Lu Ziheng

Lu Ziheng is the founder of Kaiwuji and the guest in “你有一把能够挖出金子的铲子，肯定不会先给别人用”｜对谈开物纪陆子恒：用AI发明新材料. The episode describes him as a Nanjing University of Science and Technology undergraduate, Hong Kong University of Science and Technology PhD, former Microsoft Research AI-for-science researcher, and participant in building the materials direction at Beijing’s Zhongguancun Institute.

Lu’s core argument is that AI Materials Discovery will matter only if models can generate or identify material candidates that survive expert judgment, synthesis, validation, scale-up, and customer use. He repeatedly pushes against a pure-tool framing: if a model can discover high-value material IP, the company should use it to build materials first.

Key Claims

• Lu separates materials work into original compound/IP discovery and downstream process scale-up plus commercialization.
• He thinks AI’s largest near-term leverage may be discovering “blockbuster” material IP rather than only optimizing known production processes.
• He treats scalable models, diffusion generation, simulation, and broad property prediction as a route toward useful generalization in materials.
• He argues that hard domain expertise still matters: model teams, simulation specialists, and senior experimental scientists need close daily collaboration.
• His talent view emphasizes hard specialization in one domain, broad vision, agency, taste, belief, and willingness to work on major technical problems.

Connections

• Kaiwuji — company he founded.
• AI Materials Discovery and Materials Pipeline Company — technical and business frames from the episode.
• MatterSim and MatterGen — model examples discussed in his technical route.
• AI For Science, Frontier Model Scaling, and Domain Expert Alignment — broader wiki themes his interview sharpens.