(Yicai) Nov. 22 -- The technology gap in artificial intelligence between China and the United States is not as broad as commonly perceived, so Chinese companies would be better off developing specialized AI models and not simply following the path of ChatGPT developer OpenAI, according to the founder of cybersecurity giant 360 Security Technology.

Among the four key elements of AI model development -- computing power, algorithms, software, and data -- Chinese companies mainly lag in computing power due to chip technology constraints, Zhou Hongyi said in an interview with Yicai on the sidelines of the three-day 2024 World Internet Conference Wuzhen Summit that ends today.

"I was excited about general-purpose AI models last year, but now I realize they can only achieve undergraduate-level capabilities rather than specialist expertise," Zhou noted, referring to the approach of OpenAI of building superintelligence through massive computing power, data, energy, and human resources.

The path OpenAI took is not definitive and does not need to be followed by all companies, Zhou said. "Only tech giants like Alibaba Group Holding and Baidu with substantial resources need to pursue this direction."

Training general AI models requires nearly all available public data on the internet, much of which is of poor quality, Zhou pointed out.

This resource-intensive development approach faces mounting challenges as quality data becomes increasingly scarce, potentially affecting commercial returns, Tang Yi, chairman and chief executive of the China arm of computer application firm Evernote, told Yicai.

Chinese companies should leverage China's diverse industrial base and extensive supply chains to develop specialized AI models for vertical sectors, according to Zhou, who cited Google's success with specialized AI applications, including AlphaGo for go gaming, AlphaFold for protein structure analysis, and AlphaChip for chip design.

"This approach not only requires less computing power but also offers practical applications," Zhou noted. "AI model developers should focus on solving real business problems and making AI more accessible, allowing clients to run applications with just one or two graphics cards."

Editors: Tang Shihua, Martin Kadiev