In the standard picture of gravitational lensing, we have a picture like the one you can see as Figure 1 of this paper (I don't know if I can copy/paste it). In it, the source S and the observer O are both at a finite distance from the lens L, and both on the same side. Light curves a little bit around the lens, and an image is formed on the same side of the lens as the source.
Now, in almost every situation one makes the approximation that the source and the observer are infinitely far away from the lens, like this:
This feels weird, because as you can see the image and the source are always on opposite sides of the lens, which is not always the case in realistic situations (for example, with the lensing of a star by the sun). How can this be explained? Is it that the approximation has to be applied only sometimes?