Treating Latent Chain-of-Thought Reasoning as a Dynamical System Reveals Hidden Deliberation
The paper tackles interpretability of latent reasoning methods like CODI and COCONUT, which embed multiple candidate traces in hidden states. By modeling the latent process as a dynamical system, it provides a formal framework to decode step‑by‑step deliberations and extract interpretable trajectories from the model’s hidden space, opening a path to debug internal reasoning without explicit prompts.