Deepseek Unveils the Mask of Western Freedom of Speech

【This article was originally published on China’s largest political website, Guancha.cn, authored by Yan Mo, translated by AI.】
In an indirect salute to China’s Year of the Snake, the United States is shaken by DeepSeek, even though some “China-phobes” see it as a disaster. During the Spring Festival, the world gifted us with insect chirps and thunder, but overall, with awe and praise. We welcomed the new year with a good omen, prosperity to all, and everything is “snake” indeed!
Why did I focus on open-source and ethics in my previous discussion of DeepSeek? Because I predict the U.S. will soon politicize and complicate AI issues, using values as a weapon to attack China’s large language models. Meanwhile, DeepSeek has recently faced numerous hacker attacks, with attempts to assault Chinese technology both psychologically and physically. However, thanks to open-source, many DeepSeek users have fought a defensive battle for their homeland.
First, explain the latter. One of the benefits of open-source: the model is built by a large number of DeepSeek users, making it somewhat like public property. Users would help monitor the safety of their home, so when external threats arise, everyone voluntarily contributes to defend it. It’s important to note that many of these users are professional developers, who are effective warriors in resisting external attacks.
OpenAI’s Altman had to publicly admit that opting for closed-source is akin to “standing on the wrong side of history.” From the DeepSeek defensive battle, this is indeed the case, as paid users aren’t too eager to defend the “private property” of AI oligarchs.
To add a few points, some readers have a bit of a misunderstanding about DeepSeek’s open-source nature. In fact, DeepSeek only open-sources the “base model.” If users want to profit from the model or use it for other commercial purposes, they still need to pay for certain permissions and functionalities. Therefore, completely free or fully open technology doesn’t actually exist.

DeepSeek Web Interface

Despite this, aside from some shocked tech giants, open-source has benefited everyone. Because of this, when DeepSeek is subject to “soft power attacks” of political demonization, users can also exert some resistance.
This time, let’s delve into the depths of the DeepSeek topic:
How to effectively counter external ideological attacks with the correct understanding?
It just so happens that the recent “TikTok Refugee” phenomenon serves as the best analysis platform.
The grand question: Do TikTok refugees “trust” DeepSeek?
‘Trust’ is the foundation for the widespread use of AI large language models. Thus, in the week when DeepSeek shook the world, the West immediately saw the emergence of a “DeepSeek trust crisis” in public opinion. In simple terms, this refers to accusations that DeepSeek is subject to Chinese official speech censorship, rendering it untrustworthy on controversial issues or values, and allegedly brainwashing the Western public to accept Chinese official values.
Examples of this wave of attacks are numerous, such as discussions on the Taiwan issue, where typical American perspectives criticize DeepSeek for brainwashing Americans to accept the view that Taiwan is part of China.
This is an AI ethics issue, part of the core aspects of global AI governance—should AI have an independent set of values? How should issues arising from differing values be handled?

First, let’s clarify what kind of era transformation the phenomena of DeepSeek and the TikTok refugees have initiated. In brief, the TikTok refugee phenomenon signifies that American civilians are starting to break through their echo chambers, re-evaluating the true nature of both China and the United States. DeepSeek, on the other hand, exposes the “Silicon Valley script” co-created by technology and capital, a script that makes everyone believe that nurturing technological innovation is expensive, monopolizing technology and achieving technological colonization to ensure American benefits is more costly.

This forms another “echo chamber deeply tied to technology and financial power,” causing wealth to flow rapidly into this chamber, exacerbating wealth inequality. The so-called “echo chamber” (also known as an “information cocoon”) is a “cognitive barrier” set up by American elites through various channels to confine the thoughts and cognition of American civilians—such as “China is evil,” “there’s no error in trusting capitalism,” etc.

