# How is attention different from linear MLPs?

Each output for both the attention layer (as in transformers) and MLPs or feedforward layer(linear-activation) are weighted sums of previous layer. So how they are different?