My recommendation: learn Python and the basics of statistics, experiment with the APIs of current models, and build something small that you find useful. Understanding how embeddings, transformers, and techniques like RAG work comes more quickly and effectively when you apply them to your own project.