Comment by nri

You are right, data/concept drift affects both approaches.

What I meant to say was that retraining your model might not fix that if you baked strong assumptions into it. I edited the blog post to make it clearer.