A longstanding hypothesis held that some problems would always be hard for computers to solve quickly. But OpenAI disproved it — raising the possibility of radical new efficiencies.