China taps US AI models to train homegrown rivals as chip curbs bite
QQQ•Chinese developers are increasingly using output from leading US AI models to train domestic systems, with documents showing the practice has expanded into industrial use. China is encouraging distillation to reduce AI computing costs as US chip export curbs tighten.
1. Distillation expands
A review of more than 300 Chinese academic papers, patents, corporate disclosures, recruitment notices and government documents found that Chinese developers are using US models to generate answers, code and problem-solving examples for training domestic rivals. Companies are also developing software to automate the process, while specialist workers prepare training material.
2. Examples in research
A November 2024 paper described Shanghai Jiao Tong University researchers selecting about 5,000 answers from OpenAI's o1 model, translating them with GPT-4o mini and assembling 10,750 bilingual examples to further train Alibaba's Qwen2.5-72B model. Other papers describe Alibaba researchers using GPT-4 and GPT-4o to generate training material and a project involving Meituan that used GPT-4o to help train a smaller model to coordinate AI agents.
3. US-China tensions
US officials raised concerns about Chinese developers collecting US model output at meetings in Chengdu in July, warning some activity could amount to intellectual-property theft, according to a US official familiar with the discussions. Chinese officials called distillation a standard development technique and said they were open to acting where conduct constituted IP theft, but made no concrete commitments. Chinese government policies encourage distillation for industrial applications, while some developers have sought guidance on the boundaries of permitted use.




