US-China AI Competition Intensifies: AI Expert Tu Yi-chin Assesses the Gap
The American AI company Anthropic has accused Alibaba of illegally acquiring its Claude AI technology, using model distillation to replicate its capabilities on a large scale. Ethan Tu, founder of the Taiwan AI Labs, evaluated the AI gap and competitive landscape between the U.S. and China, noting that China's political system and ideological constraints may hinder its long-term development.
Video link:
Full text:
According to details from a letter disclosed by Reuters, CNBC, and other U.S. media outlets, the American AI company Anthropic has accused Chinese tech giant Alibaba of illegally acquiring its Claude AI technology, calling it the largest attack of its kind.
Anthropic stated that Alibaba carried out model distillation operations between April and June of this year, which involves using the output of a powerful Claude AI model to train a lower-cost model, thereby effectively replicating its capabilities.
In a letter dated June 10 to Senators Tim Scott and Elizabeth Warren, Anthropic claimed that this operation utilized nearly 25,000 fake accounts to interact with Claude AI more than 28.8 million times. The company pointed to individuals associated with Alibaba and its AI laboratory's artificial intelligence models as being behind the operation.
Ethan Tu, founder of the Taiwan AI Labs, stated that model distillation is common in the industry, but this incident shows that Alibaba is actively catching up with the U.S. "Alibaba wants its models to perform as well as those of U.S. entities, so it learns from Anthropic's models; based on this statement, that is indeed happening. It's a bit like the best American graduates taking the GRE, and Chinese graduates also taking the GRE. To get a better score, the Chinese graduates look at the American graduates' test questions and study their answers. This indicates that both sides are very close in performance and are in a competitive relationship. China may be using methods that the U.S. considers unacceptable to win this academic test competition."
Anthropic warned that this distillation technique could help China narrow the gap in advanced AI capabilities, coming at a time when the U.S. Senate Banking Committee is about to hold hearings on AI issues.
As early as February this year, Anthropic had stated that it had been targeted by Chinese AI companies, including DeepSeek.
Despite U.S. concerns over the rapid development of China's AI, Ethan Tu believes that China's political system may limit the further development of its artificial intelligence. "In terms of freedom, it is easier to succeed. In China, because there are actually many restrictions on the development of artificial intelligence—some of which are quite interesting—it must answer in a way that conforms to the country's ideology and cannot violate these rigid frameworks. In fact, it is very easy for this to happen when very talented people develop to a certain level."
Earlier this month, the U.S. Pentagon released the 1260H list, naming Alibaba among 188 Chinese entities with ties to the Chinese military. Alibaba did not immediately respond to requests for comment regarding these latest allegations. VOA News, Washington.