Anthropic–Qwen distillation debate turns into a licensing argument

nptacek · x · 2026-07-22

A quote-tweeted thread argues that distillation is acceptable, Anthropic likely would not need to distill from Qwen, and even if it did, Qwen’s license would allow it. It also adds a broader claim that training on broad internet data may produce priors suggesting Chinese-speaking assistants are likely Qwen.

The post is essentially a licensing-and-model-origin debate around Anthropic and Qwen, plus a speculative jab that the real target for Chinese-language assistant priors is Qwen rather than Anthropic.

Original post →

More from Models

Models channel →