Salvato in:
Dettagli Bibliografici
Autori principali: Carson, Jack David, Reisizadeh, Amir
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2506.04374
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916780433211392
author Carson, Jack David
Reisizadeh, Amir
author_facet Carson, Jack David
Reisizadeh, Amir
contents Transformer LMs show emergent reasoning that resists mechanistic understanding. We offer a statistical physics framework for continuous-time chain-of-thought reasoning dynamics. We model sentence-level hidden state trajectories as a stochastic dynamical system on a lower-dimensional manifold. This drift-diffusion system uses latent regime switching to capture diverse reasoning phases, including misaligned states or failures. Empirical trajectories (8 models, 7 benchmarks) show a rank-40 projection (balancing variance capture and feasibility) explains ~50% variance. We find four latent reasoning regimes. An SLDS model is formulated and validated to capture these features. The framework enables low-cost reasoning simulation, offering tools to study and predict critical transitions like misaligned states or other LM failures.
format Preprint
id arxiv_https___arxiv_org_abs_2506_04374
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Statistical Physics of Language Model Reasoning
Carson, Jack David
Reisizadeh, Amir
Artificial Intelligence
Computation and Language
Transformer LMs show emergent reasoning that resists mechanistic understanding. We offer a statistical physics framework for continuous-time chain-of-thought reasoning dynamics. We model sentence-level hidden state trajectories as a stochastic dynamical system on a lower-dimensional manifold. This drift-diffusion system uses latent regime switching to capture diverse reasoning phases, including misaligned states or failures. Empirical trajectories (8 models, 7 benchmarks) show a rank-40 projection (balancing variance capture and feasibility) explains ~50% variance. We find four latent reasoning regimes. An SLDS model is formulated and validated to capture these features. The framework enables low-cost reasoning simulation, offering tools to study and predict critical transitions like misaligned states or other LM failures.
title A Statistical Physics of Language Model Reasoning
topic Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2506.04374