This article presents a survey of recent advances in the application of cascaded and neural network models to optimize digital predistortion (DPD) of transmitted in-phase and quadrature-phase baseband signals to force linear performance of the PA at RF. Model order reduction techniques enable scalability of ANNs and N-stage CC behavioral models to handle strong nonlinearities and memory effects in PAs crucial for multilayer DPD topologies where the number of coefficients increases significantly with the addition of layers. DPD linearization leveraged by AI models for a high-efficiency dual-input pseudo-Doherty load-modulated balanced amplifier (PD-LMBA) are validated with a linearization testbed.