Zhang Yiming: Prefer Large Models to be Temporarily Behind Rather Than Take the Distillation Shortcut

According to Dongcha Beating monitoring, ByteDance founder Zhang Yiming clearly stated at last month's Seed team all-hands meeting that the company will not take distillation as a shortcut to catch up with large models, even if it means temporarily falling behind domestic competitors. He urged the team to be willing to sacrifice short-term gains for long-term goals.

Distillation is a process that enables a new model to learn from a stronger model's output, saving training time and computing power. Internally, ByteDance believes that not taking this approach will make it more difficult for Seed's language model to catch up with domestic peers like DeepSeek, Kimi, and Qianwen in the short term.

However, ByteDance is more concerned about the political risk. The company has long been under scrutiny by the U.S. government due to TikTok. Although TikTok's U.S. data security business has brought in U.S. capital, ByteDance still holds shares and continues to operate TikTok's global commercial activities. If accused of being able to extract U.S. cutting-edge models on a large scale, Washington may once again exert pressure on TikTok.

Anthropic previously accused DeepSeek, Dark Side of the Moon, MiniMax, Zhipu, and Alibaba of being able to extract Claude on a large scale, but did not directly name ByteDance.