Chinese AI firms make strides to keep pace with industry leaders

BEIJING | Xinhua |  After technological advances in the artificial intelligence (AI) industry marked by large language model (LLM) fever since 2022, Chinese firms in the rapidly advancing sector have cemented their position as major players and are intensifying efforts to narrow the gap with forerunners.

According to a SuperBench assessment report for LLMs by Tsinghua University, among 14 representative models in the world, models like GPT-4 and Claude-3 continue to lead in several capabilities. However, leading Chinese models such as start-up Zhipu AI’s GLM-4 and Baidu’s Ernie Bot 4.0 have shown impressive performance, approaching the top level of their foreign counterparts and gradually closing the gap.

Industry insiders say that China, with its enormous data resources and diverse application scenarios, has gradually established itself as a prominent trendsetter and a second-to-none magnet for related industries amid the trend of exploring new frontiers of AI.

Of the 1,328 AI LLMs globally, 36 percent are from China, the second-largest proportion after the United States, according to a white paper on the global digital economy released by the China Academy of Information and Communications Technology at the Global Digital Economy Conference 2024, which runs from Tuesday to Friday in Beijing.

From China tech giant Baidu’s Ernie to TikTok owner ByteDance’s Doubao, Chinese tech firms have commercially launched a batch of LLMs and are leveraging extensive data to revolutionize AI applications across various platforms.

In May alone, Alibaba Cloud officially released Tongyi Qianwen 2.5 at the start of the month, comparable to GPT-4 Turbo in overall performance, and less than a week later, ByteDance introduced the Doubao family of models, featuring multimodal models.

Xue Lan, dean of Schwarzman College at Tsinghua University, attributes the success of China’s AI sector to widespread societal acceptance of AI, a robust business-led ecosystem, and supportive government policies.

Speaking of China’s advantages and how they benefit other countries, Kang Xi, an assistant professor at Vanderbilt University in the United States, underscores China’s computing power and vast consumer data, which are refining the foundations of AI models and facilitating the creation of tailored models adaptable for other countries’ implementation.

These advantages finally fostered an energetic generative AI landscape in China. A growing number of companies, including startups, tech giants and state-owned enterprises, are rushing to adopt generative AI tools to enhance their productivity and efficiency.

Meanwhile, nearly 30,000 AI companies are operating worldwide as of the first quarter of this year, with approximately 34 percent based in the United States and around 15 percent in China, according to the white paper.

According to a report released by the World Intellectual Property Organization on Wednesday, 54,000 generative AI patent applications were filed globally in the past decade. China filed over 38,000 patents, ranking first in the world.

Among the top 10 patent applicants are Chinese entities including Tencent, Ping An Insurance, Baidu, the Chinese Academy of Sciences, Alibaba and ByteDance, demonstrating that China’s AI sector has made significant strides in keeping up with AI industry leaders.

“When ChatGPT-3.5 was released, we estimated that the technological gap between us and OpenAI would take us one to two years to catch up,” said Yang Mingyuan, an AI developer at Baidu AI Cloud. “However, after just one year, we’ve already narrowed the gap to six months. In some areas, such as Chinese language-processing capabilities, we have even surpassed them.”

Highlighting Baidu’s edge of its full-stack layout in its AI framework, including chips, deep learning frameworks, large models and search applications, Yang noticed rich commercial opportunities in China and the commercialization capabilities of Chinese companies, which will foster positive sector development.

“OpenAI has paved an important path, but we shouldn’t just follow blindly. While general AI advances through extensive models, computer power and data, it often lacks understanding of specific business operations,” said Zhou Hongyi, founder of Chinese internet security firm Qihoo 360.

Zhou noted that many Chinese companies have begun deploying large models on various devices, such as computers, smartphones and cars. Additionally, these companies are working to improve data quality, guiding LLMs toward a more specialized, precise, high-performance and efficient direction. ■