ByteDance, the parent company of TikTok, has stopped using knowledge distillation in its artificial intelligence development. The company will instead rely entirely on data generated internally, a move designed to strengthen its competitive position and reduce exposure to geopolitical risks tied to the US-China tech rivalry.
Why the technique was dropped
Knowledge distillation is a method where a smaller AI model learns from a larger, often more capable one. It's widely used to speed up development and improve performance without starting from scratch. ByteDance's decision to ban the practice means its AI teams can no longer borrow capabilities from external models — they must build everything from their own data.
The company sees this as a way to protect its technology stack from potential restrictions or dependencies that could arise amid escalating tensions between Washington and Beijing. By cutting off reliance on outside models, ByteDance aims to keep its AI development self-contained and less vulnerable to sanctions or export controls.
Shifting to homegrown data is a significant change. It could slow down some projects initially, as teams adjust to working without the shortcut that knowledge distillation provides. But ByteDance believes the long-term payoff is worth it — a more independent AI pipeline that can compete globally without being tied to foreign technology.
The company has not said which specific AI products or services will be affected. ByteDance operates a range of AI-driven platforms, including TikTok's recommendation engine, its Douyin app in China, and various enterprise tools. The ban likely applies across all of them.
The geopolitical backdrop
ByteDance's move comes as the US and China tighten controls on advanced technology. Washington has restricted exports of AI chips and software to Chinese companies, while Beijing has pushed domestic firms to reduce reliance on foreign tech. ByteDance itself has faced scrutiny over TikTok's data security, with some US lawmakers calling for a ban.
By going its own way on AI data, ByteDance is trying to sidestep those pressures. The strategy mirrors a broader trend among Chinese tech giants — Tencent and Alibaba have also invested heavily in proprietary AI models and datasets.
ByteDance hasn't detailed how it will replace knowledge distillation. The company may need to invest more in data collection and model training from scratch. Whether this shift will slow ByteDance's AI progress or give it an edge in the long run remains an open question.