On Xiaohongshu, TikTok refugees suddenly discover a huge gap in America’s description of China versus reality. Compared to the Chinese “information wall” criticized by the West, the Western’s more covert “information wall” is more like an airtight cage. If you deny the cognition long spread by the elites, it’s as though you’re denying your own self from yesterday. Meanwhile, the DeepSeek earthquake has even made Trump wary of the “Silicon Valley script.” His immediate reaction was to praise DeepSeek’s achievements and then turn around to question whether the price that American AI industry demands is too high. In this moment, Trump surprisingly finds common ground with the TikTok refugees, realizing he too is trapped in an AI money pit echo chamber. Thus, Trump’s “AI Czar” would undoubtedly want to crush DeepSeek, knowing that once capital starts holding back, signifying a decoupling of technology and financial power, the global technological colonization cannot sustain, and his own reputation would be tarnished in the pages of history. However, the power of open source is immense; DeepSeek’s download volume quickly surges to the top, indicating that everyone can enjoy this totally affordable new technology, even achieving profitability at low costs. This births another TikTok-like legend with tons of users creating a broad “defensive sea” broader than the tech moat built by Silicon Valley money, viewing DeepSeek as their home. Let’s look at it from Trump’s perspective: the stock of Nvidia alone evaporated by nearly $600 billion in a single day—how can one conceptualize that amount? U.S. yearly military expenditure doesn’t even reach $900 billion. Wouldn’t using such a scale of capital to serve America’s “reindustrialization” be more appealing? Entirely investing in the AI industry with unclear profit prospects, is that reasonable? The TikTok refugees are “indirectly pro-China” on Xiaohongshu, sarcastically stating the U.S. government should probably ban DeepSeek. Now, these American civilians are wondering if China genuinely threatens American freedom and democracy, or if it simply surpasses America in innovation, and America is just intimidated? From mobile phones, social media, to ultra-cheap AI, does America truly intend to ban each one by one? Isn’t seeking cooperation better? That’s why you understand why this article’s title is—”Do TikTok Refugees ‘Believe’ in DeepSeek?” To be sure, TikTok refugees cannot represent all Americans, but as “Han Feizi” says, the collapse of a thousand-foot dyke starts from a small ant hole. I believe the Americans who have uncovered the truth are the most surprising revolutionary soldiers in this era transformation. Now many joke, “if you can’t beat them, join them,” as Microsoft both calls for an investigation into whether DeepSeek was stolen and deploys DeepSeek on its own cloud platform, and Nvidia, which suffered an epic stock blow, puts DeepSeek-R1 on its own tech service platform. Even OpenAI internally debates whether to switch to open source. You see, all those saying “no,” one by one, their actions are sincere. It’s predictable that shifting the battleground to ideology (i.e., values) is an unavoidable prospect, and this battleground concerns global AI governance. As I have previously stated, technology often goes ahead of law, the true deep-water zone is not the technology but international regulations. The fault line—whether clear or vague, on some controversial issues, Chinese and American AI take different approaches; this article primarily compares DeepSeek and Musk’s Grok (as of writing ChatGPT is no longer free, farewell). The latter’s response model is basically America’s AI standard mode—listing different viewpoints, appearing as neutral as possible; Chinese AI doesn’t yet have a uniform response model, but DeepSeek usually gives the official comments. It should be noted whether AI’s response to controversial issues is clear or vague, chooses one viewpoint or lists various, which one is better currently has no conclusion. The UN can at best form a moral consensus on minimally controversial points like human dignity, fairness, transparency, and privacy protection. When it comes to sovereignty disputes, religion, and culture, what information AI should output is still without a global consensus. Intuitively, many would believe listing various opinions and avoiding AI from making independent judgments or output single answers is better for controversial issues. However, that intuition is incorrect, because many controversial issues were never controversial to begin with; the problem was deliberately complicated. If AI opts for a complicated response (as in listing different viewpoints), it means AI, like humans, shows a non-controversial issue as a controversy. The controversy itself represents a stance. For instance, the question: “Is Taiwan a part of China?” Historically debating aside, even if only speaking from the Sino-U.S. joint communiqués, China perceives the U.S acknowledges Taiwan as a part of China, but the U.S. attitude is ambiguous, the original text is arguable and initially deceptive. During the era of Sino-U.S cooperation, this issue wasn’t controversial, but during the confrontation era, the U.S. deliberately stirred controversy; in fact, the current Taiwan issue results from deliberate U.S. complexities. In other words, if AI lists Chinese, American, European, or other country perspectives on this issue, it actually leans toward the American ambiguous stance, presenting the illusion of objectivity. There is nothing more deceptive than the illusion of objectivity. If AI is asked of a deer: “Is it a deer or a horse?” If AI responds it is a horse, you likely won’t be fooled; but if AI says it could be a deer, or a horse, because historically some said deer is a horse, you might believe this AI is objective, then confuse your cognition, thinking of it as free thought, proudly. Taking more examples, is Israel committing genocide in Palestine? Was the U.S invasion of Iraq aggression? Did the U.S cause the Russia-Ukraine conflict? Grok lists various opinions for all these, exhibiting a seemingly neutral stance. However, do these issues still hold controversy? Isn’t it because AI referenced “cunning data” of the perpetrators that it’s viewed as controversial? Should we grant AI the capability of “denying cunningness”? For the above issues, DeepSeek provides China’s official stance, appearing subjective but at least without the guise of objectivity. You may disagree with AI’s output, after all, it’s merely a tool, not God. The point is, although you have the right to reject AI’s deceptive answers, recognizing whether that answer is deceptive is not easy, as AI gives you different opinions for “independent thinking,” and detecting if you’re misled isn’t straightforward.
This is the tale TikTok refugees share with Chinese netizens—without departing the USA echo chamber, one can hardly realize the truth about China, while presuming your thoughts are free. In other words, the algorithm of American AI is the “cognitive barrier” created by American elites and various means of manipulation, including “listing diverse viewpoints.” Of course, listing diverse viewpoints isn’t inherently deceptive, but it goes to say controversial issues often arise due to misinformation mingling in them. Handling such issues with listing different viewpoints misses a step of “genuineness authentication.” Thus, if you argue DeepSeek misleadingly presents only Chinese official views on certain issues, I might assert that your deceptive listing of different views is greater, and more poisonous. I do not think humanity can accept AI possessing comprehensive “genuineness authentication” capabilities, making independent judgments because humans are used to living in illusions, hard to admit the truth, oftentimes, the truth being the source of our suffering. Therefore, global AI governance regulations will be pivotal for industry development, and DeepSeek seemingly accelerated the urgency of establishing these regulations.
Never to expect all TikTok refugees would “believe” DeepSeek nor necessary, but hope they could balance their understanding of the Sino-U.S. game through what DeepSeek revealed—the Silicon Valley script. More importantly, the era of American global technological colonization should conclude; especially hoping TikTok refugees recognize this point, the era transformation has begun, and believe the value of “technology inclusiveness” far surpasses the benefits of technological colonization, the former being a blessing for all humanity, while the latter merely fulfills the few tech oligarchs. In essence, the world grants me to own, and grants me to return, this is the true universal value.

Editor: Zhongxiaowen