Why do we still need to retrain Rasa while BERT and other models are already trained?

I’m new with RASA and NLP bots in general. I’ve been coding conversational bots with RASA and I have a silly question.

Since we can plug BERT, GPT-X and the likes in the pipeline, I don’t understand why we need to ‘re train’ Rasa with things such as entity recognition in the nlu.yml file or even chit chat phrases?