The global race to develop the most powerful AI models is intensifying, and China has found a notable way to keep pace with its American rivals.
Rather than relying entirely on expensive electronic chips, or developing every technology from scratch, Chinese developers are using the answers produced by advanced American AI models to improve their own systems.
According to a new investigation by Reuters, which reviewed more than 300 research papers, patents, business records and government documents, the practice has become more widespread across China’s AI sector.
The development comes in a market already marked by fierce price competition between Chinese AI companies.
The technique is known as “AI distillation”. It is not new or illegal in itself, but the scale of its growing use by Chinese companies, according to the reports, has raised concerns among officials in Washington.
How China turns American AI into a competitive advantage
Distillation can be compared to a student learning from a more experienced teacher.
In this process, an advanced AI model generates answers, writes computer code or solves complex problems. Developers then use these outputs to train smaller models, so that they can carry out similar tasks without needing the same computing power.
This method is especially important for Chinese companies facing American restrictions on access to advanced electronic chips.
Although some companies, such as ByteDance and Tencent, have managed to obtain shipments of Nvidia’s H200 AI chips, access to the latest American equipment remains restricted.
The use of this technique is not merely theoretical. Researchers at Shanghai Jiao Tong University have previously used answers produced by OpenAI’s o1 model to create thousands of training examples for Alibaba’s Qwen AI model.
Other research projects involving Alibaba and Meituan have explored similar techniques using GPT-4 and GPT-4o.
Notably, these efforts are no longer limited to universities and academic institutions.
The Reuters investigation found evidence that companies are developing automated tools to collect the answers produced by AI systems, and are employing specialised workers to prepare and organise this data for training.
The Chinese government also encourages the wider use of distillation, with the aim of reducing the costs of AI systems and making them easier to deploy, particularly in the industrial sector.
However, distillation cannot automatically turn a small model into an exact copy of a more advanced one.
It may help transfer some capabilities from large models to smaller ones, but developing competitive systems still requires advanced technical expertise, large computing resources and original research.
Why this trend worries Washington
The growing spread of AI distillation poses a complex challenge for the United States, which has spent years imposing restrictions on China’s access to the advanced electronic chips used to develop AI systems.
Tensions between the two countries have also extended into other areas. The United States recently banned the use of DeepSeek models within civilian federal agencies, including third-party programs that rely on the Chinese company’s models.
If Chinese developers can improve their models using the outputs of American AI systems, export restrictions on chips may prove less effective than Washington hopes.
The issue of intellectual property rights also arises here. American companies, including OpenAI and Anthropic, restrict the use of their models’ outputs to train competing systems without appropriate permission.
American officials have previously expressed concern about large-scale distillation, warning that activities involving the theft of intellectual property could lead to sanctions.
Chinese officials, for their part, have defended distillation as a legitimate method of developing AI systems, while acknowledging that actual violations of intellectual property rights should be subject to legal action.
The timing of these developments is especially significant, with a meeting expected between US President Donald Trump and his Chinese counterpart Xi Jinping during the Asia-Pacific Economic Cooperation (APEC) summit, scheduled for 18 and 19 November in the Chinese city of Shenzhen.
For now, distillation remains a legitimate and widely used method of training AI models, but the way companies obtain training data and how it is used have become increasingly contested.
As Chinese AI models continue to develop, a more important question arises: will American restrictions on exports of advanced electronic chips be enough to preserve the technological lead of the United States?
Source
اضافة اعلان