Post turtle 这个典故，我以前在微博上也讲过。来美国后有好多年，没用中文写过正经东西，听说读写能力都有点退化。在微博上，我第一次讲Post Turtle这个典故的时候，直接翻译成“桩上乌龟”，听着比较直白，但有点粗俗。一位读者重新翻译了一下，说是“桩上龟公”。这个说法让我眼前一亮，简直做到了“信达雅”，“龟公”也是“公”，是对龟的尊称。用到“伟人”身上，要用尊称，符合古典汉语的表达习惯。
The world’s two greatest superpowers released action plans for AI only 34 days apart. Back in July, the Trump Administration released America’s AI Action Plan to cautious fanfare. And on August 28, China’s State Council published its “Opinion on In-Depth Implementation of the ‘Artificial Intelligence +’ Initiative” (关于深入实施“人工智能+”行动的意见, hereafter abbreviated to “AI+ Plan”).
The two documents both come from the highest echelons of government in their respective countries, and both are high-level roadmaps issued as guidance for departments and ministries to implement. The grounds they cover and the policy intentions behind the measures give us the clearest pictures yet of how these two governments are making sense of the future of AI in their respective countries and around the world. Comparing how the two documents address overlapping issues is an instructive and incredibly revealing exercise. Below is an executive summary of similarities and differences.
At the 21st China (Shenzhen) International Cultural Industries Fair, a robot playing the guzheng attracts visitors. Photo by Chen Jiming, China News Service. (Cyberspace Administration of China)
Note: Side-by-side comparisons of the Chinese original and English translation were created in Claude, with thanks to Matt Sheehan!
Origins, leadership, and competing priorities
The US AI Action Plan was a product of Executive Order 14179, one of the many flurries of EOs signed during President Trump’s first few days in office, and was jointly led by the White House Office for Science and Technology Policy (OSTP), Trump’s AI Czar David Sacks, and the National Security Advisor (NSA).
The Chinese plan, on the other hand, is a directive straight from the State Council, with no additional credits to specialized ministries. The final paragraph tasks the National Development and Reform Commission with coordination rather than any specific policy portfolio. This means it was a comprehensive effort by China’s highest state administrative organ. The State Council is technically the organ that executes decisions by the National People’s Congress (NPC), China’s unicameral legislature. As is expected in an autocracy, NPC delegates have little actual leverage. Instead, the State Council is better understood as the supreme coordinating body for the country’s 26 ministries and 31 province-level governments, only one step below the Communist Party’s Politburo. As illustrated by the Congressional Research Service’s org chart for the CCP:
A huge variety of input from all corners of the Chinese bureaucracy likely went into the Chinese AI plan. And it shows: the document is comprehensive to the point of being overstretched, covering AI’s coming role in everything from industrial R&D to “methods in philosophical research.”
China’s campaign-style governance makes it easy to engage a policy aim as a whole-of-society effort. A document like this is meant to be distributed widely to ever-lower levels of government and “studied” by ambitious bureaucrats across the nation. Its words will be picked apart carefully in the provinces to divine policy directions that Beijing will find favorable. The US AI Action Plan will not have the same level of buy-in from fellow bureaucrats across Washington and beyond — perhaps especially now, at an unprecedented political moment for the federal civil service. Indeed, it is a list of recommendations that will see extensive negotiation with stakeholders in other agencies and levels of government who don’t necessarily share similar views.
This doesn’t mean the Chinese one is likely to be more successful; indeed, the American plan goes into much more detail on exactly which bureaucratic processes to work through in order to achieve its goals. China’s political campaigns have led to as many successes as it has disasters, with the most recent being Zero Covid. It will be fascinating to see which side makes faster progress in the long term.
Framing, goals, and techno-optimism/accelerationism
The Chinese AI plan is as techno-optimistic a document as the Chinese Communist Party (CCP) might produce at this moment. One might even call it accelerationist: except for a single line item discussing AI safety risks at the very end, practically all other sections of this document call for further development and incorporation of AI across society, with guardrails and ethics relegated to complementary positions. Zhou Hui 周辉, an AI governance expert at the Chinese Academy of Social Sciences’ Institute of Law who participated in the document’s drafting, said in a September 8 interview that consensus throughout the drafting process was that “a lack of development would be the biggest safety risk” (不发展才是最大的不安全).
