Two Knobs, Two Steps

A model has two parameters and the loss

$L(a, b) = a^2 + 4b^2 + 2ab - 6a - 16b$

Gradient descent starts at $(a_0, b_0) = (1, 1)$ with learning rate $\alpha = 0.1$, updating both parameters simultaneously from the same gradient:

$a \leftarrow a - \alpha \frac{\partial L}{\partial a}, \qquad b \leftarrow b - \alpha \frac{\partial L}{\partial b}$

What is $(a_2, b_2)$ after two steps?