Chinese AI Researchers Curate Their Public Image on X
The prevailing story in Silicon Valley is that Chinese AI models like Kimi K3 and DeepSeek R1 appeared without warning, as if produced inside a sealed research culture with no contact to the outside world. It is a convenient story, because it turns the Chinese advance into something closer to a natural event than a planned outcome. The truth is more accessible than the story suggests. You do not need a visa, a translator, or an insider contact to sample what Chinese AI researchers are thinking. You need an X account and a willingness to scroll.
A quick search for Moonshot AI, the startup behind Kimi K3, surfaces roughly 30 accounts operated by people who claim current affiliation, including two cofounders, half a dozen former employees, and a rotating cast of collaborators. [3] The count comes from reporting by Wired. These are not dormant profiles kept alive by a communications department. They discuss releases and incremental papers, they befriend Western researchers, and they volunteer details about their weekends and hobbies. The accounts carry the texture of real people, which is precisely what makes them convincing as a window into a secretive industry.
The visible statistics are easy to collect, and that is the problem. Follower counts, post frequency, and the expanding network of Chinese researchers on X have created an impression of radical transparency. The impression is not false, but it is structured. Every post that reaches us has traveled through a chain of incentives that decides what can be said, what should be said, and what would cost someone their passport if it were said aloud.
What a platform like X measures is visibility, and visibility is the raw material of branding, not the currency of understanding. The gap between those two is the story here. It is a gap that Chinese AI researchers will happily discuss, because some of them have drawn the exact conclusion that the rest of us are still forming. They say the AI industry is now like finance. They say branding is the game. They say it with a smile, in perfectly good English, on a platform owned by an American billionaire.
The Migration That Followed Deep
Seek’s Shockwave
The turning point came in January 2025, when DeepSeek’s R1 model went viral around the world and forced a global audience to confront the fact that China’s AI labs were no longer a footnote. [1] For the Chinese researchers who had been working in relative obscurity, the moment was both vindication and a wake-up call. The world was suddenly watching, and watching meant they had a decision to make about how they appeared to that world.
Meng Fanqing, cofounder of the startup Evolvent AI and a former intern at Moonshot who worked on the Kimi K2.5 model, remembers the shift clearly. [4] He says he started posting actively on X in mid-2025, and he describes the period with the calm precision of someone who has already processed the implications. “Around that time, all the Chinese researchers are starting to think about branding on an international level.” [4]
Notice the word he chooses. Not communication, not exchange, not education. Branding. The migration to X was not primarily an intellectual project. It was a professional response to an audience that had materialized overnight, and it was coordinated in the loose, competitive way that markets coordinate when they sense an opening.
The DeepSeek employees themselves illustrate the strange limits of this openness. Several of them post regularly on X, sharing the latest AI news and advertising job openings to recruit new talent. At the same time, they reportedly cannot leave China, because the government has confiscated their passports. They can address a global audience from their desks, but they cannot walk into a conference hall in another country. The voice travels. The body does not.
The Platforms That Chased Their Thinkers Away
There is a reason Chinese researchers describe X as the best available community, and it has as much to do with the decay of China’s own platforms as with any virtue of Elon Musk’s app. Meng puts it plainly: compared with Chinese platforms, where serious technical discussion is hard to sustain, researchers would rather spend their time on X. “X is the best community,” he says, citing the mix of Chinese and overseas audiences and an algorithm that surfaces content from your own circle.
The history behind that preference is a story of unforced errors. Zhihu, the Quora-like platform that once branded itself as the premier destination for educated professionals, began pivoting to fiction around 2020. The experts drifted away, and the content followed them out the door. Deng Yu, a Chinese mathematician, had already made his exit in 2018, publicly writing on Zhihu that he was done because the quality had collapsed.
The younger generation migrated to Xiaohongshu, also known as RedNote, where junior AI researchers have started accounts to share details of their work. But the audience there is not technical, and the tone is not rigorous. The result is a strange inversion: on a Chinese-owned platform, a Chinese PhD student cannot talk about their research without being dragged into gossip, while on an American-owned platform, they suddenly find people who want to discuss model design.
The contrast became explicit in March 2025, when Moonshot released a new research paper and watched the reactions split along platform lines. [2] On Zhihu and Xiaohongshu, the discussion centered on the authors’ backgrounds and ages — one of them was only 17 — and on their parents. [5] A current Moonshot researcher, speaking anonymously because he was not authorized to talk to media, describes the Chinese platform reaction as superficial and gossipy. “It was the foreign readers who actually discussed technical details and praised its breakthroughs.” [2]
There is an irony here that should not be lost. The same researcher who criticizes Zhihu for its mean-spiritedness and AI slop works for a company whose international visibility is now built on that foreign audience. The technical praise coming from X, which feels so much more serious than the Zhihu gossip, is also exactly what a branding strategy hopes to harvest. Both audiences, in a sense, are giving Moonshot what it needs: the domestic one supplies attention, and the foreign one supplies legitimacy.
The Silence of OpenAI, the Stage of Moonshot
The Chinese migration has been made easier by a simultaneous withdrawal on the American side. Researchers at OpenAI and Anthropic, explains Tiezhen Wang, an independent AI analyst and former researcher at Hugging Face, have become more reluctant to share their thinking online as their employers have grown into behemoths. “I guess they feel that their model architecture and how they train the model is kind of a secret, so they’re posting less.”
The vacuum is real, and Chinese researchers have filled it with visible enthusiasm. Wang points to a hunger among X audiences for people who can explain what is happening in the fast-moving world of AI, and Chinese researchers are now the ones doing the explaining. But this is not a story about national character. Openness, here, is an allocation of resources, and the resource being allocated is attention.
