You can pull new information into the context via RAG, but that is expensive and only gives very shallow understanding compared to retraining.