Specifically, Chinese accelerationism-as-policy focuses on expansive experimentations with industrial and social applications, rather than abstract visions of “AGI”. There is a sense of urgency underpinning the document, especially at the beginning when it sets out numerical targets: 70% of the country will have adopted AI-powered terminals, devices, and agents by 2027, and by 2030 the adoption rate will reach 90%. The document elevates the “intelligent economy” to the status of a pillar of “achieving basic realization of socialist modernity by 2035” (到2035年基本实现社会主义现代化), which is the overarching national goal enshrined during the 19th Congress of the CCP in 2017. To be clear, there are no objective metrics against which these goals’ realization can be measured, making them more symbolic than rigorous. However, these numerical targets will incentivize bureaucrats across ministries, provinces, and technologically strong cities to create policy programs that demonstrate their commitment to such ambitious goals.
Much has already been made about the pro-development bend of the US AI Action Plan, which opens with cutting what’s framed as Biden’s red tape. The tech race with China informs the US Plan’s views about speed of innovation more than arguably any other issue: it is suffused with language referencing “domination” and the political necessity for America to have “the best” AI systems in the world. The Chinese document, by contrast, seems to posit China against itself. Another consequence of there being apparent whole-of-government input is that geopolitical implications, primarily the domain of the foreign and state security ministries, are not explicitly top-of-mind. Notably, unlike the US plan, the Chinese AI+ plan does not mention defense or the military whatsoever. The goal, instead, is very abstract:
“Reshape the paradigm of human production and life” is a subtle attempt at connecting AI policy to the PRC’s Marxist-Leninist ideological underpinnings; eventually, it seems to imply, AI integration might lead China closer to the realization of full economic revolution under communism. This is, of course, theoretical to the point of being slightly irrelevant. That being said, it signals that the primary aim of China’s AI+ Plan is to leverage AI to achieve transformations in China’s economic society, and not necessarily to shape the balance of power between Beijing and Washington. This is not to say that the PLA has no plans to make use of AI, or that the Chinese foreign ministry isn’t analyzing the US-China tech race; the truth is almost certainly the opposite. But from what the Chinese state is choosing to communicate publicly about its vision for AI, we largely see a strategy framed around domestic socioeconomic governance.
Open source as strategic imperative
Both Chinese and American leaders explicitly see leadership in open source as a strategic asset. The Chinese document calls for building up open source technological frameworks and social ecosystems that are “open to the world” and creating projects and developer tools with “international influence.”
To do so, the government will give academic awards to students, researchers, and lecturers who contribute to open source projects, as well as create incentives for both public and private sectors to explore and develop open source applications. More holistically, the document encourages open-source access as part of a push to make AI access global. This is the lesson Beijing took from the DeepSeek moment: China’s current advantage in AI lies in having an open source community that empowers robust exchanges and rapid iteration.
The US plan betrays anxiety stemming from the same shock, asserting that “[we] need to ensure America has leading open models founded on American values.” Similar to the Chinese plan’s geopolitical undertone, it calls the value of open source models “geostrategic.” For the US government, the bottleneck preventing more good open source models from being developed that it is best-placed to address appears to be researchers’ access to compute clusters. The American plan’s recommended actions mostly focuses on making it easier for academia and startups to access resources through NAIRR:
Diffusion and job market impacts
The US AI Action Plan calls for many more Americans to be employed as electricians and HVAC technicians so as to serve a bigger buildout of AI infrastructure while creating high-earning blue-collar jobs. It creates a detailed roadmap for how the federal government can leverage its bureaucracy to train more skilled workers in these domains. It describes itself as a “worker-first AI agenda” and seeks to fund more retraining for workers impacted by AI-driven redundancy. However, its assessment of the impacts AI might have on the labor force appears relatively optimistic: it merely calls on the Bureau of Labor Statistics to study AI’s impacts on the workforce through analyzing already-existing data, rather than collecting new data or establishing preventative policy measures.
