ByteDance to Persist with Self-Developed LLMs and Restructure Enterprise AI Teams
ByteDance CEO Liang Rubo announced that the company will persist with self-developed large language models (LLMs) and accept temporary lags behind competitors instead of relying on AI distillation. Additionally, the company is integrating its Doubao, Feishu, and Volcano Engine teams to form a new "Creativity Service Platform" to enhance B-end enterprise synergy. This strategy highlights ByteDance's commitment to building foundational AI capabilities from scratch, avoiding shortcuts like knowledge distillation which could pose regulatory risks amid US-China tensions. The organizational restructuring also signals a major push to monetize AI by deeply integrating it into enterprise productivity and cloud services. Founder Zhang Yiming's decision to avoid AI distillation was reportedly influenced by ByteDance's complex relationship with the US government over TikTok. Under the new structure, the Feishu product team will merge into the Doubao team, while the Feishu GTM team will merge with Volcano Engine under the leadership of Tan Dai.
## BACKGROUND
Knowledge distillation is a machine learning process where a smaller, faster model is trained to reproduce the behavior and knowledge of a larger, more complex model. While it speeds up development, using proprietary foreign models (like OpenAI's GPT-4) to distill local models has faced scrutiny and licensing restrictions. ByteDance's Volcano Engine provides cloud and Model-as-a-Service (MaaS) solutions, while Feishu is an enterprise collaboration platform similar to Slack or Microsoft Teams.