The point is not to improve it until you actually need to change it.
Like you said, you're probably bad at predicting which code is going to need changes in the future, so if it's working now and you don't need to change it then it's a silly risk to try to improve it right now.
Only if you don't care about being able to understand and debug the code in the future!
And even then it's if you're confident in predicting how much code will or won't need to change, which, well...