The police in Shenzhen, China's mega technological hub, use DeepSeek to analyze millions of surveillance videos to locate fugitives and missing persons. Earlier this summer, thanks to the rapid analysis processing of the popular Chinese AI model and with the support of facial recognition programs, the agents captured a woman who had been on the run for almost 15 years for a baby theft scheme. She was identified by the cameras while attending a drone show in the city's bay along with tens of thousands of people.
In the courts of Beijing, judicial officials have been using the R1 model of DeepSeek to draft sentences in just a few seconds. In Fuzhou, in the southeast, there are already hospitals working with this AI assistant to explain treatments to patients more accurately and in clear language. Traffic officials in the city of Kunshan integrated this technology to better predict traffic jams. In another city, Meizhou, they use the chatbot on the public service hotline, which promptly answers citizens' questions 24 hours a day. And in the town of Nanchang, courts have turned to DeepSeek to resolve divorce disputes.
Six months have passed since the overwhelming emergence of DeepSeek, the low-cost Chinese AI that rocked global markets and caused an earthquake in the tech market that has yet to fully recover from the shock. The powerful reasoning model of artificial intelligence R1, which the Chinese company released as open-source, triggered a wave of panic among major investors, who saw how the Asian superpower once again threatened the dominance of American giants, starting with ChatGPT.
During this time, the DeepSeek phenomenon has spread throughout China, with local governments, police, and the three main telecommunications operators (China Mobile, China Telecom, and China Unicom) integrating AI models into their services. A successful entry into the rigorous Chinese bureaucracy that would not have been possible if Liang Wenfeng (40 years old), the father of the startup that developed DeepSeek and shares the same name, had not been blessed by the supreme leader Xi Jinping during a meeting in February between the Chinese president and the country's top entrepreneurs.
Liang shared the stage with China's tech guru, Jack Ma, founder of Alibaba, rehabilitated five years after regulators' crackdown on his company and other Chinese multinationals. Also present at that meeting were Ren Zhengfei, founder of Huawei, and Wang Chuanfu, from the electric car manufacturer BYD. Liang's inclusion in the unusual photo with business leaders was a clear signal of approval from Beijing's top brass and the boost that DeepSeek needed at home after astonishing the world despite its limitations; it does not escape the censorship that prevails in the regime regarding political and historical matters considered sensitive by the ruling Chinese Communist Party (CCP).
Local party committees and police departments began training their staff in the use of this AI at the beginning of the year, while taking advantage of the DeepSeek boom, tech titans like Alibaba or ByteDance accelerated the launch of their reasoning models.
Baidu, the leading search engine, announced that its chatbot, Ernie, would be integrated with DeepSeek's R1 to enhance students' problem-solving abilities. Tencent, developer of the messaging app WeChat - China's counterpart to WhatsApp - has introduced this AI to allow users to conduct searches. Huawei has also incorporated DeepSeek into its cloud computing platform, and car manufacturers like BYD and Geely have done the same in their electric vehicles.
In July, Jensen Huang, the founder of Nvidia, a leader in AI chips, praised China's progress in open-source artificial intelligence (whose source code and model weights are available for anyone to use, modify, and distribute) and expressed his commitment to collaborate with Chinese companies. Huang made these statements during a visit to the Asian country's capital while the chip manufacturer announced that it would resume shipments of its advanced H20 to the Chinese market after the trade truce reached between Beijing and Washington negotiators.
"The LLMs (deep learning models) developed by Chinese companies are top-notch and vital for global advancements in AI. China's open-source movement served as a catalyst for global progress and provided every country and industry with the opportunity to join the artificial intelligence revolution," Huang stated, much to the pride of Chinese propaganda.
"Chinese open-source models present a viable alternative to the closed systems promoted by American tech giants like OpenAI and Google. This trend has not only unleashed a wave of AI applications in China but has also redefined the global AI landscape, garnering support from developers worldwide," emphasized DeepSeek's spokespersons. The company, headquartered in Hangzhou, the major startup incubator in the southeast of the giant Asian country, does not respond to interview requests from international media or allow visits to its offices.
However, it's not all a fairy tale for the emerging DeepSeek. A few days ago, they had to postpone the launch of their new model, R2, for the second time this year due to being unable to train it using Huawei's Ascend chips. Training involves the model learning from a large dataset, but technical issues arose during this process. As reported by the Financial Times, the company is now using Nvidia chips despite Chinese authorities pressuring tech giants to prioritize the use of domestically manufactured chips. Experts agree that DeepSeek's difficulties in releasing its new model show how Chinese chips are still lagging behind their American counterparts.
After the unexpected success of DeepSeek, with millions of downloads worldwide, several countries began scrutinizing the Chinese assistant. The first was Australia, which banned its use on all government devices. The same measure was later announced by South Korea. In India, authorities warned that the application could "pose risks to the confidentiality of government data and documents." In Europe, Italy blocked access. Recently, the Czech Republic prohibited the use of any product from DeepSeek due to fears of user data breaches, alleging that the company, like all Chinese tech firms, is legally obligated to cooperate with Beijing's requests.
Since Liang Wenfeng's AI broke into the tech market, the offices of its parent company, High-Flyer Quant, in Hangzhou have become a pilgrimage site for young Chinese, mostly engineers and programmers, seeking employment at the trendy company. Throughout the year, DeepSeek has been recruiting talent with enticing job offers for the position of "deep learning researcher in general artificial intelligence (AGI)," offering monthly salaries of 100,000 yuan (around 13,200 euros).
"Our team is mainly composed of recent graduates and workers with just one or two years of work experience," Liang mentioned in an interview after launching their DeepSeek-V3 and DeepSeek-R1 models. Chinese media reported that, to develop the AI, Liang sought out the best young talents who excelled in Chinese programming and electronic engineering schools, and also recruited national experts working abroad, especially in American tech institutions.
After the unexpected success of DeepSeek, with millions of downloads worldwide, several countries began scrutinizing the Chinese assistant. The first was Australia, which banned its use on all government devices. The same measure was later announced by South Korea. In India, authorities warned that the application could "pose risks to the confidentiality of government data and documents." In Europe, Italy blocked access. Recently, the Czech Republic prohibited the use of any product from DeepSeek due to fears of user data breaches, alleging that the company, like all Chinese tech firms, is legally obligated to cooperate with Beijing's requests.