CritICL Boosts Big Model Reasoning Using Small Model Failure Modes
rohanpaul_ai · x · 2026-08-31
Paper "CritICL" proposes using small models' mistakes to improve large model reasoning. Since failure modes are structured across model scales, CritICL runs small models on math problems, saves wrong answers with critiques, and retrieves relevant critiques for the big model's prompt during inference. Results show this method achieves accuracy comparable to test-time scaling (e.g., majority voting) with just 1 generation, significantly reducing token costs and compute requirements.
Related event: CritICL Uses Small Model Failures to Boost LLM Reasoning(2 posts)→
More from Research
- A beginner-friendly introduction to causal inference kicks off, Part 1 — Nasereliver · 2026-10-16
- FBI method trains for Pass@K with one sample per update at 12.3% of the cost — EmtiyazKhan · 2026-10-16
- Berkeley researchers sample humanoid robot policies directly from weight space with diffusion — berkeley_ai · 2026-10-16
- OneWM: unified causal training shows world and action modeling improve each other — IAAI233 · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16
- 5.8M-param Speech Event Detector adds laughs, breaths and背景 sounds to Whisper transcripts — Ayush Kumar Bar · 2026-10-16