That GHC might reject an older version of base isn't a reason to switch for every GHC release, is it?
I mean, if it breaks then sure, require a newer base. But in my experience GHC (thankfully!) doesn't change the semantics of Haskell often enough to warrant a new version of base for every new GHC version.
It's hinted at in the section you quoted. A newer ghc might reject older base code as invalid.