Great results. Would an AI build for this not be better, or is it just meant as a kind of benchmark for LLMs?