Dissecting Failure Dynamics in Large Language Model Reasoning

TL;DR

This paper investigates how and where large language models fail during reasoning, revealing that errors often originate from specific early points in the reasoning process and proposing a framework to detect and correct these failures.

Contribution

The paper introduces GUARD, a novel inference-time method that identifies and redirects critical reasoning transitions to improve LLM reliability.

Findings

Errors often originate from early transition points in reasoning trajectories.

Localized spikes in token entropy correlate with reasoning failures.

Interventions based on failure dynamics improve reasoning accuracy.

Abstract

Large Language Models (LLMs) achieve strong performance through extended inference-time deliberation, yet how their reasoning failures arise remains poorly understood. By analyzing model-generated reasoning trajectories, we find that errors are not uniformly distributed but often originate from a small number of early transition points, after which reasoning remains locally coherent but globally incorrect. These transitions coincide with localized spikes in token-level entropy, and alternative continuations from the same intermediate state can still lead to correct solutions. Based on these observations, we introduce GUARD, a targeted inference-time framework that probes and redirects critical transitions using uncertainty signals. Empirical evaluations across multiple benchmarks confirm that interventions guided by these failure dynamics lead to more reliable reasoning outcomes. Our findings highlight the importance of understanding when and how reasoning first deviates, complementing existing approaches that focus on scaling inference-time computation.