I found the Lyapunov condition for applying the central limit theorem, which is useful in settings where one has to deal with non-identically distributed random variables:
Lyapunov CLT. Let $s_n^2 = \sum_{k=1}^n \text{Var}[Y_i]$ and let $Y=\sum_i Y_i$. If there exists $\ell>0$ s.t.: $\lim_{n\rightarrow\infty}\left( \frac{1}{s_n^{2+\ell}}\sum_{k=1}^n \text{E}\left[ |Y_k - \text{E}[Y_k]|^{2+\ell}\right] \right) = 0$, then $Z=(Y - E[Y])/\sqrt{Var[Y]}$ converges to the standard normal distribution.
While I have no problem showing that this condition holds for certain exercises, I'm wondering what the intuition is behind this condition.