In mid-February, OpenAI launched its innovative text-to-video generation tool, Sora, intensifying the global race in artificial intelligence. This development raises questions about how quickly China can develop a similar tool and whether AI could emerge as a critical new productive force in the country.
Sora is designed to generate intricate scenes with remarkable accuracy, including multiple characters and specific movements. Zhou Hongyi, founder of 360 Security Technology, noted that Sora could disrupt industries such as advertising, movie trailers, and short videos, potentially accelerating generative AI advancements from a decade to just one or two years.
Despite progress in large-scale AI models in China, Zhou pointed out that there remains an 18-month gap between China’s models and OpenAI’s latest iteration, GPT-4.0. He believes that OpenAI may still have significant innovations, including future models like GPT-5.0.
As of October last year, the Beijing Municipal Science & Technology Commission reported that China had developed at least 254 large language models. However, most of these still lag behind GPT-4.0. The focus for China, according to industry experts, should be on applying these technologies in specific industries to bridge the gap with the U.S.
Jia Jiaya, founder of SmartMore, highlighted that while GPT-4.0 may be less efficient in industrial applications, this presents opportunities for Chinese firms to innovate. His company has developed an industry-specific model designed to enhance efficiency across industrial sectors, asserting that no current large language model serves high-end manufacturing.
China’s comprehensive supply and industrial chains position it uniquely to leverage AI technologies across various applications. However, as experts emphasize, technology must be coupled with commercial viability to be effective.
In recent advancements, Chinese AI startup ModelBest Inc. introduced its lightweight large model, MiniCPM-2B, designed for targeted commercial applications. This model, with 2 billion parameters, offers competitive performance against larger models like OpenAI’s GPT-4.0, which has 1.7 trillion parameters.
Li Dahai, CEO of ModelBest, noted that the new model significantly reduces inference costs, making it more accessible for deployment on mobile devices. The trend towards smaller, efficient models could provide China with an opportunity to leverage AI as a productive tool without excessive resource expenditure.
As the global AI landscape evolves, China appears determined to enhance its capabilities and close the gap with established leaders, setting the stage for potentially transformative applications in various sectors.