中國 AI 圈聲音愈來愈多，全球讀者是否更瞭解？

Liu Shengyu is a machine learning system and kernel design engineer at DeepSeek. He probably never imagined that his name would hit the headlines of the world’s mainstream media, such as the South China Morning Post and Business Insider, after his blog on WeChat sharing anxieties about being replaced by AI unexpectedly went viral.

By conventional standards, Liu is a genius. He served as captain of Peking University’s supercomputing team and represented his university at an international competition. But a young genius like him said in his blog, I Had to Bury My Talent in Yesterday, that frontier AI is now fast enough to optimize its own low-level GPU code. He said he is rethinking his career.

Yet that is not the main reason his name spread. At the end of his blog, Liu made a controversial analogy comparing Anthropic CEO Dario Amodei to Adolf Hitler:

I don’t trust Anthropic or OpenAI to do this, and I particularly don’t want Anthropic to control the most advanced AI or AGI. To put it in more extreme terms, the stakes would be no less serious than letting Hitler acquire atomic bomb technology before the Allies (in WWII).

It hit headlines as an insider at the world’s most mysterious AI lab—a company whose executives rarely make public appearances—voiced fierce opposition against one of the top U.S. frontier labs. Coincidentally, this blog was released right after Amodei’s We Must Pace the Frontier essay, in which he again advocated for slowing the growth of China’s AI capabilities.

As global attention on China’s tech and AI industry rises, local, Chinese-language content is increasingly placed under a microscope, flowing into western media and social channels both intentionally and unintentionally. The question is how we view this content and what its broader implications truly are.

Share

The Evolution of China Tech Storytelling

For a long time, understanding China tech required heavy interpretation through layered media filters.

Traditional outlets like Reuters, Bloomberg, CNBC, and South China Morning Post, alongside correspondents from the Wall Street Journal, The Information, Financial Times, Nikkei Asia, etc., acted as the primary ground-level reporters. As China’s tech footprint expanded beyond China, new publications like Rest of World and The Wire China emerged to cover China tech and its global impact.

Today, reporting on China tech is no longer confined to traditional “China beat” reporters. A Silicon Valley-based WSJ reporter might cover Moonshot AI CEO Yang Zhilin on why he decides against a U.S. career, while a New York Times’ San Francisco reporter analyzes how Z.ai’s latest LLM GLM-5.3 tests the world’s cybersecurity. Chinese tech companies are no longer boxed in by a “China” label; when their models and products impact the global market, they are increasingly evaluated simply as tech companies.

Just last week, I attended a great event in San Francisco hosted by Rest of World, titled “Do Chinese AI models signal a new world order?” It was the only event focused on Chinese AI during SF Tech Week, but I hope to see more events like it in the future.

But at the same time, some dedicated China tech publications and newsletters, such as WIRED’s Made in China and The China Project, have unfortunately shut down due to editorial decisions and weak sponsorship interest amid geopolitical tensions.

Outside traditional newsrooms, independent expert voices have ramped up. Observers such as Kyle Chan (High Capacity) and Rui Ma (Tech Buzz China), and a network of thoughtful journalists, think tank researchers, and commentators have long provided multi-polar perspectives rather than zero-sum narratives. Recently, Matt Sheehan’s appearance on Ezra Klein’s New York Times podcast further proved how independent experts are playing vital roles and shaping the narrative.

We also see examples of deep, on-the-ground reporting. Afra Wang recently published an immersive look at China’s AI development in Lingzhu—a small town near Alibaba’s headquarters in Hangzhou—featuring local applications an app that reads your bazi, a traditional chart tracking birth dates and elemental forces. Wang brings both her Chinese origins and western analytical skills to serious China AI coverage, but this level of on-the-ground depth remains rare.

Inflow of New Chinese Tech Voices

What’s changing the old media gatekeeping model over the past three years is the inflow of new China tech content.