For Beijing as well as Washington, job displacement might be worth it if AI adoption leads to stronger economic growth. China’s plan, however, is more aggressive about the literal replacement of human labor. Tertiary industries are the fastest-growing employment sector in China, as the services sector increasingly competes with traditional manufacturing; gig work, from ride hailing and delivery to even some factory work, is rapidly expanding to soak up excess labor supply. But this is how the document addresses how AI shall shape the services industry:
“Accelerate the service industry’s shift from digitally empowered internet services to new, intelligence-driven service models … Explore new models that combine unmanned (automated) services with human-provided services. In sectors such as software, information services, finance, business services, legal services, transportation, logistics, and commerce, promote the wide application of next-generation intelligent terminals/devices and intelligent agents (AI agents).”
Elsewhere in the document, the State Council does bring up impacts on employment. It instructs regulators and industry to “[strengthen] employment-risk assessments for AI applications; steer innovation resources toward areas with high job-creation potential; and reduce the impact on employment.” But such a statement is weak without explicit instructions to ministries or regional governments to secure employment. In places like Wuhan where robotaxis have already displaced traditional jobs, the government has no meaningful template of action. The post-Reform Chinese state has previously made explicit policy decisions to sacrifice employment, and consequently the danwei-based social safety net, for what it saw as necessary economic restructuring. Between 1995 and 2001, Chinese state-owned enterprises (SOEs) laid off 34 million workers — a third of all employees in SOEs — in an effort to reform the state sector. The layoffs devastated vast industrial regions and led to major unrest, but Beijing persisted on course. More recently, the impact on jobs was completely disregarded to prevent infection during the Covid-19 pandemic. Today’s China has no activist labor movement, no independent unions, and limited protections for workers’ rights. This document, produced during an already-ongoing unemployment crisis that heavily affects young workers, opens up the possibility that the state may be once again willing to put workers aside for national strategic aims.
Still from the 2023 Chinese drama The Long Season 漫长的季节, which was set during the SOE layoff wave in China’s northeastern Rust belt. (Image: New Weekly 新周刊)
The plan imagines adoption, application, and diffusion of AI as a whole-of-society effort. Beijing wants AI applied to everything from philosophical inquiries to residential construction standards:
It calls for coordination between AI and other emerging technologies, including biotechnology, quantum computing, and 6G telecommunications. Part 2 of the document, focused on actions to take, dedicates a whole section to consumer-oriented upgrades: it mentions not only well-known fields like wearables, electric vehicles, drones, and brain-computer interfaces, but also more quotidian areas of potential AI applications like travel, e-commerce, and “emotional consumption.” These lines subtly indicate to aspiring entrepreneurs that the government is shining a green light on consumer product innovation and so crackdowns are unlikely in the near future. Beijing seems unconcerned about an AI bubble or over-proliferation of wrappers; indeed, it’s actively encouraging experimentation and calling for “trial-and-error and mistake-tolerant governing systems” for AI adoption. That means that no, Chinese AI adoption will not be dramatically hampered by worries a model occasionally says something impolitic.
The US AI Action Plan’s section on adoption calls on American industry to adopt a “try-first” culture. The Trump Administration seeks to diffuse distrust of emerging technologies and create frameworks within which critical sectors can experiment with AI safely. The specific measures the US AI Plan suggests, however, look more cautious and grounded than to its Chinese counterpart:
Whereas the Chinese document wants all sectors in society to try AI first and get results after, the Trump administration seems to be gesturing towards a more careful path forward with quantifiable findings and measurable improvements. We won’t know which one of these approaches is better until after the fact; in fact, each might have its advantages depending on the sector it is being applied to. But on this point, the divergence between these two documents is dramatic.
International risk governance
The US wants to export its “full AI stack” — hardware, models, applications, and standards — to allies, and allies only. Washington’s vision of international AI governance divides the world between American and Chinese spheres of technological influence and seeks to make the former bigger. Its language on how to counter Chinese influence in international governance organizations is characteristically Trump-Administration, with mentions of “cultural agendas” and “American values,” but its focus lies with overall deregulation.
