I'm not sure how to go about this. I know if a sequence $\{a_n\}$ converges to $L$, then given $\epsilon > 0$:
(1) $\exists N$ such that for $n \geq N$, $|a_n - L| < \epsilon$.
Somehow, I must show:
(2) for large $n,m$, $|a_m - a_n| < \epsilon$.
I think by (1), we can choose $m > n \geq N$ such that $|a_m - L| < \epsilon$ and $|a_n - L| < \epsilon$. I am not sure how to reconcile these inequalities to get (2) though.
Let $\{a_n\}$ be a non-Cauchy sequence. Then there exist a (fixed) $\varepsilon \gt 0$ so that for any $N$ we can find $s \ge N$ and $t \ge N$ such that $|a_s - a_t| \ge \varepsilon$. We can certainly take $s$ and $t$ so that $a_s \le a_t$.
We can actually construct two subsequences of $\{a_n\}$ by taking the 'next' $N$ to be larger than all the $s ^{\;'}$s and $t ^{\;'}$s seen so far, with the two subsequences satisfying
$\tag 1 \{a_{s_k}\}_{\, k \ge 0} \text{ and } \{a_{t_k}\}_{\, k \ge 0} \text{ with } a_{s_k} + \varepsilon \le a_{t_k}$
Suppose now that $\{a_n\}$ converges to $L$. Then every subsequence converges to $L$. But if we apply the limit operation to $(1)$, we get
$\qquad L + \varepsilon \le \lim\inf\, a_{t_k} = L = \lim\sup\, a_{t_k}$
which is absurd.