After Years of Stagnation, China Is Closing the Gap in the AI Race. It Hasn’t Stopped Developing Advanced Models, but With a Socialist Twist

China has made significant strides in AI systems after trailing the U.S. for two years. Companies like Alibaba are leading this charge, showing how the nation is addressing challenges in data centers and delivering competitive results.

Models that “reason.” Recent developments in China’s AI landscape include models designed to “reason,” similar to OpenAI’s o1 model. DeepSeek was the first local competitor, and now Alibaba has launched QwQ-32B-Preview, a promising new model. According to its developers, QwQ-32B performs exceptionally in mathematics and programming, though language comprehension still requires improvement.

Socialist AI systems. The Chinese government mandates that all AI models must adhere to “fundamental socialist values.” The Cyberspace Administration of China, the country’s internet watchdog, has instructed tech giants like ByteDance, Alibaba, Moonshot, and o1.AI to ensure their systems align with these guidelines.

Machines that reflect government ideology. AI companies in China face government oversight, with officials testing and reviewing their models. These evaluations often include politically sensitive questions, such as those related to President Xi Jinping or events like Tiananmen Square in 1989. AI models are programmed to avoid such topics, deflecting with responses about “sensitive questions.” This mirrors the control China exerts over internet access via the “Great Firewall,” where websites and platforms deemed unsuitable are blocked. Similarly, generative AI models in China must produce content deemed “appropriate” by the government.

Centralization vs. decentralization. China’s approach to AI systems development is highly centralized, with government-directed funding and resources allocated to projects aligned with national objectives. In contrast, AI systems development in the U.S. is decentralized, driven primarily by private companies, although government agencies collaborate with major tech firms on specific projects, such as military applications.

Research and development. Despite facing trade restrictions, including limited access to Nvidia chips, China continues to invest heavily in AI research. A recent study by the Information Technology and Innovation Foundation shows that China now leads in publishing AI research articles. However, the influence of these publications remains below that of U.S. institutions.

Chinese LLMs, like Alibaba’s Qwen-72B, are also rapidly closing the performance gap with Western models. Various benchmarks comparing the performance of these models in standard tests prove this. However, there’s no exceptionally reliable way of knowing whether one model is better. According to Hugging Face CEO Clement Delangue, Qwen-72B ranks among the best. Open-source Chinese models are also advancing rapidly.

Overcoming obstacles. China’s progress in AI models development is notable, especially given the constraints imposed by the U.S.-China trade war. Despite limited access to key resources, Chinese developers and researchers continue to make advancements, demonstrating resilience in the face of these challenges.

Image | Declan Sun (Unsplash)

Related | Nvidia Already Rules the AI Hardware Market. With the Launch of Its Own Open-Source LLM, It’s Now Going After GPT-4 and Llama