GeoFaith: A Spatio-Temporal Dual View of Faithful Chain-of-Thought

2026-05-26

AI Summary

This work addresses the post-hoc rationalization of reasoning chains in large language models induced by outcome supervision, proposing the GeoFaith framework to enhance the faithfulness and trustworthiness of reasoning processes. GeoFaith jointly optimizes process faithfulness, outcome correctness, and trajectory consistency through a novel spatiotemporal dual-perspective modeling paradigm that integrates geometric representation learning with dynamic entropy analysis. The framework establishes a scalable mechanism for faithfulness evaluation and enhancement, incorporating bootstrapped data augmentation, an 8B-parameter faithfulness detector, and faithfulness-aware reinforcement learning. Experimental results demonstrate that GeoFaith surpasses GPT-5 in faithfulness detection on standard benchmarks while generating shorter, more interpretable reasoning chains that maintain high accuracy.

Abstract

Chain-of-Thought (CoT) reasoning has advanced large language models (LLMs), but outcome-based supervision leads to pervasive post-hoc rationalization, producing plausible yet unfaithful reasoning chains. Most prior faithfulness assessment methods are either unscalable, expensive, or unreliable. We propose GeoFaith, a spatio-temporal framework that leverages latent geometric structure and entropy dynamics to diagnose and enforce faithful reasoning. We develop a scalable bootstrapping pipeline expanding step-level annotations from 1k to 20k samples across four domains, train an 8B faithfulness detector outperforming GPT-5 on standard benchmarks, and design a faithfulness-aware reinforcement learning framework jointly optimizing outcome correctness, process faithfulness, and trajectory consistency. Experiments show the proposed method achieves superior performance on both faithfulness detection and downstream reasoning, producing shorter, more interpretable chains without sacrificing accuracy. Our code will be made available publicly.

Problem

• Chain-of-Thought
• faithfulness
• post-hoc rationalization
• reasoning
• large language models

Innovation

• faithful reasoning
• spatio-temporal framework
• entropy dynamics
• bootstrap annotation
• reinforcement learning