Quanquan Gu, an associate professor of computer science at UCLA and one of ByteDance’s most prominent AI researchers, announced his departure from ByteDance Seed on June 2, according to South China Morning Post. Gu joined the company in 2023 and spent his first roughly eighteen months building ByteDance’s AI-for-science stack: SeedFold, the first biomolecular structure prediction model to surpass AlphaFold 3 across a broad range of FoldBench benchmarks; SeedProteo, a de novo protein binder design model that outperformed AlphaProteo, RFDiffusion, Chai-2, BinderCraft, and BoltzGen; and the DPLM protein language model series. In early 2025, following DeepSeek’s breakout moment, he pivoted to co-found ByteDance’s LLM optimization and scaling team, building the pretraining infrastructure that powered Seed 2.0. In his farewell post on X, Gu expressed gratitude to teammates and leadership “for an incredibly rewarding journey” and signed off: “The best model is yet to come. Scaling continues.” His next destination was not disclosed.
The departure comes as ByteDance shifts emphasis from foundational research toward monetization, including subscription plans for its consumer AI app Doubao — which integrates Seed 2.0 and has reached roughly 336 million monthly active users. SCMP described the exit as “unexpected.” At UCLA, where he leads the Artificial General Intelligence Lab and holds an associate professorship on leave, Gu has published influential work on deep learning theory, reinforcement learning, and language model alignment, including the Self-Play Fine-Tuning (SPIN) method. His ability to simultaneously advance biological AI and frontier LLM pretraining — disciplines that rarely overlap in a single researcher — was widely cited as what made him distinctive.