefit of watching a lot of smart people do a lot of things. I’m surrounded by 60 people that are doing smart things all the time. They probably don’t realize it but I’m learning constantly from every single one of them. So I don’t like giving up on people, because I think they could improve. So there’s... It’s kind of tongue in cheek, but people know that I’d rather torture them into greatness.
PATRICK COLLISON: That was the phrase that I was hoping to uncover. Yeah, I remember you mentioned that.
JENSEN HUANG: Yeah. So I’d rather torture you into greatness because I believe in you. And I think coaches that really believe in their team torture them into greatness. And oftentimes they’re so close. Don’t give up. They’re so close. Greatness, it comes all of a sudden. One day it’s like, “I got it.” Do you know what I’m saying? That feeling that you didn’t get it yesterday and all of a sudden one day something clicked, and “Oh, I got it.” Could you imagine if you gave up just that moment right before you got it? So I don’t want you to give up on that. So let’s just keep torturing you.
