The former people tend to gravitate toward things like Haskell, which is very pure, conceptually beautiful, and feels like using the LHC to crack nuts. But hey, you're using LHC, great fun.
The latter will pick whatever ugly PHP framework is closest to the final product they want to ship. Most great startup ideas aren't very sophisticated algorithmically wise: existing mature libs in ugly languages beat writing/adapting everything in a cool one, although the latter is a lot more fun; an active and friendly online community, which is familiar with problems close to yours, beats having macros and first-class closures in your language; etc.
In theory people could love both beautiful things and getting things done, but in practice I find that the more strongly they feel about one side, the less they care about the other.
A corollary is that, in fun-to-use languages, everyone writes fun stuff rather than useful ones: frameworks, meta-programming stuff, continuation-based contraptions, etc. So you end up with a dozen half-baked neural network libs, but no decent (and _de facto_ standard) HTTPS client.
TL;DR: the overwhelming majority of startups are about social and economic hacking, not about software hacking. Over-focusing on the language issue is a clue that one hasn't come to terms with this sad truth.