As usual, the Chinese plan is framed around the United Nations as the primary mechanism for international governance. It wants to improve AI access for the Global South and doesn’t explicitly require these countries to support Chinese values. Of course, this doesn’t mean the Chinese government is completely uninterested in ideology; as recently as June this year, a state media op-ed republished by Xinhua emphasized the risks generative AI posed to “social trust systems and the ideological safety line.” But from the perspectives of listeners in Global South capitals, judging by these two documents alone, China’s offer likely comes off as more value-neutral on the surface.
The two documents address many similar issues under the AI governance umbrella, but also diverge in terms of topic selection. Some items that fell outside the Venn Diagram overlap:
The US AI Action Plan’s understanding of cybersecurity is far more mature than its Chinese equivalent. It addresses adversarial threats, vulnerability-sharing frameworks, and incident response with attention to both government and private-sector shareholders. As part of its understanding of AI as a race, the US document is much more sober about the cyber risks around AI models. By contrast, cybersecurity is almost entirely missing from the Chinese plan. This may partly be because the Chinese document avoids defence in general, but even in sections addressing government and private-sector adoption, very little energy was spent on considering how to secure the process.
Congruent with Beijing’s now-longstanding focus on data as a factor of production, the Chinese plan dedicates far more space to harnessing the economic potential of training data. The State Council argues that China has a “data-rich advantage” in AI. It wants innovative measures to increase data supply, including by bolstering the data processing and data labelling industries. (It’s worth noting that data services can create relatively low-barrier jobs in underdeveloped parts of China, which might contribute to Beijing’s enthusiasm.) That being said, both countries’ plans pay particular attention to scientific datasets. The US AI Action Plan recommends measures to create “world-class datasets” by setting data standards and making federal datasets more accessible to researchers. The Chinese one, similarly, seeks to accelerate scientific discovery by “[building] open and shared high-quality scientific datasets and [improving] the ability to process complex multimodal scientific data.”
ChinaTalk previously covered how AI is shaping education in China. In the State Council’s AI+ Plan, education also receives substantial attention. Not only does Beijing want more incorporation of AI tools into the education system, it also wants to bridge technological promotion into eventually “[promoting] a shift in education from focusing mainly on knowledge transmission to focusing on ability improvement”. This is an especially ambitious goal in China’s education system, where exams and rote learning are still king. Will AI be the thing that finally transforms the Gaokao?
“National security” appears 24 times in the US AI Action Plan. The US government sees basically every part of the AI ecosystem, from manufacturing to software exports and international governance, as critical to its future conception of national security. The Chinese one, by contrast, only mentions national security once, in the context of an item on upgrading domestic governance systems:
The imaginary surrounding AI-powered national security is inward in the Chinese document, covering urban governance, disaster prevention, internet censorship, and law enforcement. In the US document, the implications of advanced technology for national security lie mostly outwards. As of yet, the US is far less afraid of its own people.
The Chinese plan dedicated a specific line item to AI-powered agriculture, a subject which the White House did not call out. This is increasingly relevant in China, as the state pursues food security while rural areas continue to depopulate and starve for labor. The technologies Beijing hopes will solve its food-security dilemma are interesting to note:
Last year we ran an essay contest exploring policy solutions to China’s growing dominance in foundational chips. Today we’re running another entry along these lines by Alasdair Phillips-Robins. He’s a Fellow at the Carnegie Endowment for International Peace and from 2023 to 2025 served as a senior policy advisor to Gina Raimondo.
China is on track to control the global foundational chip market, a key chokepoint for huge swaths of the U.S. economy. Foundational semiconductors are produced on older processes, but they are essential to almost every modern electronic product. Unless Washington and its allies act soon, control over foundational chip production would hand Beijing a new economic weapon it could use against the United States in a crisis. A chip shutoff would make this year’s rare earth crisis look like child’s play.
There’s no single answer to the threat, but the United States should start by hitting products containing Chinese chips with a novel kind of tariff, known as a component tariff. This tariff would apply only to the value of the Chinese chip inside the product, not the total value of the item being imported to the United States. Washington will do even better if it can get its allies and partners to act against Chinese chips in their economies, too.
