Abstract
Generative artificial intelligence (GAI) is the art of harnessing the power of data to model probability distributions. Probability distributions are foundational in various domains, including computer vision and statistical physics. Their accurate characterization is essential for generating high-quality images and predicting complex fluid dynamics. This thesis delves into the interplay between GAI and statistical physics through probability distributions. We explore how the physics of the diffusion process helps the denoising diffusion probabilistic model (DDPM) create better images and an application of AI in disease diagnosis. Moreover, we investigate the methods, limitations, and challenges of using generative models to model non-equilibrium gas flows.