China Closes Ai Gap With This Simple Trick (allegedly)

After Anthropic accused Chinese tech company Alibaba of a “distillation attack” to improve its products, Alibaba this week said it will ban its employees from using Anthropic’s AI tools, the latest chapter in an escalating battle between the US and China for AI supremacy.

Distillation is when a complex AI model is used to train and improve a weaker one—in this case, without permission—and is at the heart of this kerfuffle:

• In a letter to US senators last month, Anthropic accused Alibaba of using 25,000 fraudulent accounts to generate ~28.8 million exchanges with Claude in an attempt to boost its own AI products (using Claude in China violates Anthropic’s terms of service).
• In response, Chinese authorities held meetings with the country’s top AI firms—including Alibaba, ByteDance, and startup Z.ai—to discuss restricting overseas access to their latest and future AI models, per Reuters.

National security issues: In those meetings, authorities were “deeply worried” about Anthropic’s powerful Mythos model exploiting vulnerabilities and becoming a weapon for the US against China. US lawmakers share similar concerns about American tech falling into the hands of geopolitical rivals like China and Russia.

Feeling the pressure?

Unlike the insurmountable obstacle of being six months behind on a TV show, experts say that China is only six months behind the US in AI development, increasing the urgency around distillation. One analyst who spoke to CNBC said that some Chinese models that rival Anthropic and OpenAI can be “60% to 90% cheaper.”

Bottom line: Per the New York Times, some experts believe that cracking down on distillation wouldn’t matter much, and that it alone can’t build the types of powerful models being produced in China.—DL

Become smarter in just 5 minutes. Subscribe to Morning Brew today.