When a small Chinese company called Deepseek has created an AI system that matches the major AI products created in the United States, the world’s gap between China to build artificial is the gap between China. As a warning that he was, many circles were greeted. Intelligence.
DeepSeek also builds new AI technology more expensive, has less computer chips that are more difficult than US competitors in the United States, which costs billions of dollars and hundreds of millions of dollars. He stated that he had shocked the industry he began to believe.
However, AI experts in Tech Giant Meta considered DeepSeek’s break -throughs from the other side of the world to be more agile and new competitors.
In 2023, Meta was widely criticized, spending millions of people, building it, and distributing state -of -the -art AI technology. DeepSeek used some of the technology and other AI tools that can be freely used on the Internet through a software development method called open source.
Meta executives indicate that Deepseek’s breakthrough has a chance to compete with high -tech jians, which has most of the AI stadiums, because AI is very expensive to build. I think the breakthrough is shown. That was what meta executives wanted to happen when they distributed their technology.
“Our open source strategy has been verified,” Meta Vice President Lagavanslinibasan said in an interview on Tuesday. “The more people can access technology, the more they need to move things forward.
Meta also looks closely at DeepSeek’s work. Following the meta lead, Chinese companies have released their technology in the open source technical community. According to the two people who are familiar with the efforts on the condition of anonymity, Meta has created several “war rooms” in which employees have reversible engineering of deep seek technology.
META employees are looking for ways to reduce software training costs. This is the term used to explain how AI technology learns from data. And the information previously reported in War Room applies to the meta’s own AI.
Before handing over AI technology before the meta that owns Facebook, Instagram, and WhatsApp, the company focused on virtual reality projects. When Openai introduced ChatBot Chatgpt in late 2022, it was caught flat. Microsoft, Openai’s close partner, other high -tech companies, such as Google, were quite ahead of AI.
(NEW YORK TIMES complains of Openai and his partner, Microsoft, claiming the copyright infringement of the news content related to the AI system. The two high -tech companies denied the lawsuit.)
By freely sharing the code that promoted AI technology called lama, the meta hoped to accelerate the development of the technology and to build others to build it. Meta engineers believe that AI experts can make more progress than the expert teams that are silent in the company, as they are in Openai and other high -tech gifts. I did it.
Meta can afford to do this. I made money by selling online ads, not AI software. By accelerating the development of AI provided to consumers for free, you can pay more attention to online services such as Facebook and Instagram, and you can sell more advertisements.
“They were the only major American companies that adopted this approach. CHRIS V. Nicholson, investor in Venture Capital Firm Page One Ventures, focusing on AI Technologies. He added that Meta could attract customers and increase the cost of building other services, or even less than the cost.
Many Silicon Valley stated that the movement of the meta has set a dangerous precedent because chatbots help to spread fake information, hate speeches and other toxic content. However, Meta stated that the risk was much higher than the open source. Most AI development added that Chatgpt was shared through open source until ChatGpt was supposed to indicate that the company is working on it.
Currently, if you can recreate DeepSeek’s job, especially the claim that you could build AI at an affordable price than most people, increasing the opportunity for more companies to expand what META has done. You can do it.
“These dynamics are not visible to US consumers,” Nicholson said. “But they are very important.”
Yann Lecun, a pioneer in AI, a meta chief AI scientist, misunderstands the situation in Linkedin’s post, who believes that the takeout from Deepseek is defeating the United States in AI development. He said he was. “The correct reading said,” The open source model is beyond its own model, “he said.
Dr. Lecun said, “Because their jobs are published and open source, everyone can make a profit from it. That is the power of open research.”
By the summer of last year, many Chinese companies chased Meta’s leads and regularly procured their work. These companies contained DeepSeek created by a quantitative trading company called High-Flyer. (On Wednesday, Openai claimed that DeepSeek could have harvested data inappropriately).
Some Chinese companies have provided a “fine -tuned” version of technology procured by companies in other countries such as meta. However, other people, such as Kai-Fu Lee and other famous investors and technicians, have built a more powerful technology using a part of the meta code.
US technical experts argue that US companies, such as meta, should not procure technology because they supply AI in China. But others say that if an American company stops providing his skills freely, the epicenter of open source development will only move to China anyway.
Earlier this year, students at the University of California have built an AI system comparable to the performance of the latest Openai system. They did this by building on two open source technology released by Chinese technical giant Alibaba.
“When participating in competition to build technology, the best way to compete is to share code, strengthen the foundation, and accelerate the speed of progress.” Clémentdelangue, the highest executive officer of Hugging Face, said. Open source AI project.
