Hah, yeah.

But on a real-world data set, even if the underlying phenomenon is e^x, you'll keep amplifying sample noise until the derivatives are basically random. Assuming you even have enough data to keep taking derivatives.