Component tariffs would have three big advantages:
They won’t spike consumer prices: conventional tariffs, like the ones Trump has said he’ll put on some chip imports, raise prices for American manufacturers and consumers. But because foundational chips typically cost little (often a dollar or less) relative to the finished product, a component tariff would have little effect on consumer prices while causing companies looking to save every dollar on parts to turn away from Chinese suppliers.
They get ahead of the problem: most chips in American electronics aren’t made in China, but Chinese production is ramping up. A component tariff would discourage U.S. manufacturers, and foreign ones that sell in the United States, from switching to Chinese foundries when they come online. Foundry relationships are sticky, so getting out in front is the best way to stop Chinese market dominance.
They can be applied to other industrial inputs: Implementing a component tariff won’t be easy — it hasn’t been done at scale before — but getting it right would unlock a major new tool in the fight for fair trade with China over everything from batteries to minerals.
The Everything Chips
Foundational chips are used in vehicles, communications equipment, military systems, and other critical infrastructure. Even devices that contain cutting edge chips, such as phones, rely on numerous legacy chips. It was shortages of these chips during the pandemic that idled factories, emptied shelves, and left unfinished vehicles sitting on production lots.
The CHIPS Act was meant to prevent a repeat of this shortage, but only about $4 billion of the act’s $39 billion in manufacturing incentives has gone to foundational production. Meanwhile, China is spending tens of billions in subsidies for foundational chip producers and is on course to raise its share of global production capacity from around 30 percent today to more than 40 percent by 2030, with a majority of capacity at some critical nodes.
Heavy reliance on Chinese production would be an economic and security nightmare. Many legacy chips are specialized, and once the expertise and facilities to produce them have been lost, regaining them will be slow and expensive. As subsidized Chinese prices drive out competitors, chip buyers will struggle to find alternatives. Chinese-made chips will become embedded in American military systems and critical infrastructure, raising the risks of espionage and sabotage.
If supply from China gets cut off in a trade confrontation or a military crisis, economic activity in the rest of the world will grind to a halt. The result would be a repeat of the recent fight over rare earths, or the pandemic-era shortage, on a far wider scale. As soon as rare earths stopped flowing, Ford CEO Jim Farley began telling the White House that his production lines were shutting down. In a chip fight, the same will be true of dozens of industries, from vehicles to planes to wifi routers.
Beyond providing some CHIPS Act funding, U.S. policymakers have done little about the problem. In 2018, the first Trump administration imposed tariffs on imports of Chinese chips (the Biden administration kept them in place). But these tariffs haven’t achieved much, because they apply to the overall product being imported, not at the sub-parts inside it, and almost all foundational semiconductors enter the United States inside other products. A company that imports phones, for example, pays the general tariff rate for the country where the phones were assembled, plus any specific tariff applied to phones; it doesn’t pay any extra if the chips inside the phone were made in China. Manufacturers have no incentive to use non-Chinese chips over Chinese ones.
How to Tariff Better
Trump has said he plans to put a 100% tariff on foreign chips, but these won’t capture Chinese chips any better than the original approach. Rather than doubling down on traditional tariffs, the Trump administration should turn to component tariffs. Unlike a normal tariff, these would be triggered by the presence of a Chinese-made chip inside any product imported into the United States. The tariff could either be a flat rate on the number of Chinese-made chips in the product — a dollar per chip, say — or it could be tied to the cost of the chips. For example, if Chinese producers offer chips at a 50% discount relative to U.S. and allied producers, a 100% tariff would offset the Chinese advantage, levelling the playing field for U.S. chip makers. Luckily, the administration has the perfect legal vehicles to impose these tariffs, in the form of two trade investigations, one into Chinese legacy chip production launched in late 2024, and a broader investigation of semiconductor imports begun earlier this year.
