Decomposing LLM Self-Correction: The Accuracy-Correction Paradox and Error Depth Hypothesis

TL;DR

This paper analyzes the self-correction abilities of large language models, revealing a paradox where weaker models correct errors more effectively than stronger ones, and proposing the Error Depth Hypothesis to explain this phenomenon.

Contribution

It systematically decomposes LLM self-correction into detection, localization, and correction, and introduces the Error Depth Hypothesis to explain the inverse relationship between model strength and correction rate.

Findings

Weaker models achieve higher intrinsic correction rates than stronger models.

Error detection capability varies widely across architectures and does not predict correction success.

Providing error location hints can negatively impact model correction performance.

Abstract

Large Language Models (LLMs) are widely believed to possess self-correction capabilities, yet recent studies suggest that intrinsic self-correction--where models correct their own outputs without external feedback--remains largely ineffective. In this work, we systematically decompose self-correction into three distinct sub-capabilities: error detection, error localization, and error correction. Through cross-model experiments on GSM8K-Complex (n=500 per model, 346 total errors) with three major LLMs, we uncover a striking Accuracy-Correction Paradox: weaker models (GPT-3.5, 66% accuracy) achieve 1.6x higher intrinsic correction rates than stronger models (DeepSeek, 94% accuracy)--26.8% vs 16.7%. We propose the Error Depth Hypothesis: stronger models make fewer but deeper errors that resist self-correction. Error detection rates vary dramatically across architectures (10% to 82%), yet detection capability does not predict correction success--Claude detects only 10% of errors but corrects 29% intrinsically. Surprisingly, providing error location hints hurts all models. Our findings challenge linear assumptions about model capability and self-improvement, with important implications for the design of self-refinement pipelines.