Samsung Electronics interview question

How are the recent BERT based embedding models different from word2vec?