Sampling 50/50 choices would be a binary distribution that (very crudely) approximates a normal distribution.

But the counterintuitive thing about the CLT is that it applies to distributions that are not normal.