From scratch, maybe. But nothing stops Haskell and Clojure from binding to the same C-libraries every other language binds to. And I think you're overstating t

From scratch, maybe. But nothing stops Haskell and Clojure from binding to the same C-libraries every other language binds to. And I think you're overstating the difficulty of even a top-to-bottom new system. Re-implementing deep learning would be much, much easier than reinventing deep learning. Still probably not worth doing! But your estimate sounds about right for the latter, not the former.