Current wireless networks have been largely designed as a combination of dedicated processing blocks, such as channel estimation, equalization, and coding/decoding blocks, where each block is designed separately on the basis of mathematical models that define the statistical behaviour of the wireless channels and the underlying data traffic. This model-driven approach will be challenged by the complex and diversified scenarios in which 5G networks and beyond are expected to operate. In addition, with the deployment of ultra-massive Multiple-Input Multiple-Output (MIMO) systems, optimising the physical layer functionalities based on mathematical models and solutions will become prohibitive due to the computational complexity and associated control overhead. Therefore, it is anticipated that conventional mathematical models and solutions will not be able to provide the required enhancement in capacity and performance. To solve some of the above-mentioned problems we intend to use generative AI to assist 5G MIMO systems. The project is part of Dino Pjanics PhD thesis work at LU.