First, Chinese researchers and scientists are opening accounts on X, communicating directly while cross-posting thoughts just as they would on China’s social media like Weibo or Zhihu. As WIRED reported, researchers from labs like DeepSeek, Moonshot, MiniMax, and Z.ai are active on X to explain their work, recruit talent, and build global brands. We have already seen Z.ai’s Tang Jie exchanging views with Elon Musk on the platform. While U.S. labs like OpenAI and Anthropic have grown quieter or more guarded, Chinese teams post continuously about breakthroughs.

Second, not just researchers, domestic media and influencers are also going global. Chinese-only tech and business outlets such as Chaping差评, Zhidongxi智东西, LatePost晚点 are expanding onto X and Substack. China’s tech and business journalism is indeed high quality. For example, I am a big fan of LatePost whose exclusive, investigative stories often shape the narrative of China’s tech ecosystem. I personally also translated a few LatePost reports in my newsletter.

Influencers like 数字生命卡兹克 post primarily in Chinese for local and overseas audiences. Thanks to X’s instant translation features, even Chinese-language content can also reach a broader demographic.

Simultaneously, AI translation has made raw primary sources like DeepSeek researcher’s blog, leaked closed-door transcripts from CEO Liang Wenfeng, and interviews with figures like Yang Zhilin, spread in the western media shortly.

The Paradox of Abundance

One content creator I deeply follow is Zarbot. While I don’t know his full background, he appears to be an AI infrastructure expert with years of experience at companies like Oracle. His WeChat blogs on GPU kernels and LLM architectures are among my favorites (even though they are difficult to understand). This year, he joined X. I believe global users will be amazed by his deep technical expertise.

This level of granular depth and diversity from talented insiders was previously locked behind language barriers and platform restrictions. What’s more important is we could be no longer bottlenecked by the overarching “US-China AI race” macro-framing. Global audiences can access disaggregated, bottom-up content straight from local Chinese communities. Here I want to quote a tweet from Kyle Chan before.

We’re living in a golden age of China tech analysis.

This accessibility comes with a cost, however, because content abundance without context could create distortion.

AI translation closes the language gap, but it carries little cultural or subcultural literacy needed to read correctly. When a researcher’s personal venting is spread out of its domestic setting and broadcast globally, foreign readers could mistake it for industry consensus.

Liu’s blog is an example. Most of it is about a kernel engineer watching AI absorb the skills he spent years building. What distributed was the final paragraph compressed into a headline: DeepSeek engineer says Anthropic AGI would be “like Hitler getting nukes.”

It does admittedly reflect the dislikeness many in China feel toward Amodai, given his repeated portrayal of China as a threat. But in my opinion, comparing Amodai to Hitler does not carry the same weight in China as it does in the west.

Although many draw comparisons between WeChat blogs and Substack, the two platforms are quite different. WeChat hosts Chinese media outlets, including state-backed publications, as well as government services. It is also a space where people write personal blogs under pseudonyms without necessarily seeking wider attention. Liu, for example, writes under the pseudonym intlsy的狗窝 and has published only seven posts on WeChat.

The second problem is that abundance is not a random sample. Content does not cross the language barrier on its own. Someone chooses to translate it, and social media’s algorithm rewards the translations that provoke the strongest reactions. For example, U.S. hawks would surface material that confirms Chinese hostility. Those favoring engagement prefer content that shows openness. Readers then consume whatever fits their priors. In that light, my previous belief that we are no longer bottlenecked by the US-China AI race framing may look too optimistic.

The Liu event may have been reviewed inside DeepSeek and other Chinese labs, and it will likely make researchers more careful about what they post. Would a disclaimer have helped? Had Liu said that the blog reflected only his personal views, DeepSeek might have been somewhat insulated. I doubt it would have made much difference.

Going forward, I expect people at high-profile labs, in China and elsewhere, to write with a global audience in mind and to think twice before posting, which is the irony: The more closely western readers watch Chinese insiders, the less candid those insiders could become.

Subscribe now