At the recent World Artificial Intelligence Conference (WAIC), Baidu CEO Robin Li emphasized a pressing issue within China’s artificial intelligence landscape: an excess of language models coupled with a deficit of practical applications. He observed that the surge in Chinese language models has led to “intense competition,” as reported by the South China Morning Post.
In the aftermath of the emergence of ChatGPT, many major Chinese technology firms raced to develop their own AI language models, largely due to the restrictions placed on OpenAI's technology in China. Baidu, under Li’s leadership, developed the ERNIE model, which they assert rivals the capabilities of ChatGPT. Beyond just the foundational model, Baidu has made ERNIE available as a chatbot tailored for enterprise use, accessible through the Qianfan Foundation model platform.
Li highlighted that the competitive landscape is populated by over 100 AI models striving to capture portions of the Chinese market. Notable contenders include Alibaba’s Qwen, Zhipu’s Qingyan, and Baichuan from Tencent’s subsidiary, Baichuan Intelligence. Despite this abundance, Li expressed concern that developers have concentrated more on outperforming their competitors rather than creating meaningful AI applications. “I’ve noticed that many still focus primarily on foundational models,” Li stated. “But I ask: what about real-world applications? Who stands to benefit from them?”
To attract more users, local tech giants like Alibaba, Tencent, and ByteDance (the parent company of TikTok) have slashed operational costs associated with their models. However, Li pointed out that the saturation of models has squandered substantial resources, particularly computational power. In contrast to American firms engaged in developing generative AI systems, Chinese companies face significant hurdles due to constraints on computing resources, exacerbated by recent U.S. export restrictions on powerful chips from manufacturers like Nvidia.
Li’s insights echo remarks he made during a previous forum at Shenzhen University, where he warned that China possesses “too many foundational models and very few AI-native applications.” This sentiment was shared by Xu Li, CEO and co-founder of SenseTime, who joined Li on the WAIC panel. Xu noted that the future trajectory of the AI sector hinges on the development of applications that can incite significant change across various industries.
In addition to traditional model offerings, SenseTime has expanded its portfolio to include services such as SenseChat, a conversational AI, and platforms for generating images. At WAIC, the company introduced Vimi, an innovative AI avatar video generator capable of transforming still images into dynamic short clips.
Xu added that the AI industry has yet to reach a pivotal moment, claiming that it has not penetrated vertical markets in ways that prompt widespread transformation. The ongoing discussions among industry leaders emphasize the need to pivot from mere language model development to the creation of applications that deliver tangible benefits and drive the practical implementation of AI technologies across diverse sectors.