Not All Errors Are Created Equal: ASCoT Addresses Late-Stage Fragility in Efficient LLM Reasoning

TL;DR

This paper identifies late-stage errors as more damaging in LLM reasoning and introduces ASCoT, a method that improves efficiency and robustness by focusing verification on critical late reasoning steps, reducing resource use with minimal accuracy loss.

Contribution

The paper proposes ASCoT, a novel approach combining semantic pruning and adaptive verification to address late-stage fragility in LLM reasoning, enhancing efficiency and reliability.

Findings

Reduces token usage by 21-30% on GSM8K and MATH-500

Maintains accuracy with less than 1.8% drop

Effectively reallocates computational resources to critical reasoning steps

Abstract

While Chain-of-Thought (CoT) prompting empowers Large Language Models (LLMs), ensuring reasoning reliability remains an open challenge. Contrary to the prevailing cascading failure hypothesis which posits that early errors are most detrimental, we identify a counter-intuitive phenomenon termed \textbf{Late-Stage Fragility}: errors introduced in later reasoning stages are significantly more prone to corrupting final answers. To address this, we introduce ASCoT (Adaptive Self-Correction Chain-of-Thought), a method harmonizing efficiency with robust verification. ASCoT first employs semantic pruning to compress redundant steps, then utilizes an Adaptive Verification Manager (AVM) to prioritize high risk, late-stage steps via a positional impact score, triggering a Multi-Perspective Self-Correction Engine (MSCE) only when necessary. Experiments on GSM8K and MATH-500 demonstrate that ASCoT effectively reallocates computational resources: it reduces token usage by 21\%--30\% for LLaMA-3.1-8B with negligible accuracy drops ($<1.8\%$), achieving a superior trade-off between inference efficiency and reasoning fidelity.