Deepseek, the Chinese artificial intelligence start -up, has climbed the top of the download ranking on the App Store, surprising analysts for its ability to match the performance of the main American rivals. This “chatbot” is putting the High-Tech industry in agitation, in particular the American giants such as Nvidia and Meta, which have spent colossal figures to dominate the in full expansion of the AI.
What’s known
Deepseek was designed by a startup based in Hangzhou (Eastern China), a city known for the high concentration of technological companies. Available as an application or on computer, it offers a series of functions similar to those of its western competitors: It can write songs, help to face daily situations or suggest a recipe suitable for the content of the refrigerator. Deepseek is able to communicate in different languages, but has declared to the FP that he is competent especially in English and Chinese. However, it shares the limits of many Chinese conversational agents. When asked to talk about sensitive arguments, such as President Xi Jinping, he prefers to avoid the topic and suggests “talking about something else”.
Despite this, its performance, both in the writing of complex code and in the resolution of difficult mathematical problems, surprised the experts. “What we found is that Deepseek (…) is the best or on par with the most performing American models,” he told CNBC Alexandr Wang, CEO of the American company Scale Ai. This success is even more surprising if we consider the resources used. According to an article that describes its development, the Deepseek model has been trained with only a fraction of the chips used by its western competitors.
Open source
Like its western rivals, such as chatgpt, Llama and Claude, Deepseek is based on a large linguistic model (LLM), made up of immense quantities of text, to master the subtleties of natural language. But unlike competitors, which develop ownership models, Deepseek is open source. This means that the application code is accessible to anyone, allowing you to understand its operation and modify it. “A non -US company continues the original Openii mission: an open and forward -looking research for the benefit of everyone,” said Jim Fan, head of the research of Nvidia, on X. Deepsiek claims to be “the avant -garde of the Open models Source “and to rival” the most advanced owner models in the world “.alexandr Wang of Scale Ai declared on X that Deepseek is “an alarm bell for America”.
An ‘incredible’ discovery
Many analysts thought that the advantage of the United States in the production of high -performance chips and their ability to limit China’s access to this technology would have guaranteed their domain in the AI. However, Deepseek claims to have spent only 5.6 million dollars to develop its model, a negligible sum compared to the billions invested by the American giants. The actions of the main US and Japanese technological companies collapsed in front of the challenge launched by Deepseek. Nvidia, a world leader in the production of components and software for artificial intelligence, saw its title descend by over 3% to Wall Street on Friday. The Japanese SoftBank giant, key investor in an US 500 billion dollars project for the development of IA infrastructure, lost more than 8% today.
Marc Andreessen, investor and close councilor of the President of the United States Donald Trump, defined Deepseek a turning point “for the Ai”in the same way that the “Sputnik” was, with reference to the launch of the Soviet satellite that started the race for space during the Cold War. “Deepseek R1 is one of the most incredible discoveries I’ve ever seen,” he wrote on X.
Deepseek denounces a hacker attack on vast scale ‘
Deepseek denounced a “harmful and vast scale” computer attack and which for this reason had to temporarily limit users’ recordings. The company has specified that those who already have an account will be able to access normally.