DeepSeek: Chinese AI that makes the West tremble

For some time, the Chinese application DeepSeek, an AI assistant very similar to ChatGPT, has gained significant popularity among users, causing concern for its competitors in the U.S. Professors Andrea-Emilio Rizzoli, Adjunct Professor at the Faculty of Informatics at Università della Svizzera italiana (USI) and Director of the Dalle Molle Institute for Artificial Intelligence (IDSIA USI-SUPSI), along with Giovanni Barone Adesi, Professor Emeritus at the USI Faculty of Economics, have commented on the platform's rising popularity in various media outlets in Ticino.

DeepSeek comes from China but also speaks Italian and, of course, English. It has become the most downloaded app on the App Store, causing Big Tech shares to plummet. "The markets are beginning to understand what many speculated last year: the remarkable performance of these companies—Nvidia, for example, gained over 170% in the stock market in 2024—was driven more by projections of future earnings rather than their actual business performance," explained Professor Giovanni Barone Adesi. To address this issue, the newly elected U.S. president, Donald Trump, announced the Stargate programme. This initiative plans to invest USD 500 billion over the next four years to enhance U.S. AI facilities, ensuring they remain at the forefront of the digital world. The world of Artificial Intelligence (AI) is closely tied to significant investments. In 2024, a total of $500 billion was invested in AI globally. Notably, half of the performance gains of the US S&P 500 Super Index in 2024 were driven by companies in the information technology and communication services sectors. This was particularly true for the industry leaders known as the "Magnificent Seven": Google, Amazon, Microsoft, Meta, Nvidia, Apple, and Tesla.

DeepSeek's basic model was developed with an investment of 5.6 million, a drop in the bucket compared to the hundreds of millions invested by U.S. companies. "I consider this to be good news," commented Professor Andrea-Emilio Rizzoli, "we are faced with the development of a model that matches U.S. performance but at a lower cost and in an open-source format, although the training codes have not yet been released". A further advantage of the Chinese platform is the reduced energy consumption needed to train and operate it. The United States has expressed some concern following DeepSeek's appearance on the market: "Up until now, investments and profits have primarily been concentrated in the United States. Therefore, it's understandable that the notion of sharing profits beyond its borders raises concerns. However, from the perspective of technological advancement, more competitors are expected to develop products of equal quality," explained the Director of IDSIA USI-SUPSI.

Commenting on DeepSeek's performance, Professor Rizzoli showed some enthusiasm: "DeepSeek performs similarly to or even better than Western systems; however, its cultural context sets it apart. Currently, it seems unable to answer questions considered inconvenient for the Chinese government. Each model reflects the values of the culture from which it originates. Given that China is a totalitarian government with restricted information circulation compared to Western countries, it is not surprising that DeepSeek avoids responding to certain inquiries". However, for companies, the use of DeepSeek is advantageous, as explained by the USI professor: "China gives companies an open model and gives them the freedom to use it, in return it does not ask for data, but it has the advantage of having a tool based on the Chinese model of thinking used on a large scale". For now, it is not yet known precisely how DeepSeek works: "We stick to the information in the article published by the developers on arXiv, which lacks scientific review. Many researchers are now beginning to test this model on their own machines. However, based on my observations, I have no reason to believe that the claims made are highly misleading or false," said the Director of IDSIA USI-SUPSI.

The current dissemination of DeepSeek has highlighted how technological progress is becoming increasingly polarised: "We are facing a two-way race," commented Professor Rizzoli. "While countries like India and various European states are making progress, the United States and China hold a technological advantage due to their economic resources. In contrast, Europe is fragmented, which hinders its ability to unite and respond quickly". According to the USI professor, competition between the two superpowers is a positive and stimulating factor: "The US and China are competing for technological dominance, aiming to achieve general AI or even superintelligence as quickly as possible, which could also serve as a military asset for both nations". A superintelligence must be a general AI, i.e. capable of reasoning on many domains, as a human being would, but with a much higher capacity. "We are still a long way from this goal, and DeepSeek has not come close, although it has good reasoning capabilities. Large language models (LLM) are not yet able to explain all the steps of the procedures they adopt to solve a problem; in most cases, their most surprising results are obtained by analogy: they solve a problem by re-proposing a solution already seen in another context. For now, the ability to develop innovative, never-seen-before solutions remains uniquely human," Professor Rizzoli concluded.