A new AI assistant from China is getting people talking in Silicon Valley

It has sparked a heated debate in American tech circles: How could a small Chinese company so dramatically outperform the AI ​​industry’s best-funded players? And what does this mean for the future of the field?

Meta’s chief AI scientist Yann LeCun wrote in a Threads post that this development does not mean China is “outperforming the US in AI,” but rather serves as evidence that “open source models are outperforming proprietary models.” He added that DeepSeek has benefited from other open-weight models, including some metas.

“They developed new ideas and built them on the work of others. Because their work is published and open source, everyone can benefit from it,” LeCun wrote. “That is the power of open research and open source.”

(Although many companies, including DeepSeek and Meta, claim their AI models are open source, they have not actually made their training data available to the public.)

OpenAI CEO Sam Altman also appeared to mock DeepSeek last month after some users noticed that V3 was occasionally confused with ChatGPT. A day after V3’s release, Altman wrote on I don’t know if it will work.”

Some figures circulated unsubstantiated claims online that DeepSeek’s success was a “psyop,” or psychological operation, by the Chinese government, raising suspicions about the small team’s ability to “beat all the world’s top researchers as a side project.”

Soumith Chintala, co-founder of PyTorch, the machine learning library developed by Meta AI, was one of many to deny these allegations this weekend.

“I’m strangely impressed that people are keeping up with Deepseek by spreading bizarre conspiracy theories, despite Deepseek being open source and writing some of the most detailed articles ever,” Chintala wrote on X. “Read. replicate. compete. Don’t be salty, it will just make you look incompetent.”

Others in the technology and investment industries echoed the praise and expressed excitement about the impact of DeepSeek’s success.

“That’s what makes the DeepSeek thing so fun. “A bunch of scammers have been selling AI secret sauce for years – creepy mystery juice that has never been fully explained,” wrote macroeconomist Philip Pilkington on Tent burned down.”

Nat Friedman, the former CEO of Github, wrote similarly: “The Deepseek team is obviously really good. China is full of talented engineers. Every second take can be managed. Excuse me.”

DeepSeek’s models feature bilingual proficiency and excel in both Chinese and English. However, they appear to be subject to censorship or certain political leanings on topics considered sensitive in China.

When asked about the sovereignty of Taiwan, a self-governing island democracy that Beijing claims as its territory, DeepSeek’s R1 sometimes states that the topic is “beyond my current horizon.” Elsewhere, the model describes Taiwan as “an inalienable part of Chinese territory” and adds: “We firmly reject any form of separatist activities for ‘Taiwan independence’ and are determined to achieve the complete reunification of the motherland by peaceful means.”

Hot on the heels of DeepSeek’s latest models, other players in China’s tech sector are already launching new competitors in the race for AI dominance.

Alibaba on Sunday introduced its latest Qwen2.5-1M model, an upgrade from Qwen2.5-72B.

Kimi AI, owned by Beijing-based Moonshot AI, also announced on Saturday the launch of its latest multimodal reasoning model Kimi k1.5, touted as comparable to OpenAI’s o1.