Work

Observable Patterns Are Not Explanations: A Causal-Geometric Analysis of Latent Reasoning Models

Year: 2026

Type: preprint

Abstract: Latent reasoning models (LRMs) replace explicit chain-of-thought with continuous thoughts. Recent work treats observable latent-state patterns, such as BFS-like frontiers and decodable arithmetic computation, as evidence for internal reasoning mechanisms. Evaluating two LRMs (Coconut and CODI) against controls lacking the proposed recurrence or curriculum, we find these patterns also appear in the controls and do not always causally affect behavior. Causal interventions reveal that latent-thought utilization is not binary but graded, scaling with a thought's causal effect on model behavior. Geometric analyses reveal this effect concentrates in low-rank directions whose step-to-step geometry grows more structured as their behavioral influence increases. Latent thoughts should therefore be treated as hidden computation, not hidden explanation: decodability, attention, or static structure alone cannot establish mechanism. LRM interpretability thus requires matched controls and causal tests.

Source: arXiv (Cornell University)

DOI: 10.48550/arxiv.2606.12689

Authors: Aswal, Darpan, Ferraz, Thomas Palmeira, Zhou, Yongxin, Peyrard, Maxime

Institutions: Institut polytechnique de Grenoble, Centre National de la Recherche Scientifique, Université Paris-Saclay, Laboratoire d'Informatique de Grenoble, Université Grenoble Alpes, Naver (South Korea)

Language: English

Topic: Explainable Artificial Intelligence (XAI)

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

Funders: Agence Nationale de la Recherche, Centre National de la Recherche Scientifique

Awards: MIAI @ Grenoble Alpes, G3010527862, G4510828401, MIAI Cluster, Causal Abstraction Benchmark: Rigorous Evaluation of Algorithmic Interpretability

OpenAlex ID: W7164548276

Cites:

Cited by:

Related to: