An assistant to artificial intelligence created by the Chinese startup Deepseek became the most -downloaded application at the weekend in the American Apple App Store, causing a wave of shock in the Silicon Valley and causing a pointed decrease in the stock prices of the main technology companies. Nvidia saw more than $ 460 billion On Monday, erased from market capitalization, which Bloomberg described as “the largest in the history of the American stock exchange.”
The shocks result from the Open Source model developed by Deepseek called R1, which debuted at the beginning of this month. The company said that it can compete with the current industry leader: OPENAI 01. However, the technology industry was the most surprised that Deepseek claimed that he had built its model, using only a compact part of specialized computer chips, whose companies dealing with artificial intelligence usually need to develop for development cutting tools. Break systems.
On Monday, Deepseek stated that it temporarily limits fresh registrations, citing “malicious attacks on large -scale” on the company’s services, according to the report message on your website.
The R1 Deepseek model “undermines the view that Western companies dealing with artificial intelligence have a significant advantage over Chinese”-Jack Clark, co-founder of the Anthropic start-up, dealing with artificial intelligence, wrote in your newsletter. High -risk capital investor Marc Andreessen called This is “AI moment for Sputnik.”
Cheng Lu, a researcher at Openai, said that Chatbot Deepseek showed impressive conversational skills in Chinese. “For the first time I can feel the beauty of the Chinese language created by Chatbot,” he said In the post x on Sunday.
The AI Deepseek assistant is currently available free of charge and offers three main functions. First of all, users can ask chatbot questions and receive direct answers. For example, when Wired asked for ideas for a recipe with grenade seeds, Chatbot Deepseek quickly presented a list of 15 options, from yogurt parfits to rice pilaf “inspired by the Middle East”, but did not give any specific boss or recipe.
The Deepseek application also has a search mode that displays answers from the Internet. When Wired asked: “What are important news today?” Chatbot Deepseek referred to the suspension of weapons between Israel and Hamas and gave links to several western information services, such as BBC News, but not all articles seemed to be related to the topic. Ironically, one of them concerned the Modern York Times article about the impact of Deepseek on the stock exchange.
Finally, the “Deepthink” mode is available, which allows users to apply the Deepseek R1 model, which was built based on the existing company technology V3 model. The difference between them is that R1 has the so -called “reasoning” abilities, which allow him to explain step by step how he came to his conclusions. For example, to the question: “What are the most important historical events of the 20th century?” Deepseek initially gave a long, complicated answer, which began with a number of general questions.
“It’s a hundred years, so a lot happened,” we read in the answer. “I should probably divide it in decades or main topics, such as wars, political changes, technological progress, social movements, etc.” Then Chatbot Deepseek began to quote World War II, Frigid War and Holocaust.
However, before R1 managed to finish his answer, the whole answer disappeared and was replaced by the message with the content: “I’m sorry, I’m not sure how to approach this type of question. Instead, let’s talk about problems with mathematics, coding and logic! ” Many experts and pioneers they noticed Deepseek seems like other technological platforms operating in China Censorship comprehensively Topics recognized by the communist Party of China as sensitive
However, despite these restrictions, free chatbot Deepseek can pose a grave threat to competition such as OpenAI, which charges a fee of $ 20 a month for access to its most powerful models of artificial intelligence. Unlike its Chinese counterpart, OpenAI does not reveal the basic “scales” of its models that determine how artificial intelligence processes information. She also refused to make the full “thought chains” produced by her own reasoning models.