Language Models are Injective and Hence Invertible

https://arxiv.org/abs/2510.15511

really interesting article proving “almost surely” that llms are injective and therefore also invertible.

this can have some interesting results in the coming years, so worth keeping an eye on.