CNBC: Chinese AI Company DeepSeek Sparks Panic in Silicon Valley 204

CNBC spent a full 40 minutes reporting on DeepSeek, watch the video.

Title: How China's new AI model DeepSeek threatens U.S. dominance

Full translation:

An AI model released by a little-known Chinese AI company DeepSeek has triggered panic across Silicon Valley, outperforming the best U.S. models despite being built at a much lower cost and on less powerful chips.

DeepSeek (the name of the lab) released a free, open-source large language model in late December, reportedly built in just two months and for less than $6 million, using Nvidia's lower-spec H800 chips.

These new developments have set off alarm bells: whether America's global lead in AI is shrinking, and have raised questions about the massive spending by big tech companies on building AI models and data centers.

In a set of third-party benchmarks, DeepSeek's model outperformed Meta's Llama 3.1, OpenAI's GPT-4o, and Anthropic's Claude Sonnet 3.5 across accuracy ranges from complex problem-solving to math and coding.

On Monday, DeepSeek released R1, a reasoning model that also outperformed OpenAI's latest O1 in many third-party tests.

Microsoft CEO Satya Nadella said in the world: "To see DeepSeek's new model, these open-source models that are doing this reasoning time compute, both very effectively, and very efficiently," at the World Economic Forum in Davos, Switzerland on Wednesday. "We should take the developments out of China very, very seriously."

Because the U.S. government has imposed strict semiconductor restrictions on China, preventing the country from importing the most powerful chips, such as Nvidia's H100. It shows that DeepSeek has either found a way to work around the rules, or overcome Washington's export control barriers.

Benchmark general partner Chetan Puttagunta said: "They can take a very good large model and use a process called distillation." "Basically, you use a very large model to help your smaller model get smart on the things you want it to be smart on. And that's actually very cost-effective."

Little is known about DeepSeek and its founder Liang Wenfeng. According to media reports, DeepSeek was born out of a Chinese hedge fund called High-Flyer Quant, which manages about $8 billion in assets.

But DeepSeek is not the only Chinese company breaking through.

AI researcher Kai-Fu Lee said his startup 01.AI was trained for only $3 million. TikTok's parent company released an update to its model on Wednesday, which claims to outperform OpenAI's O1 on key benchmarks.

"Necessity is the mother of invention," said Perplexity CEO Aravind Srinivas. "Because they had to find a way to work around alternatives, they actually ended up developing more efficient technology."