When creating the shadow map, we only rendered back faces. This is because if we were to render front faces, points on certain faces would have nearly the same depth as the shadow map's depth, which can cause fluctuations between light and shadow across faces that should be completely lit. The following image shows an example of this effect:

Since the majority of faces that cause this issue are those that are facing the light source, we avoid much of the problem by only rendering back faces during the shadow map pass. This, of course, will only work correctly if your meshes are completely closed. If that is not the case, glPolygonOffset can be used to help the situation by offsetting the depth of the geometry from that in the shadow map. In fact, even when back faces are only rendered when generating the shadow map, similar artifacts can appear on faces that are facing away from the light (back faces in the shadow map, but the front from the camera's perspective). Therefore, it is quite often the case that a combination of front-face culling and glPolygonOffset is used when generating the shadow map.