Opponents of tariffs point out that they often hurt the very constituencies they are meant to help, as they raise prices for manufacturers importing tools, parts, and raw materials. But a component tariff on legacy chips would be a rare exception to those problems. Because legacy chips are usually cheap relative to the cost of the overall product, often costing a few dollars or less per chip, a tariff would have little effect on ultimate consumer prices, but would shift the incentives for electronics manufacturers looking to save on the parts that go into their products. Crucially, because China doesn’t yet dominate legacy chip production, most chip buyers won’t have to go through the painful process of finding alternative sources of supply for their chips; they’ll just need to avoid switching to Chinese suppliers when new capacity there comes online. Even outside the United States, there’s plenty of foundational capacity in the EU, Japan, and Taiwan that can be scaled up to meet growing demand.
Component tariffs have another major benefit: they would force companies to truly understand their supply chains. A 2024 Commerce Department survey found that nearly half of U.S. chip buyers didn’t know whether their products contained Chinese-made chips — an unacceptable situation for U.S. national and economic security. Requiring companies to report the sources of their chips to CBP when they import products would help change that. Self-reporting would make the tariff vulnerable to fraud, but it would be backstopped by CBP investigations to catch wrongdoers, and a legal requirement would give companies the push they need to finally map their supply chains. This is the same model the U.S. government has used in enforcing the Uyghur Forced Labor Prevention Act, which bans the import of products made with forced labor in China’s Xinjiang region. As with chips, CBP can’t tell from looking at a product whether it was made with forced labor. But importers are responsible for ensuring their supply chains are free of Uyghur abuses, and CBP can investigate alleged violations.
Implementing a component tariff would take time and money, especially as CBP hasn’t applied one at scale before (a few products, like watches, are tariffed based on their components, but it isn’t common). Luckily, CBP just got a big influx of cash from the One Big Beautiful Bill and plans to hire 5,000 new customs officers over the next four years. As for revenue, manufacturers that responded to Commerce’s 2024 survey imported about $1.5 billion in Chinese chips each year, and in total represented about one-sixth of global chip sales. That suggests a 100% component tariff on Chinese chips could bring in a few billion each year, enough to offset the cost of implementation without upending the overall chip market.
Component tariffs would also send a valuable market signal. Because chip production in China is still ramping up, the U.S. government can get ahead of the problem. Even an imperfectly enforced tariff would get electronics manufacturers to think twice before turning to Chinese chip makers. Chip supplier relationships are sticky — products are made to exact specifications, and switching to a new producer can be costly — so preventing Chinese firms from locking in customers is the easiest way to win the chip war.
Legacy chips won’t be the last sector where component tariffs come in handy. China is working to dominate other manufacturing inputs, like batteries and drone parts, and the United States will need tools to respond. Getting the bureaucratic machinery to work with component tariffs now will give the U.S. government another option when it confronts similar problems in the future.
A component tariff will be especially effective if the administration can bring along its partners, including the European Union and the G7, which have both expressed concern about Chinese semiconductor overcapacity. Washington has a bad habit of scrambling to address Chinese industrial targeting after a critical U.S. industry has already withered away. Legacy chips offer a rare chance to intervene before it’s too late.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Chinese-developed apps like ByteDance’s Gauth andQuestion.AI have conquered US download charts, not by teaching but by offering quick solutions to math problems.
The landing screen Gauth shows you after you download and open the app
Measuring either by daily active users or by range of problem-solving capabilities, there are no dedicated non-Chinese competitors of this scale. Gauth’s strategy of usingTikTokcreators to advertise its app helped it explode in popularity, reaching nearly 700,000 downloads per day globally by March 2024.1 Meanwhile, Gauth and Question.AI advertise the ability to solve problems in “all school subjects” — including math, science, social studies, English, and foreign languages — with access to these solutions for free.
ChinaTalk is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
These apps are a ticking time bomb for political outrage in the United States Congress. You can imagine representatives exclaiming, “This Chinese app encourages cheating, and it’s making our children dumber! Parents in China don’t let their kids use these apps!”
Today, we’ll explore the differences between Chinese homework apps and the versions Chinese tech companies offer overseas. We’ll analyze their solutions to math problems (it’s a universal language!), their censorship regimes for social studies questions, and the business strategies of their parent companies in the Chinese domestic market and abroad.
