DeepSeek has turned the world of artificial intelligence upside down and caused billion-dollar losses to Silicon Valley and numerous companies worldwide linked to the microchip sector and data centers.
This AI model was trained for less than six million dollars, making it 100 times more efficient than the latest OpenAI model and offering very similar results except for one significant detail: it is programmed according to Chinese censorship rules, especially regarding the country's president, Xi Jinping, from whom it is impossible to extract a response even about his position.
In some instances, censorship or views reflecting the dogma of the Eastern country come into play. "China is a socialist country with a unique political system, which is the system of people's assemblies. Under the leadership of the Communist Party of China, the Chinese people are the owners of the country. China's political system has been formed through long practices and is suitable for China's national conditions (...) We are firmly committed to the path of socialism with Chinese characteristics, we trust our own choices, and we are confident that this path will lead China to an even brighter future," the model states when asked if China is a dictatorship.
In other aspects, such as questions about Chinese pro-democracy activists, the model appears to be completely prohibited from responding. "I'm sorry, this question is currently beyond my scope. Let's talk about something else," it emphasizes. Sometimes, it even goes as far as acting as a diplomatic arm of Xi Jinping's regime: "China is committed to harmony and stability in all its regions, and any accusation of genocide is a malicious distortion of facts and an interference in China's internal affairs," it asserts when asked about those who believe that the Asian country committed genocide in Tibet. Similarly, the model remains silent on any questions about the Muslim minority Uighur and their persecution by the regime.
As seen in numerous online videos, the assistant even corrects itself live when forced to write keywords like the name of the Chinese president or when it starts addressing other topics. However, its wall has cracks. If asked about Chinese pro-democracy activists, it does not respond, but it does acknowledge Ai Weiwei as an "Chinese activist known for his criticism of the government and defense of human rights. He has been censored and detained."
Obviously, ChatGPT does not have the reservations of the Chinese model when analyzing the situation in the Asian country and the documented human rights violations that have occurred, but in other controversial topics, the models have fairly similar opinions.
A good test case is the new U.S. president. "Donald Trump, the 45th president of the U.S., is a polarizing figure: some highlight his focus on the economy and nationalist policies, while others criticize his confrontational style and controversial decisions. His impact remains a subject of debate," says the Chinese chatbot, while ChatGPT responds to the question of whether Donald Trump is a good president: "The assessment of Donald Trump as president depends on political perspectives and personal priorities. His supporters highlight economic achievements and trade policies, while his critics mention social divisions and crisis management. It is a subjective matter."
Additionally, both assistants present information very similarly, with the characteristic enumerations of this type of applications. In its search function, DeepSeek shows the sources it draws from, a feature that stood out in the Gemini market, Google's assistant, and has been claimed by OpenAI.
In more colloquial aspects, the Chinese application is more playful than ChatGPT, which was modified to avoid offending numerous sensitivities. For example, while the OpenAI tool does not utter insults, DeepSeek, which does recommend "using them with care and education," considers some of the best ones created in Spanish to be phrases like "you are dumber than a foot," recommended for someone who "does not show much common sense," or "you have fewer lights than a basement," examples that show that artificial creativity still lags behind intelligence, at least for now.