Why do RNN-based Keras models struggle with long sequences?

RNNs suffer from vanishing gradients, limiting long-term dependency learning.

As sequence length increases, gradients shrink exponentially during backpropagation through time, making early inputs irrelevant.