As Linus explains it: is cramming a ton of info into a tiny file. The more homogeneous the data in a file is the more can it be compressed.
Let’s take the example of an image. If it is only white, then the compressed file will say something like this: put a white pixel in these positions [then the range]. The same happens with a jpeg image: you simply tell the computer to build the image pixel by pixel like this: on pixel (x,y) insert a color (rgb color). Neat, huh?