Chinese companies are using responses generated by U.S. artificial intelligence models to train their own systems, which are cheaper and easier to integrate into industry, according to a Reuters investigation published amid intensifying technological rivalry between Beijing and Washington.
The practice, known as “distillation,” involves using a high-performance model to generate responses, code or solutions to complex problems. These results are then turned into training data for smaller models, which can reproduce some of the original system’s capabilities.
The investigation, based on an analysis of more than 300 scientific papers, patents, government documents and company materials, shows that the method is increasingly being used in China’s industry. In 2024, researchers at Shanghai Jiao Tong University selected approximately 5,000 responses generated by OpenAI’s o1 model and turned them into more than 10,000 bilingual examples to train Alibaba’s Qwen2.5-72B model.
In April 2025, Alibaba researchers used GPT-4 and GPT-4o to produce training data in several languages. In February 2026, university researchers and Meituan employees presented a GPT-4o-based method for training models capable of coordinating AI agents in programming.
Distillation is not illegal in itself, but OpenAI and Anthropic restrict the use of their models’ outputs to develop competing systems. U.S. officials raised the issue in talks with Chinese authorities in Chengdu in July. Beijing maintains that the technique is common practice and promises to intervene in cases of proven theft.
The dispute comes after the United States announced new restrictions on China’s access to advanced chips, including products from Nvidia. At the same time, Beijing is supporting the development of smaller models that require less computing power and can be used in factories. The differences could be discussed by Donald Trump and Xi Jinping at the APEC summit in Shenzhen, scheduled for November 18-19.
بشكلassistant to=event_translation code_久久爱 (json.) 久久精品 bitstarz({Sources
Latest News
08:53
08:50
08:40
08:33
08:24
See more news