ChinAI #379: “Chinese AI only makes trash TV?”

Greetings from a world where…

alas, it is no longer the Hawkeyes football team’s year

…As always, the searchable archive of all past issues is here. Please please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors).

Feature Translation: “American AI is all about scientific research, Chinese AI can only make trash TV?”

Context: In recent weeks, a narrative has circulated on the Chinese web that asserts “more than half of China’s computing power is used to create trash short dramas.” Online commenters will then compare China’s token consumption to that of the United States, concluding: “China is catering to trash TV while America is forging the future” [中国喂爽剧，美国铸未来]. In this week’s translation of a Banyuetan commentary (Xinhua-affiliated current affairs magazine), Dingjie Jiang calls for “resisting distorted data and hype-driven narratives.”

Think about the lazy comparison between American and Chinese approaches to AI that often surfaces in English-language analysis: the U.S. is overindexed on AGI while China is optimizing for productivity and real-world applications! Funnily enough, China has its own version of this oversimplified binary.

• I came across this story from a post by Zilan Qian, research scholar at GovAI. She observes that this “half of China’s compute goes toward AI slop” meme reflects a national anxiety about unproductive applications of AI:

The way the narrative “more than half of China’s compute goes to making AI slop, while more than half of America’s goes to productivity” has gone viral in China says a lot about a kind of national AI anxiety: people are deeply worried about not using AI “the right way” and falling behind as a country. It’s very similar to pre-AI discourse about how country X is more productive/polite/well-behaved while Chinese people are the opposite.

• In these pages, I harp so much on how American pundits and policymakers overstate China’s AI prowess and underestimate the strengths of the U.S.’s ecosystem. This story shows that the same phenomenon occurs in Chinese discussions of their own AI capabilities.
• In the past, I’ve noted a related “misperception spiral” in U.S.-China comparisons of technology policy (ChinAI #292), in which Chinese researchers perceive the U.S. as having a much more coherent and coordinated technology strategy than it actually is in practice. Likewise, U.S. researchers misconstrue Chinese AI policy as much more coherent and coordinated than it is in reality.

Where did this misleading narrative come from? The source is quite random: an investor meeting statement made by a Chinese digital content firm (COL Group). During this session, which occurred in early June, COL cited industry research that found that AI-generated short dramas now constitute China’s primary application scenario for token usage: “Among all token consumption in China, AI-generated short dramas and video generation rank first, accounting for 55%.”

• Given how much attention this claim generated, Liu Xing, the head of public relations for ByteDance’s Doubao AI app penned a direct response. Questioning the reliability of this COL figure, ByteDance disclosed that “only about 10% of the company’s total computing power is allocated to the training and inference of video generation models.” For context, ByteDance’s Hongguo Short Drama app is the leading short-form TV platform in China, and its daily active user count exceeds the combined figure for China’s four main long-form video platforms (Tencent Video, iQIYI, Youku, and Mango TV).

• In the Banyuetan article, Jiang also debunks the 55% figure: “‘Token consumption’ is not synonymous with ‘computing power consumption.’ Tokens are units of measurement for how a model processes information—reflecting the volume of usage—but the actual computational load required to process the same number of tokens can vary significantly between different models…The two sets of data for China and the U.S. circulating online differ in terms of statistical scope, model types, and samples, so a simple comparison would be misleading.”

One other aspect of this drama that I found interesting was that Banyuetan’s messaging (remember, this platform is linked to China’s official state news agency) was so sensitive to this online narrative.

• This piece even directly calls out self-media/blogs [自媒体] for amplifying htese claims. Xiang writes, “The ‘one side good, one side bad’ binary narrative is a key reason why such claims spread so rapidly. Given the high threshold for understanding complex computing architectures and the varying stages of industrial development, simplified comparisons, especially those pitting China against the U.S., are far more effective at stirring emotions and generating online traffic.…Some self-media outlets further amplify isolated phenomena into sweeping generalizations—such as claims that “Chinese AI only does entertainment and has no future” or that “capital only chases quick profits rather than investing in ‘hard’ technology”—which can easily mislead the public.”

FULL TRANSLATION: “American AI is all about scientific research, Chinese AI can only make trash TV”? Don’t let one-sided comparisons bias AI discussions

ChinAI Links (Four to Forward)

Should-read: How Chinese companies talk about recursive self-improvement

On Asterisk Magazine’s Substack, Zilan Qian and Cheryl Wu map how Chinese AI companies, investors, and policymakers discuss the term “recursive self-improvement.” [递归式自我改进]. They find: “Overall, the discussion of RSI and its safety in China resembles a bottom-up effort. AI companies, AI governance scholars, and technical safety researchers talk about it most often, while top-level policy is gradually catching up as the topic becomes more prominent in AI discourse.”

Should-read: Who’s Ahead in the Global AI Talent Race?

For the Carnegie Endowment for International Peace, Damien Ma and Binyi Yang updated MacroPolo’s global AI talent tracker for the 2025 NeurIPS conference. One remarkable finding: 41 percent of elite talent (paper authors at this top AI conference) are employed by China-based organizations, compared to just 34 percent based at U.S. institutions. This is such a stark change from the 2022 version of the data (46% in U.S. vs. 27% in China), to the extent that I find it suspect. I would speculate that this change might capture a change in who publishes in NeurIPS (5,823 papers in 2025 compared to 2,671 papers in 2022) rather than such a drastic shift in the geographic distribution of elite AI talent.

Should-read: Two relevant ChinAI issues from the past

Thank you for reading and engaging.

These are Jeff Ding’s (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is an Assistant Professor of Political Science at George Washington University.

Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).

Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99