Deepseek shows that the goal strategy is working

When a small Chinese company called Depseek revealed that it had created an ai system that could match the leading products of the United States in the United States, the news was received in many circles as a warning that China was closing the gap in the global race To build artificial intelligence.

Depseek also said that he built his new ai technology more profitably and with fewer compute .

But the experts in ai within the technological giant taro saw the advance of Deepseek as something more than the arrival of a new and agile competitor on the other side of the world: it was a claim that an unconventional decision taken almost two years ago was the correct call .

In 2023, goal, in a widely criticized movement, he gave his vanguard's technology after spending millions to build it. Deepseek used parts of that technology, as well as other ai tools, available for free on the Internet through a software development method called open source.

Executive targets believe that Deepseek's advance shows that the upstarts now have the opportunity to innovate and compete with the technological giants that have mostly had the playing field to themselves because the ai ​​costs a lot to build. It was something that the executive goal expected to happen when they gave their own technology.

“Our open source strategy was validated,” said Ragavan Srinivasan, a target vice president, in an interview on Tuesday. “The more people have access to the necessary technology to advance faster, the better.”

Goal is also looking closely at the work done in Deepseek. After the goal example, the Chinese company also launched its technology to the open source technological community. Meta has created several “war rooms” where employees are Deepseek's reverse engineering technology, according to two people familiar with the effort they talked under anonymity.

The goal used are looking for ways to reduce the cost of training their software. – A term used to describe the way in which the technologies of ai learn from the data, and apply it to the goal itself <a target="_blank" class="css-yywogo" href="https://www.theinformation.com/articles/meta-scrambles-after-chinese-ai-equals-its-own-upending-silicon-valley?rc=fiplau” title=”” rel=”noopener noreferrer” target=”_blank”>reported In war rooms.

Before the goal, owner of facebook, instagram and WhatsApp, gave its ai technology, the company had focused on projects such as virtual reality. He was caught with flat feet when Openai presented the chatbot chatgpt at the end of 2022. Other technological giants such as Microsoft, Openai's close partner and Google were also very much ahead in their artificial intelligence efforts.

(The New York Times has sued Openai and his partner, Microsoft, claiming the infringement of copyright of the news content related to ai systems. The two technological companies have denied the claims of the demand).

By freely sharing the code that promoted its ai technology, <a target="_blank" class="css-yywogo" href="https://ai.meta.com/blog/large-language-model-llama-meta-ai/” title=”” rel=”noopener noreferrer” target=”_blank”>called flameMeta expected to accelerate the development of its technology and attract others to build on. The engineers believed that the IA experts who work in collaboration could advance more than the teams of experts together within the companies, since they were in OpenAI and the other technological giants.

Goal could afford to do this. He earned money selling online ads, not ai software. By accelerating the development of the ai ​​that offered consumers for free, it could attract more attention to online services such as facebook and instagram, and sell more ads.

“They were the only important US company to adopt this approach. And it was easier for them to do this, more defensible “, Chris V. Nicholson said, an investor of the Risk Capital firm Page One Ventures, which focuses on ai technologies. Goal can offer IA below the cost to build it, or even give it away, to attract customers and increase sales of other services, he added.

Many in Silicon Valley said the goal measure established a dangerous precedent because chatbots could help spread misinformation, hate speech and other toxic content. But goal said that any risk was far exceeded the benefits of open source. And the majority of the ai ​​development, they added, had been shared through an open source until ChatgPT made companies turn to show what they work in.

Now, if Depseek's work can be replicated, particularly his statement that he could build his ai most affordable than most of them had believed possible, that could provide more opportunities for more companies to expand what he did.

“These dynamics are invisible to the American consumer,” said Mr. Nicholson. “But they are very important.”

Yann Lecun, an early pioneer who is the main scientist of the Meta, said in a publication about LinkedIn That people who think that the conclusion of Deepmind's work should be that China is defeating the United States in the development of ai is reading the situation badly. “The correct reading is:” Open source models are surpassing the owners, “he said.

Dr. Lecun added that “because his work is published and open source, everyone can benefit from him. That is the power of open research. “

For last summer, many Chinese companies had followed the example of a goal, regularly open obtaining their own work. These companies included Depseek, which was created by a quantitative commercial company called High-Flyer.

Some Chinese companies offered “refined” versions of technology Open of origin by companies from other countries, as a goal. But others, such as the beginning 01.ai, founded by a well-known investor and technologist named Kai-Fu Lee, used parts of the goal code to build more powerful technologies.

American technology experts still argue that American companies as a goal should not be open to their technologies because they were feeding ai in China. But others say that if US companies stop freely by providing their technology, the epicenter of open source development would simply change to China anyway.

Earlier this year, the students of the University of California, Berkeley built an artificial intelligence system that, in many ways, rivaled the performance of the last OpenAI system. They did it building on two open source technologies launched by the Chinese technological giant Alibaba.

“When you are in a race to build technology, the best way to compete is to share the code, strengthen the base and accelerate the progress rate,” said Clément Delangue, executive director of Hugging Face, a company that houses many of the worlds in the world Open source ai projects.