I think you underestimate how much of conceptual space is accessible to linguistic computations and just hasn't been reached that way yet.

Calculators actually allowed us to calculate things that we couldn't before--not because they did math that we couldn't, but because they made it easy enough and/or fast enough to bother.

LLMs can in principle act similarly. We don't have many human experts in all three of history, the chemistry of aromatic hydrocarbons, and fencing. But LLMs are (mediocre) experts in (how to sensibly talk about) all three. If there's anything important to be found via linguistic computations in that area, LLMs are more likely to find it than humans are.

Thus, there's no reason to think that LLMs would be unable to expand the frontiers of human knowledge. Their strengths might lie in different places than humans', but that's why they can likely expand our frontiers: the computations which are comparatively easy for them might be hard for us.