Tweeted By @tunguz

Introducing MASS – A pre-training method that outperforms BERT and GPT in sequence to sequence language generation tasks https://t.co/BzuZd7MKIL

— Bojan Tunguz (@tunguz) June 24, 2019