This is a homework question for Humphrey's book on Lie algebras. It has solutions online here, on page $20$. There is a typo in the solution, it should be $[u_i,u_j]=0$ for all $i\neq j$.
As @JyrkiLahtonen commented, it suffices to show that $t_{\alpha}\in L_i$ and $t_{\beta}\in L_j$, since this implies that $\operatorname{ad}t_{\alpha}\operatorname{ad}t_{\beta}(x)=[t_{\alpha}[t_{\beta}x]]\in L_i\cap L_j=0$ for all $x\in L$, i.e., $\operatorname{ad}t_{\alpha}\operatorname{ad}t_{\beta}=0$, so $$(\alpha, \beta)=\kappa(t_{\alpha}, t_{\beta})=\operatorname{tr}(\operatorname{ad}t_{\alpha}\operatorname{ad}t_{\beta})=\operatorname{tr}(0)=0.$$
To see that $t_{\alpha}\in L_i$, let $\kappa_i$ denote the Killing form of $L_i$. By Humphreys Lemma 5.1, $\kappa_i=\kappa\vert_{L_i\times L_i}$. By Humphreys Corollary 8.2, $\kappa\vert_{H\times H}$ and $\kappa_i\vert_{H_i\times H_i}$ are nondegenerate. This allows us to identify $H$ with $H^*$, by associating to $\gamma\in H^*$ the unique element $t_{\gamma}\in H$ such that $\gamma(h)=\kappa(t_{\gamma}, h)$ for all $h\in H$. Similarly, we can identify $H_i$ with $H_i^*$, by associating to $\delta\in H_i^*$ the unique element $u_{\delta}\in H_i$ such that $\delta(h_i)=\kappa_i(u_{\delta}, h_i)$ for all $h_i\in H_i$. With this in mind, $u_{\alpha}\in H_i$ and $\alpha(h_i)=\kappa_i(u_{\alpha}, h_i)$ for all $h_i\in H_i$. However, by an argument similar to that in the previous paragraph, $\kappa(L_i, L_k)=0$ for $i\neq k$, so given $h=h_1+\cdots+h_t\in H$, $h_k\in H_k$, we have $$\alpha(h)=\alpha(h_i)=\kappa_i(u_{\alpha}, h_i)=\kappa\vert_{L_i\times L_i}(u_{\alpha}, h_i)=\kappa(u_{\alpha}, h_i)=\kappa(u_{\alpha}, h).$$ Then, by uniqueness, $t_{\alpha}=u_{\alpha}\in H_i\subseteq L_i$. Similarly, $t_{\beta}\in L_j$.
Best Answer
If you are asking why the $u_i$ are in $H$, here is an argument:
Take another arbitrary element $y\in H$. Write $y = v_1+...+v_t$. Then because $H$ is abelian we have $[x,y] =0$, and because the $L_i$ are ideals and the decomposition is direct, this implies $[u_i,v_i] = 0$ for all $i$. But then, again because the $L_i$ are ideals and the decomposition is direct, we also have $[u_i, y] =0$. So we have shown that $u_i$ commutes with all elements in $H$, and you say you have shown it is semisimple (as element of $L_i$, but then also as element of $L$). That implies that $H+span(u_i)$ is an abelian subalgebra and consists of semisimple elements, i.e. is toral. But $H$ was maximal toral, so $u_i \in H$.