Deepsek shows that Meta’s AI strategy is working

When a small Chinese company named Deepsek revealed that it had created an AI system that could match the major AI products made in the United States, this news was congratulated as a warning in many circles that China Closing the difference in global race to make artificial. intelligence.

Deepsek also said that it effectively constructed its new AI technology and with a lower hard-to-gate computer chips than its American rivals, shocking an industry, which believed that the big And better AI will cost billions and billions of dollars.

But inside the tech veteran meta, AI experts saw the success of Dipsek as more than the arrival of an agile, new rival from the other side of the world: it was considered that about two years ago an unconventional decision Meta There was the right call.

In 2023, the meta, in a widely criticized move, removed its state -of -the -art AI technology after spending millions to make it. Deepsek used other AI tools along with parts of the technique, which is freely available on the Internet through a software development method called Open Source.

Meta officials believe that the success of Deepsek shows that the upstarts now have a chance to innovate and compete with technical veterans who have mostly AI playgrounds, because to build AI It costs a lot. This was something when Meta officials hoped when they gave their technology.

“Our open source strategy was valid,” said Meta Vice President Ragavan Srinivasan in an interview on Tuesday. “The more people have access to the technology, the better it needs to be pursued, the better.”

Meta is also taking a close look at the work done in Deepsek. After the Meta lead, the Chinese company also released its technology to the Open Source Tech community. Meta has created several “war rooms”, where employees are the technology of reverse engineering deepsake, according to two people, who are familiar with the effort that spoke on the situation of anonymity.

Meta employees are looking for ways to reduce the cost of training their software – Apply a word used to describe AI technologies to describe how to learn from data – and apply it to your AI information of Meta Informed On war rooms.

Prior to Meta, who owns Facebook, Instagram and WhatsApp, left his AI Tech, focused on projects such as company virtual reality. In the end of 2022, Openai introduced the chatbot chatgpt, when other technical giants like Microsoft, Openai, and other technical giants such as Google were well ahead in their AI efforts, then it was caught flat-foot.

(The New York Times has sued Openai and its partner, Microsoft, which claims copyright violation of news material related to the AI ​​system. Both technical companies have denied the claims of the suit.)

By sharing the code independently, which ran his AI technology, Is called lamaMeta hoped to accelerate the development of its technology and draw others to build on top of it. Meta engineers believed that AI experts working collaborated could make more progress than teams of experts, as they were in openi and other technical veterans.

Meta could have risked the risk of doing so. It earned money by selling advertisements online, not AI software. By accelerating the development of AI offered to consumers for free, it can pay more attention to online services such as Facebook and Instagram – and can sell more advertisements.

“She was the only major American company to take this view. And it was easy for them to do so – more defensive, ” An investor Chris V with Venture Capital firm Page One Ventures. Nicholson said, which focuses on AI technologies. Meta can offer AI below cost to make it – or even give it away – to attract customers and increase sales of other services, he said.

Many of Silicon Valley said that the Meta’s move set a dangerous example as chatbots can help spread disintegration, abusive language and other toxic materials. But Meta said that any risk was removed from the benefits of the open source. And most AI development, he said, was shared around through the open source until the chat did not let companies show what they are doing.

Now, if the work of the lampsak can be repeated – especially it claims that it was able to build its AI, most possibly possibly more companies for more companies to expand more companies. Meta could provide which Meta did.

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

Yan Lake, an early AI Pioneer who is the head of Mata AI scientist, said in a post Linkedin Those who think that the work of Deepmind should be a tech that China is beating the United States in AI development, doing the situation wrong. “The correct reading is: ‘Open Source models are crossing the ownership people,” he said.

Dr. Lakeun said that “Because his work is a published and open source, everyone can benefit from it.” This is the power of open research. ,

Until the last summer, many Chinese companies followed the lead of the meta, opening their work regularly. The companies included the lampsac, which was built by a quantitative trading firm called High-Flire.

Some Chinese companies offer “Fine-Tuned” versions technology Pouring open by companies from other countries like Meta. But others, such as start-up 01.ai, established by a famous investor and technologist named Tech-Fu Lee, used parts of the Meta code to manufacture more powerful technologies.

US tech experts have still argued that American companies like Meta should not source their technologies as they were fueling AI in China. But other people say that if American companies stop providing their technology independently, the sub -center of the open source development would just move to China anyway.

Earlier this year, students from the University of California, Berkeley, created an AI system, in many ways the performance of the latest system of Openai was rivals. He did this by constructing two open-sources technologies released by Chinese tech veteran Alibaba.

“When you are in the race to manufacture technology, the best way to compete is to share code, strengthen the foundation and speed up the rate of progress,” said Clament Delangue, Chief Executive Officer of Hugging Face, said, a The company that hosts many people of the world Open-SUS AI Projects.