Former ByteDance Engineer: The AI Gap Between China and the US is Widening, Lacking Distillation Shortcuts and Feedback Loops is the Main Reason

According to monitoring by Dongcha Beating, Zhang Chi made a direct assessment of the AI gap between China and the US in the same interview: “I don’t even agree with the notion that China is catching up; I believe we are still far behind. The gap is widening, which is very unfortunate.” His colleagues and students generally agree, but he also acknowledges that the leadership of listed companies like Zhipu and MiniMax would not agree with this assessment. He attributes the reasons to three aspects. First, distillation shortcuts: he believes many Chinese companies directly use outputs from Claude, GPT, or Gemini as training data, stating, “Claude recently mentioned detecting a large number of distillation attempts; I guess that’s how some companies are taking shortcuts.” However, he also admits that DeepSeek demonstrated real architectural innovation in V3 and R1. Second, the lack of a user feedback loop: US models are user-friendly, leading to more users, and user feedback improves the models; Chinese models start off not being good enough, resulting in fewer users and limited data, creating a vicious cycle. Third, the infrastructure gap: he felt that the infrastructure during his internship at Google was “excellent, with code running very smoothly,” which is a huge difference compared to ByteDance.