Researchers have developed a method to use aggregate mobility statistics to inform language models used in crowd simulation. The approach involves fine-tuning a model to match observed destination compositions and reducing errors by resampling training data. This technique was tested on mobile network counts from two baseball games, resulting in improved accuracy without requiring inference-time correction. --- Why it matters: This work matters because it provides a way for pedestrian simulators to generate realistic crowd behavior using limited data, which is often a challenge due to privacy concerns. By fine-tuning language models with aggregate statistics, researchers can improve the accuracy of post-event crowd simulation. Source: https://arxiv.org/abs/2608.19778