Text this: Pseudo-derandomizing learning and approximation