Descent With a Weight Penalty

A single-weight model is trained on an objective that combines a data-fitting term with a penalty that discourages large weights:

$L(w) = (w - 5)^2 + 4w^2$

Gradient descent starts at $w_0 = 3$ with $\alpha = 0.05$ and applies $w \leftarrow w - \alpha L'(w)$.

What is $w_3$, and what value do the iterates settle on if the run continues?