The effectiveness of large model anti-distillation policies is questionable: distillation is just a shortcut for data in independent labs, and restrictions cannot prevent China's AI from catching up.

According to Beating Monitoring, regarding Washington and Anthropic's attempt to cut off China's large model "distillation" channels by blocking cutting-edge models, former GitHub International Strategy Director and Interconnected Capital founder Kevin S. Xu pointed out that adversarial distillation is only a desperate shortcut for some independent Chinese labs suffering from data scarcity. Relying on API blocking cannot fundamentally halt China's overall AI progress.

The named entities DeepSeek, Moonshade, and MiniMax are all independent labs lacking support from a group ecosystem, facing the critical issue of scarce high-quality retraining data such as reasoning steps. In contrast, large labs backed by Alibaba (Qwen), ByteDance (Seed), or Xiaomi possess massive proprietary scenario data comparable to Google and Apple, and do not rely on distillation. Therefore, blocking policies at most cause short-term setbacks for independent labs and cannot shake the foundation of China's major tech companies.

The so-called "data advantage" claimed by outsiders is a misconception: in terms of high-quality knowledge annotation and evaluation data required for training cutting-edge large models, China not only lacks advantages but is also severely deficient in mature commercial data supply chains like Scale AI or Surge. Due to the poor quality of domestic data service providers, independent labs, in despair and out of laziness to take shortcuts, have resorted to API distillation as a cheap data acquisition strategy.

However, the data annotation industry is a low-threshold commercial model issue, not a technical bottleneck like lithography machines. The domestic supply-demand gap can be easily filled. In the long run, student models trained solely by distillation cannot theoretically surpass their teachers. But since large models are still built by human engineers, whether the U.S. forcibly cuts off API channels or not, clever and diligent Chinese developers will ultimately break this upper limit, designing models that surpass their mentors. U.S. sanctions are not only ineffective but may also prematurely cut off the theoretical constraints that could keep Chinese models confined to the "student" ceiling.