How do you think you can design a benchmark to solve truly novel problems?