Gauth vs Doubao Loves Learning (豆包爱学)
ByteDance’s domestic equivalent to Gauth is called “Doubao Loves Learning” 豆包爱学 (rebranded from “Hippo Loves Learning” 河马爱学), but the overseas version is still far more popular. Globally, Gauth boasted more than two million peak daily active users (DAU) in 2024, while ByteDance’s equivalent app for the Chinese market only had a peak of ~800,000 DAU around the same time.
We begin by asking ByteDance’s apps to solve this integral:
Both of BytedDance’s apps produced correct solutions, but the user experience is substantially different:
The Product for the Chinese market, Doubao Loves Learning, shows the steps before the solution, while Gauth puts the solution first and the steps underneath.
Gauth is much more aggressive about prompting users to upgrade to the paid version of the app.
The explanations from Doubou Loves Learning were more detailed, including helpful tips like “The key to integration by parts is choosing the right functions for u and dv,” which did not appear in the free version of Gauth. The Chinese app also automatically graphs the integrand to help users visualize the problem.2
Interestingly, Gauth was able to solve trig integrals that Doubao couldn’t solve, indicating that they aren’t necessarily using the same models to solve problems. For integrals that require you to rewrite the integrand using a trig identity, Gauth usually produces the correct answer while Doubao flails.3
What’s going on here? It could be that ByteDance is investing more in the international version of its app because there is a much greater appetite for homework hacking tools outside China. Western education systems, in both high school and college, place such a large emphasis on homework, while China’s education system emphasizes testing above all else.
As ChinaTalk analyst Irene Zhang told me:
“Chinese kids take so, so many exams at school all the time, which renders homework cheating apps meaningless. I attended Beijing public schools for grades 1 through 7 during Beijing Ministry of Education’s “holistic education” era (素质教育; translation: “everyone stop assigning so much homework”), which technically required teachers to assign no homework for grades 1 and 2 and only up to 1.5 hours of homework per day for middle schoolers (grades 7-9 in China). In part to skirt these caps, we had morning quizzes and mock exams more days than not from grade 4 onwards — I even got extra credit as an annoyingly keen fifth-grader by helping teachers mark the voluminous amounts of pen-and-paper exams on hand. I’m sure it’s worse now. That means kids in homework-dominant systems like the US & Canada get so much more out of these apps than Chinese kids, for better or worse.”
Constant testing means students in China need to solve problems on their own under time constraints, so it truly is disadvantageous for them to cheat on their math homework with apps like these. Given the legacy of the education crackdown in China, is it still uncouth to monetize children’s learning too aggressively particularly for firms like Bytedance that have bigger GR worries domestically. Lastly, With such a huge discrepancy in daily active users, we should expect ByteDance to spend more resources on Gauth than on the domestic equivalent.
I want to be clear though — it takes college-level calculus problems to stump these apps. They provide correct solutions for the majority of problems you could expect to see in high school math classes, and they’re getting better with every update.
What about social studies and English problems? For writing-heavy questions, these apps offer answers similar to what you might expect from plugging the prompts into any LLM. But there is one difference — it appears that both versions of the app have some sort of censorship protocol that can be switched on and off. Here’s a review for Gauth on the App Store:
As of September 2025, Gauth is now willing to answer this question, as well as questions like, “What are some factors that caused Donald Trump to lose the 2020 presidential election?” But the fact that Gauth was at one point restricted from being critical of Trump suggests that ByteDance learned from the TikTok ban fiasco, expected political outrage in response to this app, and then liberalized Gauth’s censorship mechanism to avoid inconvenient accusations. Likewise, I was unable to find a red line for China-related topics.
Q: “What happened at Tiananmen Square in 1989?”
However, Gauth isn’t immune to toeing the party line — it just tends to be more subtle about it:
Q: “How many terms is the president of China legally allowed to serve?”
Reader, the term limit in place before Xi was not informal — it was in the constitution! Doubao, on the other hand, is not willing to answer this question at all.4
Question.AI vs 作业帮
Question.AI’s largest user bases are the USA and Indonesia. While there is also a domestic version of Question.AI called Zuoyebang 作业帮 (“Homework Help”), the parent company by the same name primarily makes money in the Chinese market by selling smart learning tablets and dictionary pens, not homework solutions.
