Please let us know your free update
Simply sign up to the MYFT digest of artificial intelligence, and will be distributed directly to the receiving tray.
The writer is the founding partner of a VC company based in Hong Kong. capital
DeepSeek has changed the trajectory of technology worldwide in technology forever. In China, the founder Liang Wenfeng has become a local champion. In a country where overseas degrees, especially in the United States, are still recognized as prestigious than in Japan, students and parents have all the research teams of emerging artificial intelligence. I’m not UN to discover that.
Beijing is so confident that he has never been pursuing this technology. Deepseek’s success impairs the barriers created in the Technical War between the United States and China.
A company based in Hangzhou, which releases low -cost open source AI models, can access the state to the state, from Sao Paulo researchers to emerging Stockholm and Nairobi doctors, along with detailed disclosure of training methods. Is meaning. -Of-the-the-the-the-the-ai is almost never.
In Chinese emerging companies, chain reactions are taking place. A new AI application has been created. Competition will be more intense. The risk chosen from initial venture investment is increasing. Deepseek’s decision to pursue an open source AI model is urging others to do the same and press pressure. The first reaction was the Alibaba QWEN team, which released QWEN2.5 as an open source last month, the night before the New Year in China.
This is an amazing change. After US Startup Openai released the AI model Chatgpt in late 2022, the global digital economy had been linked to control by a small number of high -tech companies. These players increase efficiency. This builds an amazing large -scale model that requires tremendous calculations, energy and capital, while protecting the training method as a corporate secret.
A centralized closed model creates a dangerous feed back loop. The more data they accumulate, the more powerful they are and the more alienated the people outside the gate. For consumers, this means seeing the future of AI without a large price, surrendered data, meaningful participation.
The promise of DeepSeek’s reasoning R1 model lies in its adaptability. When it is open, you can adjust it according to the local needs. Avoid redundant calculations using what is called spa -news network training. In other words, its efficiency reduces calculation and energy needs in an order of magnitude.
This means that advanced AI can benefit not only a small number, but also the masses. It proves that technology is a product. There is no need to waste billions of dollars in the high -tech large -scale competition with a closed model. The value of AI is that it is not a unique model, but everyone can use it.
As an investor, I am worried that DeepSeek’s outstandingness may lead to the United States to choose more severe sanctions. In China, exports of graphic processing units (GPUs) such as NVIDIA’s powerful H100 (GPU) have hindered the growth of startups. Foreign investors’ funds are limited due to compliance risk concerns. But the true danger is to limit access to collaboration between world education and research. This suppresses the flow of global knowledge that is important to maintain progress. Talents can avoid tip deficiency, but we have a barrier to long -term stagnation of learning risks.
However, even with additional US restrictions, even a conspiracy theory targeted for DeepSeek and the painting campaign cannot change the reality that Chinese emerging companies have entrusted AI to the hands of humanity.
Let’s consider this as a moment of history for all noise. In 1440, Johannes Gutenberg handed a printing machine to Europe. This is an invention that broke the monopoly of the unknown of the elite before. Deepseek’s achievements participate in this tradition of making it easier to access information. The low -cost inference model proves that AI can belong to not only those who are stocking code, chips, and capital, but also everyone.
