LLMs are only capable of performing a finite amount of computation within a single forward pass. We know that much.

They are also known to operate on high level abstracts and concepts - unlike systems operating strictly on formal logic, and very much like humans.