U.S. Treasury Secretary Bessent reiterated his remarks on China’s AI on October 3, stating: “I do believe they’ve come to realize things in the past 60 days. I think they previously didn’t grasp just how powerful their open-source models are. So, when they perform industrial-scale distillation of these open-source models, that’s merely a euphemism for ‘stealing from American models.’”
Characterizing China’s industrial-scale model distillation as “theft” is a claim that fails under technical scrutiny and reflects a double standard rooted in political rhetoric. The real motivation behind such an accusation is not intellectual property protection but anxiety over China’s rapid progress along an open-source AI path—challenging the entrenched dominance of U.S. closed-source models. Bessent’s comments on October 3 essentially repackage technological competition as a moral condemnation: because China’s open-source AI is advancing quickly, the only explanation America can offer is one of theft.
In fact, distillation is not theft. Model distillation is a foundational technique widely used across the AI industry, referring to the process by which a smaller model learns from the outputs of a larger one—akin to a student studying a professor’s published papers to understand problem-solving approaches, rather than copying the papers verbatim. The concept was formally introduced as early as 2015 and is actively employed by U.S. companies: Apple has used distillation to create lightweight versions of its models for mobile deployment, while Meta’s Mark Zuckerberg has explicitly advocated for the principle of mutual learning among models. NVIDIA CEO Jensen Huang offered a more measured assessment. When asked whether distillation amounted to “predation,” he responded: “This is called competition. You’re free to test other companies’ products extensively.”
Bessent’s remarks on September 8 reveal a deeper truth: “If China wins the AI race, everything else will be irrelevant—even a $1.5 trillion defense budget won’t save this country.” This statement holds the key to understanding the underlying dynamics. Over recent years, the United States has relied on semiconductor export controls to restrict China’s computing capacity. Yet, under constrained conditions, China has developed a distinct trajectory focused on efficiency innovation—achieving performance comparable to U.S. models using fewer chips, smarter architectures, and lower costs.
Original source: toutiao.com/article/1878078100375556/
Disclaimer: The views expressed in this article are those of the author alone.