Why does GNU configure define macros for SIZEOF_INT etc. as specific integer values?

If you're going to define these at all, why not define them as

#define SIZEOF_INT sizeof(int)

etc. so they are correct even if someone tries to use them to compile against a different architecture?