Abstract: In this talk, I will cover three new (and orthogonal) ideas in modern ML -- transformers, diffusion models, and differentiable simulation. Transformers are modern neural network architectures, which enabled recent breakthroughs in various AI fields. Diffusion models and differentiable simulations are for data generation and inverse problems. I will focus on sharing high-level ideas without many technicalities, assuming minimal ML background. I will conclude the talk with potential applications of the introduced ideas for physics research.