My initial thought was that LLMs will help solve this problem, because they can munch through large inputs quickly. That just pushes the problem out a step, though. You have to trust the LLM.