Do LLMs actually think in silence? A new paper formalizes four axioms — Causality, Minimality, Separability, and Stability — to audit latent thought representations. No open-weight LLM satisfies all…