AgenticTuner Uses LLMs to Calibrate High-Dimensional Cosmological Simulations

AgenticTuner replaces black-box Bayesian optimization with an LLM-based agent that proposes parameter trials for cosmological simulations. By incorporating human-readable priors like documentation and heuristics, the agent achieves competitive performance at significantly lower CPU costs.