Tsinghua University's campus in China. Photo courtesy of Tsinghua

In their study, published April 27 in China’s aviation journal Acta Aeronautica et Astronautica Sinica, researchers from Tsinghua University’s School of Aerospace Engineering, led by Professor Chen Haixin, developed an AI agent to reduce drag by modifying aircraft wing shapes.

The team trained the AI by testing different wing shape adjustments and rewarding the system when drag was reduced.

Before undergoing aerodynamic fine-tuning, the team found that Chinese-language instructions produced slightly better results than English prompts.

"Before the aerodynamic task fine-tuning was performed, the Chinese-language instructions initially demonstrated slightly better performance than the English-language instructions," the researchers wrote in their study, as quoted by the South China Morning Post.

Chinese AI chatbot DeepSeek on a smartphone. Photo from Pexels

They noted that Chinese can sometimes describe complex engineering relationships more directly and compactly than English.

However, the performance gap largely disappeared after the AI underwent intensive task-specific reinforcement learning.

"The key conclusion is that the fine-tuning process for the aerodynamic design task has a significantly stronger effect on the agent’s performance than the language of the prompt itself," they said.

They added, citing previous studies, that English remains the dominant language for many leading AI models and scientific publications because English instructions typically provide more stable expression and broader knowledge coverage in general AI models.

"But Chinese instructions have potential advantages in engineering context expression, semantic intuitiveness and alignment with local models and engineering specifications."

Tsinghua University was ranked the world’s top institution for AI in the 2026 CSRankings released in January. CSRankings measures universities based on research output in leading computer science and AI journals rather than reputation surveys used by rankings such as U.S. News & World Report.