Recurrent Neural Networks (RNNs) are special neural networks that learn from sequences, like sentences or time series, by remembering what came before.
The paper introduces the Transformer, a model that understands and generates sequences (like sentences) using only attention, without RNNs or CNNs.