At the World Artificial Intelligence Conference (WAIC), Baidu CEO Robin Li highlighted a pressing issue within China's AI landscape: an oversaturation of language models without sufficient practical uses. According to Li, the proliferation of Chinese language models has sparked intense competition among tech giants, a trend exacerbated by restrictions on OpenAI’s technology in China.
Major players like Baidu rushed to develop their own models following the success of OpenAI's ChatGPT. Baidu introduced ERNIE, positioning it as a comparable alternative. Despite this, Li emphasized a critical imbalance: while numerous models vie for dominance, real-world applications remain scarce.
Beyond Baidu's efforts, Li noted over 100 competing models in China, including Alibaba's Qwen, Zhipu's Qingyan, and Baichuan from Tencent's subsidiary. However, Li expressed concern that developers prioritize model superiority over practical implementation.
"The focus on foundational models persists," Li stated at WAIC, "but where are the tangible benefits for real-world applications?"
To mitigate this surplus, local vendors like Alibaba and Tencent have reduced operational costs for their models. Li underscored the wastage of significant resources, particularly in computing power, due to the proliferation of models.
In contrast to American counterparts unhindered by computational constraints, Chinese firms face limitations exacerbated by restrictions on high-performance chip exports.
Li's observations at WAIC echoed sentiments expressed previously at the 2023 X-Lake Forum, where he criticized China's surplus of foundational models relative to practical AI applications.
Xu Li, CEO of SenseTime, echoed these concerns, emphasizing that the critical test for AI's advancement lies in its practical applications across industries.
SenseTime, despite facing U.S. sanctions, unveiled Vimi at WAIC—a controllable AI avatar video generator aimed at revolutionizing short video creation from static images.
Xu concluded that AI's transformative potential hinges on its application across vertical industries, an area where significant strides have yet to be made.
