March 16, 2025

ikayaniaamirshahzad@gmail.com

While the West Hesitates China Marches Forward



While the US debates whether large language models are useful, whether scaling laws have stalled, and whether their carbon footprint is too high; while Europe drafts yet another regulation to shield us citizens from frontier technology and looks for fresh excuses to underfund innovation—

China marches forward.

The Chinese government has taken a clear stance on DeepSeek: Officials must learn to use it—as a means to master AI and large language models (LLMs)—and integrate it into governance. Nationwide deployment. It’s being applied across the board to reduce workloads, enhance decision-making, and accelerate tasks wherever it makes sense. Unprecedented appreciation for this technology at the bureaucratic level.

Imagine if learning to use ChatGPT was mandatory for government staff in the West. Imagine how quickly we’d sort out where it’s useful and where it isn’t. Imagine if policymakers, the people with the power to change things, gained firsthand experience with LLMs—not for abstract policy concerns or to draft new regulations but as tools in their daily work. Imagine if they gathered the know-how result of a continued, consistent practice. Just imagine. Our world would look radically different today.

But that’s not happening. We’ve fallen behind. If not in innovation yet, then in applicability—and certainly, in our attitude toward AI. The only way to address the challenges AI may pose is through action. China has always understood this. Worry and excessive caution lead nowhere. Our government officials need to get acquainted with ChatGPT; they need to know it at the object level to understand how to use it and what to do next. And how to face China’s initiative.

Innovation and attitude—that’s where the Chinese have always outpaced us. The moment DeepSeek appeared, they moved as one, transforming the entire country into an AI-adapted, 1.4-billion-person behemoth. It has already begun in a few pioneering cities and provinces. It will quickly spread everywhere. I’ve compiled some key excerpts from their news outlets.

All the info I quote was reported as recently as last week: here, here, here, here, here, and here; h/t Teortaxes and OedoSoldier for the leads. I’ve translated them, as I don’t yet know Chinese.

Zhengzhou, Henan province:

On February 15, the first citywide “Frontier Learning” lecture series for Zhengzhou’s leaders in 2025 was held. . . . In his concluding remarks, An Wei, member of the Standing Committee of the Henan Provincial Party Committee and Secretary of the Zhengzhou Municipal Party Committee, emphasized the importance of using AI technology to enhance personal capabilities. He urged all Party members and cadres across the city to thoroughly study and master the use of large models like DeepSeek, fully leverage AI to assist in decision-making and problem-solving, and apply it to advance their work more efficiently and scientifically. By combining the “human brain + artificial intelligence,” individuals can strengthen their abilities and improve their work performance.

Foshan, Guangdong province:

According to a February 18 report by Foshan Daily, the city of Foshan has recently localized the deployment of the DeepSeek AI system, integrating it into the Yuezhi Hui · Foshan City Brain platform and launching the “Government Governance Zone” within the Yuezhengyi workspace. . . . The report highlights that DeepSeek’s application scenarios in Foshan’s government affairs continue to expand. These include promoting the intelligentization of form-filling and data reporting processes to enhance efficiency and accuracy, reducing burdens at the grassroots level, and improving operational performance—thus supporting the “Hundreds, Thousands, and Tens of Thousands Project” (Bai Qian Wan Project).

Changsha, Hunan province:

On February 14, Changsha Digital Group and the National Supercomputing Center in Changsha jointly launched CS-DeepSeek, the first city-level AI application in Hunan Province. This initiative aims to provide intelligent solutions across various industries within the smart city ecosystem. . . . In the field of urban safety, CS-DeepSeek will be deployed to support Changsha’s intelligent urban safety management, covering scenarios such as housing, gas, transportation, natural disasters, and public safety. For example, when addressing traffic congestion, fire hazards, or public safety incidents, CS-DeepSeek can analyze real-time urban operation data, identify potential risks, and enhance the intelligence of early warning monitoring, risk analysis, and emergency response.

Shenzhen, Guangdong province:

Recently, Futian District in Shenzhen, Guangdong Province, introduced 70 AI “digital employees” developed using DeepSeek. These AI employees are applied across 240 government service scenarios, including document processing, public services, emergency management, and investment promotion, covering the entire chain of government operations. . . . According to reports, after the launch of the district’s Version 2.0 government large language model, the accuracy of document format corrections exceeded 95%, document review time was reduced by 90%, and the efficiency of cross-department task assignments increased by 80%.

