Physics paper claims language models work like dissipating quantum systems, not conserving ones

A theoretical physics paper argues that generating coherent text requires systems that lose information in controlled ways, not systems that preserve it — the opposite of how most AI models are built. The authors prove this using math borrowed from quantum mechanics, suggesting that the ability to forget selectively is what lets mechanical systems produce intelligent writing.

This is pure theoretical work with no empirical validation, no real systems tested, and no way yet to know if the math maps onto how actual language models work. The paper makes a claim about the fundamental physics of intelligence without measuring intelligence in the world. It's the kind of argument that looks elegant on paper and then fails to predict anything real.