Chinese developers vaunt their intelligence at technology exhibition

OpenAI's GPT-4, developed in the US, remains the benchmark artificial intelligence model, but that isn't stopping Chinese developers from exploring creative new ways to harness the technology.

The World Artificial Intelligence Conference (WAIC), held last week in Shanghai, provided a platform to showcase progress in the domestic sector and how China may be able to circumvent constraints imposed by the US government.

China and the US are in a race to chase the Holy Grail of artificial intelligence. It's a technology that promises a digital revolution shaping the future of mankind. So far, the US is outpacing the competition with the debut of ChatGPT, but Chinese innovators are catching up rapidly.

Over 30 large language AI models were unveiled at the Shanghai AI event this year. Some are compatible with GPT-4 from OpenAI, while others were focused on practical applications across various industries, like finance, retail and energy.

Upgraded and new AI models

During the global conference, SenseTime released its upgraded AI model SenseNova 5o, with a 30 percent performance improvement over the previous version. The model's abilities in mathematics, English, command response and other core indices are now considered "on par" with GPT-4o, which is regarded as the industry's best AI model.

SenseTime Chief Executive Xu Li highlighted companion indices of AI models, including SenseNova, GPT-4 and Google's Gemini, to prove the powers of the company's AI models. Investors seem impressed. Hong Kong-listed SenseTime shares have doubled in the past three months.

MiniMax, a Shanghai-based developer of large language models, said the gap between domestic models and GPT-4 is narrowing. The general accuracy rate of GPT-4 is about 60-70 percent, while domestic models like MiniMax come in at close 60 percent.

Yan Junjie, founder of MiniMax, said he expects only five AI models to dominate the global market in the future. Since there are only two, maybe three, US firms at the top of the market – like Open AI and Google – there is room for Chinese AI models to emerge as top contenders.

Besides fast-growing AI stars like SenseTime and MiniMax, Chinese Internet giants Alibaba, Baidu and Tencent all showcased new and upgraded AI features at the global conference.

Chinese progress matters because OpenAI is planning to restrict access to its application programming interface on the Chinese mainland. The interface is a platform that allows developers of other products to integrate with its AI models, beginning next month, according to media reports. The move to snub China is seen as part of US strategy to impair China's technology development, but it may well prove to be a springboard for domestic AI development.

At the conference, SenseTime also announced a "zero cost" plan to help Chinese developers transfer from OpenAI to SenseTime AI models.

Beyond AI models: Applications

With huge amounts of data and usage scenarios, many Chinese developers are concentrating their resources on vertical applications with AI integration, moving into realms such as predicting electricity usage, managing driverless cars and drones, and combating financial fraud.

Baidu Chairman Robin Li urged the AI industry to shift its focus away from the development of large language models and instead prioritize AI adaptation and applications.

"Without applications, having only AI models is meaningless," Li told a conference forum.

4Paradigm President Hu Shiwei, echoing Li's opinion, said the priority is to integrate AI into "hundreds of industries."

His company showcased AI applications that improve the accuracy of financial fraud identification, increase sales in the retail industry and lift production capacity in manufacturing through "smart" scheduling.

One spotlight at its exposition booth was an AI model for hydropower. It encompasses power generation calculations, weather forecasting, and flood prevention and control. In terms of flood forecasting, it has been reduced from about three hours in the past to five or six minutes today with AI, making the technology impact "millions of people and billions of yuan," a company staffer told Shanghai Daily.

Shanghai-based StarBitech showcased its Media AI system that the company predicts will let media professionals finish an hour early every day with powerful insights and streamlined workflows.

The system covers AI models for data management, multi-modal content creation, AI proofreading, intelligent publishing, and public opinion analysis, boosting efficiency, accuracy, and engagement in publishing, journalism, and broadcasting.

Infinigence AI released a new infrastructure platform, akin to a mid-level software layer, to connect different brands of AI chips and models. This approach addresses the challenge of limited access to advanced chips by enabling hybrid AI infrastructure utilizing AMD, Huawei and Nvidia technologies.

That's an important development amid US sanctions that limit access to advanced chips like Nvidia's. Chinese organizations often find it difficult to procure enough training cards with Nvidia chips.