Ganzhou, Jiangxi province:

According to a February 14 post from the WeChat account Ganzhou Reform, Ganzhou City in Jiangxi Province has recently completed the deployment of the DeepSeek series of large models within its government environment, becoming the first prefecture-level city in the province to successfully launch the model.

Wuxi, Jiangsu province:

On February 16, the Wuxi Municipal Data Bureau announced on its official WeChat account that Wuxi has become the first city in Jiangsu Province to deploy the full-scale DeepSeek-R1-671B model within its government innovation environment. By combining the generalization capabilities of large language models with the specialized advantages of government datasets, this deployment provides enhanced AI capabilities for the city’s Urban Brain platform, reshaping the landscape of intelligent urban governance.

Guangzhou, Guangdong province:

According to a February 16 post from the WeChat account China Guangzhou Release, the Guangzhou Municipal Government Services and Data Management Bureau has recently deployed the DeepSeek-R1 and V3 671B large models on its government external network. The deployment, implemented through the Digital Guangzhou Innovation Laboratory, includes deep adaptation of models like DeepSeek-R1 to domestic hardware.

ChatGPT may have hit 400 million weekly active users—crazy impressive—but at this rate, DeepSeek will surpass that soon.

It all began with DeepSeek, the young underdog. It was them, not Alibaba, Baidu, Tencent, or Bytedance that caught up with top US AI labs. It was them that caused an absolute sensation in the West.

Liang Wenfeng, the company’s “low-key” CEO, reportedly traveled to San Francisco in September last year to connect with his Bay Area colleagues, including OpenAI employees, “to stay up to date.” OpenAI claims to have evidence that DeepSeek “stole” reasoning traces from o1 to distill its knowledge (not quite “suck it out”) into DeepSeek-R1, violating the Terms of Service (shameless). But whether Wenfeng took deep inspiration from OpenAI or built V3 and R1 from scratch no longer matters.

What matters is that DeepSeek now possesses the knowledge and engineering prowess to develop models on par with the best in the West. Computing power is all they need to keep going (Nvidia chips). And although export controls remain intact—and useful—suddenly wielding the industrial and manufacturing might of the Chinese government behind their ambitions makes them an unstoppable force.

Wenfeng, whose successful-quant-trader-to-successful-AI-founder pipeline is unusual by anyone’s standards, was dismissed by even his Chinese peers at first. But that’s no longer the case. He is now seeking external funding for the first time, and alongside other Chinese tech leaders, has recently met with President Xi Jinping. Given the CCP’s reinvigorated focus on AI, it’s reasonable to expect that the outcome of this meeting was extremely good. For them. (I’m low-key bullish; they’re publishing everything!)

DeepSeek plus the entire Chinese tech sector plus the CCP are working to surpass the West on the path to AGI and beyond. Good job Sam.

DeepSeek’s success is not just geopolitical (the US is undeniably nervous and the EU is…well, it is there somewhere) or political (by now Xi Jinping is probably AGI-pilled). It’s also economic and productive. This is where China’s approach to technology fundamentally diverges from the West’s. To them, it is opportunity. (JD Vance didn’t invent the idea.) As much as it pains me to say it, democracy’s greatest flaw in this race is the time lost debating what should or shouldn’t be done. When an external force has already committed to moving forward, deliberation becomes a disadvantage.

China moves in formation, like a monolith—like a hive. Of course, this is a simplification of the complexities and challenges within the CCP and the broader country, but compared to the West, it holds true. There’s no need for voting mechanisms, just orders flowing from the top down. And given China’s economic rise over the past three decades, it’s hard to argue with the results.

(At the same time, knowing the guys who would make the decisions here I’d rather have slug-slow democracy; it’s not an easy call.)

