Academia Sinica Logo

AI can be used to accelerate and even replace supercomputer simulations in astrophysics, turning complex models into easily accessible tools. I will introduce recent and ongoing work aimed at cosmology, including hybrid AI-physics codes, replacements of hydrodynamics solvers with graph neural networks, and ways to interface AI simulations with Large Language Models in multi-agent systems. I will show how such a system of agents can autonomously set up and run a complete hundred-million-particle three dimensional cosmological simulation in seconds. As a use case of this type of modelling, I will explore the Lyman-alpha forest of intergalactic absorption in high redshift quasars and how it can both be gravitationally lensed, and also used to make large-scale maps of the intergalactic radiation intensity. I will also briefly mention related simulation predictions for the "little red dot" galaxies seen by JWST at high redshift.