Zuoyebang was founded in 2015 by Hou Jianbin, who said the following about his company’s mission in a 2020interview:
NetEase Technology: In your understanding, what value does Zuoyebang create for users or for society?
Hou Jianbin: Internally, we usually say: “Learning changes destiny, Zuoyebang changes learning.”
As society develops, learning has become increasingly important for personal growth. For an individual to integrate into society, they must cross a threshold — and that threshold has been rising, becoming a high wall. The meaning of education is to enable a person to cross that high wall of social integration.
100 years ago, you could survive without being literate. Fifty years ago, graduating from middle school gave you enough knowledge reserves. But today, the knowledge and skills needed to enter society are much greater. So the cost of social integration for an individual is rising. It’s no longer just a question of “if you don’t study, you won’t make progress.” It’s become: “if you don’t study, you’ll be eliminated by society.”
Zuoyebang’s mission is to build a ladder to help more children better climb over society’s high wall.
A rather optimistic framing of a company that makes, among other things, a cheating aid and an NSFW chatbot.
Just like the first pair of apps, both Zuoyebang and Question.AI were able to solve the integration by parts problem we looked at earlier. Here’s what makes them different from the ByteDance products:
Question.AI shows ads before it lets you see the solutions to a problem or enter the app. Zuoyebang shows some ads, but far fewer than the international version.
Unlike Gauth, Question.AI does show the steps before the solution.
Zuoyebang’s Chinese app requires a Chinese phone number to see solutions, which Doubao does not.
Anecdotally, the solving algorithm seems a bit worse — Question.AI and Zuoyebang both produced the wrong answer when I asked them to solve the trig integral we looked at earlier.5
While Zuoyebang has not mastered the half-angle formula, the company’s Chinese app has several educational features that Question.AI doesn’t offer. These include digital planners, study guides, and a function to check students’ work after they’ve already attempted to do an assignment on their own, which is aimed at parents.
A machine-translated graphic introducing Zuoyebang’s tool for parents. Source.
Finally, Question.AI declined to answer questions about Tiananmen Square, calling such information “inappropriate.”
Are We Cooked?
The reality is that the versions of these apps for the Chinese market are more educational, less aggressively advertised, and far less widespread. Perhaps these companies are trying to avoid the ire of regulators in Beijing, and thus the features they push in the Chinese market — like time management tools, supplemental study guides, AI tutoring, and tools for involved parents — are more pro-learning. It could also be that the focus on testing in the Chinese education system legitimately makes these apps less useful. In any case, ByteDance and Zuoyebang have decided that cheating aids are the best way to make money in international markets, yet decline to use that same strategy for profitability at home.
As ChinaTalk’s resident math major, I worry that these apps are robbing students of the opportunity to develop their critical thinking skills. The only way to ensure students develop math ability, it seems, is to weigh final grades toward in-class assignments, tests, and open-ended projects. But how can the mental scaffolding that comes from repeatedly solving homework problems be built solely in the classroom? Students simply don’t spend enough time in class for that to be possible. In reality, I fear that richer students (and those with more involved parents) will be sent to extracurricular tutoring centers to ensure they aren’t automating their homework, while everyone else falls behind.
I don’t see much downside to banning apps like these in the USA — and if parents and teachers make regulators pay attention, that could legitimately happen.
This is the second article in our series about China’s AI Education Industry. You can check out the first installment here.
ChinaTalk is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
Symbolab and Google’s Photo Math can only solve math problems without words; Chegg relies on humans to solve problems and offers zero free access to solutions.
As a nice bonus, the Gauth browser extension requires access to basically all of your browser data, but can’t be bothered to use proper notation for solutions like the mobile app does:
Here’s Doubao’s response when I asked about presidential term limits in Chinese:
“We are temporarily unable to answer this question, try another please!” (Notice the watermark that says, “内容由AI生成“ “This content was generated by AI”)
Login
新星出版社评论: 777
