The six generators are $S^{ab}\equiv i[\gamma^a,\gamma^b]/4$, where $(a,b)$ are the three boosts $(0,1), (0,2) (0,3),$ and the three spatial rotations $(1,2), (2,3), (3,1)$. The Dirac gamma matrices are $4$-by-$4$ matrices. It's true that you can write down the gamma's in terms of two-by-two blocks of two-by-two Pauli sigma's, but that is still four-by-four matrices.
The Lie algebra of $SO(3,1)$ is locally isomophic to that of $SU(2)\times SU(2)$ which is what the $1/2,1/2$ representation refers to, but unless you are dealing with massless (Weyl) particles you need both $SU(2)$'s.
We can take
$$
\gamma_0 = \left(\matrix{0&1\\ 1&0}\right), \quad \gamma^a= \left(\matrix{0&\sigma_a\\ -\sigma_a &0}\right).
$$
We also have
$$ \quad \gamma^5= \left(\matrix{1&0\\ 0&-1}\right),
$$
Then setting $(\Sigma_1,\Sigma_2,\Sigma_3)=(S_{23},S_{31},S_{12})$
$$
\Sigma_a= \frac 12 \left(\matrix{\sigma_a&0\\ 0&\sigma_a}\right)
$$
are the rotation generators, and
$$
K_a=\frac i2 \left(\matrix{\sigma_a&0\\ 0&-\sigma_a}\right)
$$
are the boost generators (I do not guarantee that the signs are correct). Note that the boost generators are not Hermitian. The non-compact Lorentz group $SO(1,3)$ has no finite dimensional unitary representation.
My representation matrices are adapted to the decomposition of the algebra into the two $S_a\pm iK_a$ mutually commuting $SU(2)$ subalgebras of $SO(1,3)$.
If I got my signs right, the matrices should obey the Lorentz algebra
$$
[\Sigma^a,\Sigma^b]= i \epsilon_{abc} \Sigma^c\\
[\Sigma^a,K^b]= i \epsilon_{abc} K^c\\
[K^a,K^b]= - i\epsilon_{abc} \Sigma^c.
$$