Now that DeepSeek has demonstrated AI’s effectiveness across numerous applications, China won’t pause to agonize over existential risks, ethical quandaries, or whether scaling laws are true laws or a mere empirical pattern that will eventually hit a wall. Nor will they waste time debating whether DeepSeek-R1 should be classified as “AI” or whether official documents should preface every mention of the term with “so-called.” They do care about these questions—they’re not stupid—but their approach is to address them through action—safety nets, support, training—rather than through endless committees and regulatory paralysis.

And still, as much as they like to do stuff instead of talking about doing stuff, they’re not chasing “AI for AI’s sake.” It’s ironic—the only time the West moves fast is when there’s a chance to grab short-term, self-serving gains that pollute the commons.

Unlike the hand-wavers who slap AI onto everything to juice sales—no matter how counterproductive—China stands firmly against this approach. Because the goal to them is not AI—it’s never been AI—but efficiency.

DeepSeek’s greatest advantage over US labs has always been efficiency. Their engineers are talented, hard-working, smart, and knowledgeable but their defining strength is knowing how to get more done with less.

Take Grok 3, for example. It was trained on the largest cluster in the world (Colossus, with 100K H100s, then extended to 200K H100s). That’s a perfect achievement—perfect for boasting about having the largest cluster in the world. That’s what Elon Musk and xAI focused on. Grok 3 is a good model, especially considering the limited training time and the startup’s youth, but it wasn’t trained efficiently. They seem to think that scaling compensates for any inefficiencies as long as the cluster is big enough. They’re right; their attitude is not.

DeepSeek took the opposite approach. They produced a high-performing model—on par with Grok 3, perhaps slightly worse on popular benchmarks—using the smallest cluster among the world’s top AI labs (none is “small” in absolute terms, though). Even if Wenfeng did “steal” trade secrets from OpenAI, DeepSeek still had to work with what little they had and squeeze every drop of performance by tinkering with the stack down to PTX and the CUDA kernels. They made up for the lack of compute with remarkable efficiency.

That efficiency isn’t just a quirk of DeepSeek’s research and engineering teams. It reflects a broader cultural mindset. The Chinese are efficient. It’s a way of being. More efficient than the US and infinitely more so than Europeans. Their strength as a country isn’t one of numbers—though that helps—it’s about attitude. They do what they must and make the most of what they have.

In times of competition—some would say war, yet to be seen if a cold or hot one—knowing how to march in formation efficiently and with the right attitude matters more than knowing how to vote or debate.

Right now, it’s only a few cities—Zhengzhou, Guangzhou, Shenzhen, Ganzhou, and Wuxi—leading the charge. Tomorrow, every local branch of the CCP will follow. In no time, the entire party—some 100 million members—along with industry professionals, will be using the latest Chinese AI, which by that time will be much better than it is today (imagine what DeepSeek could do if they get their hands on a 100K H100 cluster). Instead of squandering time on endless debates about whether AI works or if LLMs are useful, they’re putting them to work. And learning by doing.

It’s funny how here, in the West, the people who criticize LLMs the most are the ones who use them the least. Funnier still, one of their main arguments is that you can’t use AI to do something you weren’t already an expert in—and I agree—yet they do exactly that when they attack AI without understanding how it works or where it’s useful. The Chinese don’t waste time on this. They wouldn’t even read this paragraph.

That’s the unmistakable difference in attitude. It will kill us. In China, proficiency in DeepSeek and AI will eventually be mandatory, not optional—not to use it for everything, of course, but to know, every time, whether it’s the better tool for the job. And perhaps it won’t even come as an order from the president but as a self-imposed mindset: This is how I become better. This is how I embrace the present. This is how I grow into a useful citizen of the times upon me instead of a helpless complainer.

Again, try imagining officials at the European Commission using ChatGPT like that. (I’d mention a European equivalent to ChatGPT, but we don’t have one.) I’m confident many—most, if I had to guess—haven’t even downloaded the app or used it in the browser for anything more meaningful than asking, “How many R’s are there in ‘strawberry’?”

I know this technology is not perfect—no one’s claiming it is—but please, for your sake, mine, and everyone’s, open up your attitude.

This complacency—the self-satisfaction from having once led the world in living standards (no longer the case)—will follow the West to its grave. And for those of us who outlive it, the memory will haunt us as we scramble to learn pinyin and tones back in xuéxiào.



Source link

Leave a Comment