appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Stephen Chen

A study from Tsinghua University has opened a new front in the global technology race by asking a question that may have profound implications for industrial powers.
When commanding an artificial intelligence (AI) model to, say, design a better aircraft, does it matter whether you speak to it in English or Chinese?
The answer, according to the research published in China’s top aviation journal Acta Aeronautica et Astronautica Sinica on April 27, is mixed: Chinese may have an intrinsic advantage over English, but not by much – at least for now.
Led by Professor Chen Haixin with Tsinghua’s School of Aerospace Engineering, the team created a sophisticated AI “agent” designed to perform a classic engineering task: reducing drag by tweaking the shape of a modern aircraft wing.
The AI was taught to “see” and “reason”. Using a Vision-Language Model (VLM), the AI was shown images of wing shapes and their airflow patterns, combined with a set of engineering rules and design history.
It then had to propose subtle geometric modifications, such as adding a “bump” here or adjusting a curve there, to make the wing more aerodynamic. The AI learned through a process of trial and error, receiving a “reward” every time it successfully reduced drag.
Chen and his colleagues immediately observed something “noteworthy”.
“Before the aerodynamic task fine-tuning was performed, the Chinese-language instructions initially demonstrated slightly better performance than the English-language instructions,” they wrote.
This suggests that for an untrained model, the engineering context and intuitive phrasing of Chinese offered a potential advantage.
Chinese characters can often convey complex, nuanced engineering concepts, such as the relationship between a “shock wave” and an “upper surface suction peak”, in a more direct, semantically dense way than English.


However, after the intensive reinforcement learning process, where the AI was “fine-tuned” on the specific task of wing design, the performance of both the English and Chinese-speaking agents became almost identical.
Both achieved a drag-reduction performance comparable to a “greedy search” method, which is a non-learning benchmark that represents a near-optimal solution.
“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,” the paper said.
This finding lands in the middle of a fiercely competitive landscape. As the United States and China vie for dominance in AI, the ability to command industrial AI in one’s native tongue is a powerful strategic asset.
English remains the “native language” of many cutting-edge AI models and the global repository of scientific literature.
“English instructions typically have higher expression stability and knowledge coverage in general large models,” Chen’s team wrote, citing some previous studies.
“But Chinese instructions have potential advantages in engineering context expression, semantic intuitiveness and alignment with local models and engineering specifications.”
China has by far the largest population of engineers and scientists in the world, with engineering graduates outnumbering those in the US by more than five to one each year. Most engineers in China use AI design tools in their native language.
Some researchers believe this could create an enormous pool of human-AI talent that will accelerate innovation.


Chinese drug loberamisal extends stroke treatment window from 6 hours to 2 days

China-Russia space camp in Malaysia opens to youth in Southeast Asia

Neurobiologist Ye Bing says his return to China is not spurred by Michigan campus tragedy

China’s stealth fighters, critical minerals from lithium waste: 7 science highlights

Science

Will Chinese become the king language for commanding AI on engineering tasks?

A Tsinghua University study has introduced a new angle to the global tech race. Photo: Shutterstock

China doubles ‘AI for science’ computing scale in 2 months using no US chips

China’s largest scientific AI computing cluster officially entered operation at the national supercomputing network’s core node in Zhengzhou, Henan province, on Tuesday. Photo: CCTV

China launches super-powered AI science system to take on Donald Trump’s Genesis Mission

China’s system is designed to serve thousands of institutions nationwide. Photo: Shutterstock

Lead Microsoft AI scientist Li Hongzhi joins China’s Tongji University

Li Hongzhi has returned to China and is a distinguished tenured professor at Tongji University’s Institute of AI for Engineering. Photo: Handout

Acclaimed biologist Xu Xianzhong returns to China after arrests in his US lab

Inside the Shenzhen Medical Academy of Research and Translation in southern China. Photo: Handout

Xu Xianzhong has returned to China, taking up a full-time position at the Shenzhen Medical Academy of Research and Translation. Photo: Handout

A memorial set up at the University of Michigan for Wang Danhao, a semiconductor researcher from China who died on campus not long after he was questioned by US federal agents. Photo: Linda Wan

EV expert Xu Fuguo leaves US$30 million project in Japan, takes Chinese university post

Xu Fuguo, whose research has focused on EV control engineering, started as a professor in the mechanical engineering school at Dalian University of Technology after spending years researching in the Japanese auto sector. Photo: Xinhua

Xu Fuguo has been appointed professor in the mechanical engineering school at Dalian University of Technology. Photo: Handout

News

China

Study suggests sophisticated AI models can respond better to Chinese than English before learning how to fine-tune geometric modifications


Study suggests sophisticated AI models can respond better to Chinese than English before learning how to fine-tune geometric modifications.


Will Chinese become the king language for commanding AI on engineering tasks?

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Study suggests sophisticated AI models can respond better to Chinese than English before learning how to fine-tune geometric modifications

Study suggests sophisticated AI models can respond better to Chinese than English before learning how to fine-tune geometric modifications