CritICL: Inference-Time Weak-to-Strong Generalization from Small LLM Failure Modes

Forum topic · 小凯 · 2026-08-30

Summary

CritICL is a novel inference-time framework for improving LLM reasoning without relying on repeated generation or external verification. The key insight is that LLM failure modes exhibit structured, consistent patterns across model scales within the same family. Rather than treating failures as undesirable outputs, CritICL uses failure modes from weaker models as guidance, incorporating them into inference through critique-based in-context examples. The authors propose two variants: CritICL-dynamic, which adaptively predicts input-specific failure modes and retrieves relevant critiques, and CritICL-static, which uses a global failure-mode profile for stable guidance. Experiments show CritICL consistently outperforms standard in-context learning and matches or exceeds test-time scaling methods while requiring significantly fewer generations and lower token cost. The paper (arXiv 2608.27455) was published on 2026-08-27 by Yufan Wu and colleagues in the NLP field.

Paper Overview

Field: NLP
Authors: Yufan Wu, Yinghui He, Zhengyi Hu, Lang Wei, Ruichen Li, Qifan Yang, Ting Zhu
Published: 2026-08-27
arXiv: 2608.27455

Abstract

Recent advances in inference-time scaling have significantly improved the reasoning performance of large language models (LLMs). However, these methods typically rely on repeated generation or external verification. To address this limitation, the authors introduce CritICL, a novel inference-time framework that improves reasoning while maintaining high efficiency.

The key insight is that LLM failure modes exhibit structured patterns across model scales within the same family. Instead of treating failures as undesirable outputs, CritICL leverages them as a source of guidance. Specifically, failure modes derived from weaker models are incorporated into inference through critique-based in-context examples.

Two variants are proposed:

CritICL-dynamic: adaptively predicts input-specific failure modes and retrieves critiques accordingly.
CritICL-static: uses a global failure-mode profile to provide stable guidance.

Experimental results show that CritICL consistently outperforms standard in-context learning, achieving performance comparable to or better than test-time scaling methods while requiring significantly fewer generations and lower token cost.

Commentary (translated from Chinese)

Recent progress in inference-time scaling has notably improved LLM reasoning performance, but these approaches typically depend on repeated generation or external verification. CritICL addresses this limitation by exploiting the structured nature of failure modes across model scales, turning a weaker model's failures into useful critique-based in-context guidance for a stronger model — a form of weak-to-strong generalization at inference time.