I am an intermediate level chatbot developer. Till December 2019 I was working on Lex, Dailogflow, and Watson and currently I am working on RASA. but I want to dig more deeper into NLP part. So will it be better to work directly on Tensorflow or using RASA is good.
Also there is a recent buzz about BERT, Blender and GPT-3 in last 2 months. What should be the approach from now considering these recent advancements?