人工知能と機械学習における C++ メタプログラミングの展望は何ですか?

#include <concepts>
#include <tuple>
#include <utility>

template <typename T>
concept Number = requires(T x) {
    { x + x } -> std::same_as<T>;
    { x * x } -> std::same_as<T>;
};

template <Number T>
constexpr T power(T base, int exp) {
    if constexpr (exp == 0) {
        return 1;
    } else if constexpr (exp < 0) {
        return 1 / power(base, -exp);
    } else {
        return base * power(base, exp - 1);
    }
}

template <Number T, Number... Ts>
constexpr auto partial_derivatives_at(T (*f)(T, Ts...), std::tuple<T, Ts...> point) {
    auto& [x, ys...] = point;
    return std::tuple(
        []<typename X>(X) -> X { return 1; }(x) +
            std::apply([&](auto& y) -> auto { return partial_derivatives_at<X>(f, std::make_tuple(x, y)); }, std::make_tuple(ys...)),
        std::apply([&](auto& y) -> auto {
            return power(y, 1) *
                std::apply([&](auto& z) -> auto { return partial_derivatives_at<X>(f, std::make_tuple(x, z)); }, std::make_tuple(ys...));
        }, std::make_tuple(ys...)));
}