I am curious. What makes you say generic LLMs will improve? I mean, obviously, they will, but to what extent? How can we know what they will or will not be capable of by the time we reach the limit of feasible improvement? Will they ever be able to explain in a way humans can easily digest?