Junbo Zhao (Jake)

Assistant Professor, Zhejiang University

Director, Ant Group

Bio (2026.7 Version)

Junbo Zhao is an Assistant Professor at Zhejiang University (100-Young Professor Program) and a Director at Ant Group.

Current Focus (2025+): He focuses on three pivotal frontiers:

Scaling of World models/Embodied AI: How can the physical world data, simulation data and human-centric data combined altogether to scale (just like the way of LLM’s pretraining), and can further be deployed in policies. WE DO NET YET HAVE A SCALING LAW here… stay tuned.

Next-Gen Scaling (dLLMs): Spearheading LLaDA 2.X series (Including LLaDA 2.2, Unified model, 2.1 and 2.0), a diffusion language model family that matches AR performance with 5x inference speed minimal improvement by far; and LLaDA 2.0-Uni, a unified multimodal dLLM-MoE integrating understanding and generation, producing interleaving form of output natively.

The “Second Half” of Data: Moving from quantity to density. He explores high-leverage data strategies (synthesis, distillation, and evaluation benchmarks like HeartBench) to unlock efficiency.

Background: In 2024, he led the TableGPT project (TableGPT2 & R1), achieving over 200k downloads and widespread industry adoption in finance and manufacturing. He earned his Ph.D. from NYU under Turing Award laureate Yann LeCun. Previously, he contributed to foundational frameworks (PyTorch, FAISS) at Meta FAIR and developed the first end-to-end self-driving system at NVIDIA (showcased at GTC).

Impact: His work has garnered near 30,000 citations (as of July 2026). He is a recipient of the Zhejiang Province Science and Technology Progress First Prize (2024 & 2023), the QbitAI 2025 AI Person of the Year, the WAIC Yunfan Award, and was featured on the cover of Forbes 30-under-30.