} ?>
(Yicai) Feb. 12 -- “Innovation can't be planned,” Baidu founder Robin Li has said regarding the meteoric rise of Chinese artificial intelligence startup DeepSeek.
“You don't know when and where the innovation comes,” Li said yesterday at the World Governments Summit in Dubai. “What you can do is just to foster an environment that is conducive to innovation.”
DeepSeek, an open-source AI model, rattled stock markets late last month, when its Hangzhou-based developer of the same name released the latest iteration. Comparable to major large language models such as OpenAI's ChatGPT, but reportedly at much lower cost, the release of DeepSeek-R1 set off alarm bells about the massive sums being spent on AI training and infrastructure.
Inference costs for foundation models have plunged more than 90 percent over the past 12 months, Li said. "If you can reduce the cost by a certain percentage, then that means your productivity increases by that kind of percentage.
"That's pretty much the nature of innovation,” Li said in conversation with H.E. Omar Sultan Al Olama, the United Arab Emirates’ minister for AI. “Just today it's happening at a rate much faster than before.”
“And companies in China are under certain type of constraint. So we were probably the first ones to feel the pain of those high costs," he added, referencing curbs on Chinese access to chips as a result of geopolitics.
Beijing-based Baidu was the first Chinese company to launch a ChatGPT-like model in March 2023. Ernie Bot's monthly active users have topped 12 million, behind Doubao, which belongs to TikTok owner ByteDance, and MoonShot AI's Kimi, according to data research firm QuestMobile.
Despite the efficiency gains, infrastructure investment remains important, Li pointed out. “When technology still evolves at such a rapid rate, you just cannot stop investing. You have to invest to make sure that you are at the very front of this technological innovation or revolution.”
Regarding the ongoing debate between the open-source and closed-source approaches to AI development, Li drew from recent experiences and said that open-source models such as DeepSeek can hasten technology adoption and innovation.
"What they learned in the past few months is that open source can help you get more attention," Li noted. “A lot of people will be curious enough to try it, which helps spread the technology much faster.”
But the key challenge lies not in the choice of language models, but in creating practical value through applications, Li pointed out. While chatbots are popular, their user engagement metrics still lag behind traditional social media platforms, offering much room for innovation in AI applications, he noted.
Editor: Martin Kadiev