> The question if machine learning algorithm's produced embeddings will have the acceptable quality too
there are tons of benchmarks and results which demonstrated that embeddings from language models are superior to word2vec in (almost) all scenarios.
