See Core issue 1564.

Discussed in Rapperswil 2014. EWG points out that auto deduction is moving to the opposite direction, and there's a reason why the suggested deduction isn't done, Vandevoorde showed an example of an overload set including an unconstrained template and a function taking a vector, and explained that the rules currently choose the function taking a vector, and advised against changing that. Closing as NAD.