One of the simplest problems is that of a quadratic with a large root and a small root, such as
x2 - 1E9*x + 1 = 0
The roots are 1E9 and 1E-9 and you will very likely get 1E9 and zero from most codings of a quadratic root solver such as the old formula of (1/2a)(-b±sqrt(b**2-4ac)).
Anyone should be able to get the roots of
x2 - 4*x + 4 = 0
quite easily as they are both equal to 2.
But, if we make a very small change to this equation
x2 - 4*x + 3.999999 = 0
the roots become 2.001 and 1.999. This is an example where a small relative change in the coefficients induces a much larger change in the roots.
A similar problem involving cubics is
x3 - 1E9*x2 + 3E9*x -2E9 = 0
Here, the roots are 1,2 and 1E9. Your old solver may compute that the smaller roots are zero, or that they are not accurately spaced.