Injective Preprint

New preprint out! Language Models Are Injective and Hence Invertible We prove that LLM representations are injective, and present the first exact inversion algorithm. The announcement blew up with 5M views on Twitter. Check it out!