Speaker
Zhimei Liu
(University of Cambridge)
Description
We use discrete diffusion models to learn and generate Type IIB flux vacua across multiple Calabi–Yau geometries. The project was developed with extensive use of LLMs and coding agents for implementation, debugging and experimentation, while scientific decisions and the overall workflow remained human-directed. I will discuss both the physics results and what this experience suggests about the role of LLM-based tools in theoretical physics research.