One Step Downhill

A very small model has a single weight $w$. How badly it performs is measured by a loss function

$L(w) = w^2 - 6w + 13$

and training's whole job is to make $L$ smaller. It does that by repeatedly nudging $w$ using this rule:

$w_{\text{new}} = w_{\text{old}} - \eta \, L'(w_{\text{old}})$

Here $L'(w)$ is the instantaneous rate of change of $L$ at the weight $w$, and $\eta$ (the learning rate) controls how big the nudge is. For this run,

$\eta = 0.3, \qquad w_{\text{old}} = 5$

The model takes exactly one step.

By how much does the loss decrease? Report a positive number, rounded to two decimal places.