Metrics like Github stars can be gamed, but it doesn't mean that the number has no value. I would conjecture that the # of stars correlates strongly with other desirable things.
I think of it as similar to the # of citations that an academic paper has - genius can remain ignored and undiscovered, but overall the citation count has embedded in it a whole series of factors that correlate to "should I bother reading this".
Ironically before citing a paper that studied this very thing, I checked many citations it had :)