A little-known Chinese artificial intelligence lab has sent panic across Silicon Valley with the release of an artificial intelligence model that outperforms the best models in the U.S., despite being cheaper to make and with inferior chip performance.
DeepSeek (as the lab is called) launched a free, open-source large language model in late December that explain It took just two months and less than $6 million to build, using a reduced-power chip that Nvidia calls the H800.
The new developments raise concerns about whether the United States’ global leadership in artificial intelligence is shrinking and raise questions about the huge spending by big tech companies on building artificial intelligence models and data centers.
In a set of third-party benchmarks, DeepSeek’s model outperformed YuanThe accuracy of Llama 3.1 from OpenAI, GPT-4o from OpenAI, and Claude Sonnet 3.5 from Anthropic ranges from complex problem solving to mathematics and coding.
DeepSeek on Monday released r1, an inference model that also perform better than OpenAI has achieved the latest o1 results in many third-party tests.
Microsoft CEO Satya Nadella said at the World Artificial Intelligence Conference: “Seeing the new DeepSeek model, it is impressive how they have really effectively completed an open source model that can do Inference time calculations, and supercomputing efficiency “We should take developments in China very, very seriously. ”
DeepSeek also has to deal with stringent semiconductor requirements limit The U.S. government imposed this measure on China, cutting off China’s access to the most powerful chips, such as Nvidia’s H100. The latest developments suggest that DeepSeek has either found a way around the rules or that export controls are not the containment measure Washington wanted.
“They can take a really nice large model and use a process called distillation,” said Chetan Puttagunta, general partner at Benchmark. “Basically, you use a very large model to help your smaller model perform better when you want it to It’s getting smart about things. It’s actually very cost-effective.”
Little is known about the laboratory and its founder, Liang Wenfeng. DeepSeek was born out of a Chinese hedge fund called High-Flyer Quant, which manages about $8 billion in assets. media Report.
But DeepSeek isn’t the only Chinese company making progress.
Leading artificial intelligence researcher Kai-Fu Lee explain His startup 01.ai used only $3 million for training. TikTok parent company Byte jumps Wednesday release An update to its model claims to outperform OpenAI’s o1 on key benchmarks.
“Necessity is the mother of invention,” said Perplexity CEO Aravind Srinivas. “Because they had to figure out how to solve it, they actually ended up building something more efficient.”
watch this video learn more.
