Yes, I am not arguing against model-building. However, logical inference has a tree like structure (not just a chain), which LLMs seems to master. Though logical inference can also be serialized (if we introduce some context).

It's a nice question of whether the emergent logic in LLMs, trained on human texts, can be really universal. I believe mathematics (up to isomorphism) is universal, however it still might be very different what some hypothetical extraterrestrials deal with.

For example, they might have a different foundations found, or maybe are less into discrete math, etc.

We can probably get some hints from human history of development of numbers. From natural numbers to octonions. Also for instance Aristotle logic had some flaws noticed only relatively recently. It may well be that LLMs do have some subtle "bugs", or fall for the same fallacies humans fall.