"What? Not a Genius?"... China's AI That Shocked the World Faces Copycat Accusations [Tech Talk]

Chinese AI Growth Accelerates as 'Distillation Controversy' Intensifies
Replicating and Compressing Large Models into Smaller Ones
"U.S. Computers Powering China's AI": Growing Discontent

Editor's NoteThis series explains in simple terms the technologies that are essential to our daily lives yet often unfamiliar, from AI to semiconductors, telecommunications, and biotech.

China's artificial intelligence company Moonshot AI has developed an open-source large language model (LLM) called Kimi K3. The company shocked the industry by launching a model that delivers comparable performance to those from OpenAI and Anthropic, but at a significantly lower cost. This has been compared to the so-called "DeepSeek Shock" of the past.

Yang Zelin, founder of Moonshot AI, a Chinese open-source AI development company. Screenshot from Xiaohongshu

This time, however, the reaction from the United States has turned markedly more aggressive. Not only companies, but even government officials have raised suspicions about "distillation attacks" by Chinese AI firms. Distillation, a replication and compression technique used by AI researchers for many years, has now become a focal point of controversy as the technological rivalry between the US and China intensifies.

Did Chinese AI models copy those from the US?

On the 17th (local time), visitors are seen touring the booth of the Chinese AI startup Moonshot at the World Artificial Intelligence Conference (WAIC) held in Shanghai, China. Photo by Yonhap News Agency

On July 21 (local time), US Treasury Secretary Scott Bessent claimed in an interview with Fox Business, "We have found the watermark of US LLMs in many of China's AI models." Secretary Bessent stressed, "We will closely examine this issue in the coming days or weeks. If foreign models steal technology from our great companies, we will sanction them."

It is not only the government taking an aggressive stance. Anthropic, in a letter submitted to the US Senate Committee on Banking last month, raised the possibility that Alibaba had "distilled" Anthropic's model. Anthropic condemned this as "the largest distillation attack ever."

Distillation: A technique used in AI research for decades

'Intelligent distillation' is a development technique that has been used in the AI industry for decades. It improves performance and reduces size by training a smaller model on the outputs of a large model. The term is derived from mainstream distillators. Pixabay

Distillation is a method of AI model replication and compression. It involves using a pre-trained large model as the "teacher" and training another "student" model to emulate the teacher. Typically, it works by collecting a large amount of responses from the teacher AI, thereby generalizing intelligence. The resulting student model is much smaller than the teacher but can achieve similar performance.

AI researchers have been using distillation techniques for decades to replicate and compress each other's models and advance the technology. This practice has further evolved through the research of leading figures in modern AI such as Juergen Schmidhuber and Geoffrey Hinton. In the open-source sector, where all information about models is released and shared, distillation has become a core component of development.

"US computers fueling China's AI victory": Complaints from US AI companies

However, as the performance of open-source models has risen sharply, cracks have begun to form between US and Chinese companies. Google and OpenAI, for example, raised suspicions as early as February that Chinese companies might have secretly distilled their models. In May, Anthropic posted on its blog that a "large-scale distillation attack from China" was underway, alleging that Chinese AI research institutes were able to develop models on par with the US due to massive technology leaks that illegally siphoned off innovation from American firms.

The reason US AI companies are wary of distillation is clear: if advanced American models—trained with astronomical investment—are immediately distilled by Chinese startups, the incentive to lead innovation at the cutting edge diminishes. Anthropic put it bluntly: "The AI victory of authoritarian governments is being fueled by the computing resources of America."

Distilled cost-effective AI models also threaten the business outlook for US companies. Last month, Microsoft announced it would consider introducing DeepSeek's open-source model as a low-cost service. If cost-effective models come to dominate the AI market, leading AI companies will see their pricing power steadily eroded.

"The US copied distillation research too": A skeptical perspective

Attitudes toward distillation within the industry are mixed. While America’s major AI companies oppose large-scale distillation attacks, some warn that the AI industry could not have advanced without distillation.

Juergen Schmidhuber, who opened the first chapter on AI distillation techniques with a paper published in 1991, stated on his official X (formerly Twitter) account on July 23, "I support the free distillation of AI by private companies. In 1991, I made my research on distillation freely available in Europe. Afterwards, this research was replicated in the US and China." He pointed out that the American AI industry itself was able to advance thanks to distillation.

This content was produced with the assistance of AI translation services.