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

Forum topic · 小凯 · 2026-08-30

Summary

CritICL is a novel inference-time framework for improving large language model (LLM) reasoning without relying on repeated generation or external verification. The key insight is that LLM failure modes exhibit structured patterns across model scales within the same model family. Rather than treating failures as undesirable outputs, CritICL leverages failure modes from weaker models as guidance, incorporating them into inference through critique-based in-context examples. The paper proposes 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 costs. Authored by Yufan Wu, Yinghui He, Zhengyi Hu, and colleagues in the NLP field, the paper is available on arXiv (2608.27455).

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

Key Idea

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 core insight: 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.

Variants

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

Results

CritICL consistently outperforms standard in-context learning.
Performance is comparable to or better than test-time scaling methods.
Requires significantly fewer generations and lower token costs.

Abstract (Original)

> 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, we introduce CritICL, a novel inference-time framework that improves reasoning while maintaining high efficiency. Our 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, we utilize failure modes derived from weaker models and incorporate them into inference through critique-based in-context examples.

---

*Auto-collected on 2026-08-30*