Wang knows this from the inside. During a one-off consultation in 2025, he advised Moonshot to push more of its researchers to the front on social media. His reasoning was pragmatic: when a startup releases a new model or paper, “it’s more interesting to see what they as a person want to post, instead of the official announcement.” He adds that this helps the company gather feedback, start conversations with other researchers, and potentially trigger collaboration.
Read that advice again. The goal is not transparency. The goal is a more engaging marketing surface, one disguised as a human being. A researcher who seems to speak personally is more convincing than a corporate announcement, which is precisely why the corporation wants the researcher to speak. The authenticity is manufactured, not in the sense that the person is fake, but in the sense that their presence has been positioned as part of the product.
Moonshot is well equipped for this game, because the company already understands that culture is a competitive asset. Chinese media often describe it as the AI lab with the most interesting internal culture. The CEO decorated his office with band posters and rock memorabilia. The company’s Chinese name, 月之暗面, comes directly from Pink Floyd’s album. The Dark Side of the Moon. Charisma is part of the business model, and the researchers on X are its public face.
The Finance Model of Research
Meng Fanqing’s comparison deserves to be quoted in full, because it explains the motives behind the entire migration. “X is the platform with the highest traffic and exposure. To some extent, the AI industry works just like the finance industry now; each individual and company needs their own branding.”
In finance, the distinction between making money and appearing to make money has always been blurry, because the appearance attracts the capital that makes the money possible. The AI industry, with its dependence on talent and hype, has arrived at a similar equilibrium. A researcher with a large following is not just a researcher; they are also a recruiting tool, a fundraising signal, and a defensive asset against competitors who might hire them away.
This explains why the content of the Chinese researcher accounts is so deliberately unfiltered. Sharing a hobby or a weekend photo is not a distraction from the branding work, it is an essential part of it. A stranger who knows what you look like on a hike is more likely to trust what you say about a training run. The personalization is the strategy.
And it works. The Moonshot network on X has become a functioning media operation, with posts that celebrate releases, contextualize papers, and preempt criticism. Inside China, the company enjoys a reputation for culture and charm. Outside China, it enjoys something more valuable: a set of recognizable human faces associated with its technical work.
The cost of this arrangement is harder to measure. Every Chinese researcher who posts is constrained by the same forces that confine their bodies: state oversight, corporate discipline, and the knowledge that an unguarded sentence can end a career. The sunshine on X is real, but it is the sunshine of a carefully tended garden, not the open landscape it pretends to be.
What the Window Leaves Out
Consider the fate of Su Jianlin, a well-respected AI researcher whose Chinese-language blog became essential reading for Western researchers who could not read the original. Before the migration to X, volunteers translated his posts into English, and the translations circulated through the AI community as a kind of underground press. Su was an independent voice, credible precisely because he had no company to promote. Today he works for Moonshot, and he has a public X account of his own.
The absorption of Su Jianlin into the corporate apparatus is a small parable about what openness means in this industry. The independent voice that the West craved is now part of the same branding machine that tells researchers to post about their hobbies. His knowledge is still valuable, and his technical content is still serious. But the structural independence that made his blog trustworthy has been incorporated into a marketing strategy.
There are also things the window simply cannot reveal. When DeepSeek employees share job listings on X, they are acting as recruiters, but the conditions of their employment are not on display. When researchers befriend Western colleagues, they are building relationships that might someday matter for collaboration, but the political constraints on those relationships remain invisible. The algorithm that shows you content “in the same circle that you are in,” as Meng approvingly notes, is also the algorithm that ensures you never see the circle’s boundaries.
The pattern across all of this is consistent. The Chinese AI industry has learned that openness is a positional good: it benefits the lab that opens up more than the lab that stays closed, as long as the opening is controlled. The result is a performance that is technically genuine. The researchers really do believe what they post. The papers really do contain real results. The honesty is real; what is manufactured is the equivalence between posting and disclosure.
The Door That Could Have Been
Wang’s final assessment of the situation is generous, and it is worth taking seriously. Having Chinese researchers explain their work in their own words on X is a net benefit for the United States, he says, because it reduces the need for blind speculation. “People have a lot of wrong hallucinations about what China is doing. And now, X is the door for them to better understand China.”
Maybe. But a door is a strange object, because it can be opened or closed, and its frame determines what you can see even when it is wide open. The door that exists today has been built by the incentives of the platform and the interests of the companies, and both have shaped the view that it offers.
Compare it with the door that could have been built. Imagine if researchers at OpenAI and Anthropic had continued posting with the freedom they once had, instead of withholding their reasoning behind a wall of corporate secrecy. Imagine if Zhihu had nurtured its expert community instead of chasing fiction revenue. Imagine if Chinese researchers could leave the country to attend conferences and answer questions in person, instead of delivering their carefully framed thoughts through an algorithm controlled by an American billionaire.
None of those conditions held, and so we have the door we have. It is better than a wall, which is a real achievement. But it is easy to mistake the improvement for the solution. We are no longer in the dark about Chinese AI, and we are also not being shown everything. We are being shown what serves the interests of those who control the view.
The 17-year-old behind the Moonshot paper is a good place to end. On Zhihu, readers wanted to know who his parents are. On X, readers wanted to know about the architecture. Both questions avoid the one that actually matters: how can we know what is true, when every source of information has a stake in the story it tells? The window is open, and the view is lovely. The researchers are real, and so are the results. What the window cannot show is the hand that placed it in the wall.
Sources
1. DeepSeek
2. Moonshot AI
3. Wired
4. Evolvent AI
5. Xiaohongshu
6. Anthropic
7. Hugging Face