(CLO) Chinese startup DeepSeek is shaking up the global artificial intelligence (AI) community with its impressive new AI model, developed at a remarkably low cost.
However, while many experts acknowledge this achievement, some analysts remain cautious and warn of the risk of overstating DeepSeek's success.
The AI war between the US and China is getting more and more fierce. Illustration photo: AI
DeepSeek, based in Hangzhou, capital of Zhejiang province, shocked the AI industry when it announced its R1 model on January 20. The model is said to be comparable in performance to proprietary models from OpenAI, the company behind ChatGPT, but was developed at a significantly lower training cost.
DeepSeek previously released its V3 large language model, which cost just $5.5 million to train—a modest sum compared to the hundreds of millions of dollars companies like OpenAI and Google typically spend. This has raised concerns that tech companies are wasting money on GPUs (graphics processing units) for AI, leading to a sharp sell-off in shares of Nvidia, a leading AI chip supplier, last week.
Reactions from experts
Sam Altman, CEO of OpenAI, admitted that his company may have “taken the wrong turn” in pursuing a closed-source strategy. In a Reddit Q&A, Altman said OpenAI needed to rethink its approach, but he also stressed that it was not the company’s top priority.
“OpenAI has taken the wrong approach and needs to adjust its open source strategy,” Altman said, but he also stressed that this is not OpenAI’s top priority right now.
Meanwhile, Andrew Ng, former head of Google Brain research and former chief scientist of Baidu, said China is narrowing the gap with the US in the field of AI.
"When ChatGPT launched in November 2022, the US had a large lead over China in generative AI. But over the past two years, that gap has been rapidly narrowing," Andrew Ng shared on the X platform.
He also emphasized that many AI models from China, such as Qwen (Alibaba), Kimi (Moonshot AI) and InternVL (Shanghai AI Lab), are making significant progress, with areas like video creation where China has surpassed the US.
He also warned that if the US continues to restrict open source, China could dominate the AI supply chain, leading to more businesses using Chinese models than American ones.
Morgan Stanley analyst Shawn Kim said major US tech companies are actively supporting DeepSeek.
Nvidia has integrated the R1 model into its NIM microservice, while OpenAI investor Microsoft also supports R1 on its Azure and GitHub cloud computing platforms. Amazon also allows customers to use R1 through its Amazon Web Services (AWS) service.
This support shows that DeepSeek can become a formidable competitor in the AI market, not only in China but also globally.
Skepticism about DeepSeek's future
Despite the praise, some experts remain skeptical about DeepSeek's true impact on the AI industry.
DeepSeek founder and CEO Liang Wenfeng. Photo: Weibo
Yann LeCun, head of AI at Meta (Facebook), disputes the idea that DeepSeek is a sign that China is surpassing the US in AI.
"The correct understanding is: open source AI models are gradually surpassing proprietary models, not China surpassing the US in AI," LeCun shared on the Threads platform.
Additionally, many experts have questioned DeepSeek's true cost. Professor Zheng Xiaoqing from Fudan University said that DeepSeek's report on V3 training costs did not include previous research investments.
"DeepSeek achieved success through technical optimization, but this does not mean it will change the landscape of the AI chip industry," Zheng said.
Cao Phong (according to SCMP, Reuters, NYT)
Source: https://www.congluan.vn/deepseek-buoc-ngoat-ai-hay-chi-la-cuong-dieu-post332892.html
Comment (0)