If you reduce it to bytes, yeah, humans are unreasonably efficient learners. But that's not how it works, there's loads more information embedded in a real human speech context.

LLMs learn from unlabeled data. Children definitely do not. There's a huge difference. I would not be surprised if LLMs could learn a lot more efficiently if they had carefully constructed training data, with video and sound.

But also humans have been speaking for so long it's silly to imagine we don't have some evolved language structures in the brain. I don't know why anyone would single that out for skepticism while not questioning e.g. the brain structures for sight, sound, emotions, navigation, etc.

> some evolved language structures in the brain.

That's Chomsky's argument. A small set of constraints for organizing language.

This is testable, right?

Put a kid from one language tradition in a spot with a different language tradition, and they won't be able to learn it.

Eg. kids with native mandarin speaking parents adopted to native Indo-European parents fail at learning English, and will be better at learning Mandarin than their peers with Indo-European heritage.

That doesn't work. Children will learn whatever language they are exposed to. That is why Chomsky and others talk in terms of a universal grammar.

I don't think anyone is suggesting that the brain has structures that are specific to a particular language.