Side note: I also think that weaving logical inference and llms together into a virtuous cycle is an interesting topic to explore.