AI Paper Atlas

Research map / Large language models / Agents, code and math

Large Language Models Struggle with Unreasonability in Math Problems

Jingyuan Ma, Damai Dai, Zihang Yuan, Rui Li, Weilin Luo, Bin Wang, Qun Liu, Lei Sha, Zhifang Sui

AAAI 2026 · Large language models › Agents, code and math · approach: Mathematical and structural reasoning benchmarks · 23 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add critical-thinking instruction to prompt (ablation)Qwen2.5-7B-Instruct · UMP (GSM8K+MATH)absolute score (%): 43.59 -> 66.88Improved
add critical-thinking instruction to prompt (ablation)Qwen2.5-7B-Instruct · UMP (GSM8K+MATH)original accuracy (%): 88.35 -> 86.96Hurt
add critical-thinking instruction to prompt (ablation)R1-Distill-Qwen-7B · UMP (GSM8K+MATH)absolute score (%): 63.06 -> 69.50Improved
add critical-thinking instruction to prompt (ablation)R1-Distill-Qwen-7B · UMP (GSM8K+MATH)original accuracy (%): 91.96 -> 89.88Hurt
fine-tune on unreasonableness-detection responses (ablation)Qwen2.5-7B · UMP held-out testabsolute score (%): 41.11 -> 49.01Improved
fine-tune on unreasonableness-detection responses (ablation)Qwen2.5-7B · UMP held-out testoriginal accuracy (%): 80.26 -> 79.93No clear effect
fine-tune on unreasonableness-detection responses (ablation)Qwen2.5-3B · UMP held-out testabsolute score (%): 34.94 -> 46.30Improved
fine-tune on unreasonableness-detection responses (ablation)Qwen2.5-3B · UMP held-out testoriginal accuracy (%): 70.90 -> 67.55Hurt
use reasoning model instead of general modelDeepSeek-R1 vs DeepSeek-V3 · UMP averageabsolute score: 0.594 -> 0.815Improved
use math-specialized training instead of general modelQwen2.5-Math-7B vs Qwen2.5-7B-Instruct · UMP averageabsolute score: 0.429 -> 0.267Hurt
present unreasonable instead of original problem to reasoning modelsreasoning models · UMPreflection countHurt
present unreasonable instead of original problemreasoning models · UMPnormalized token entropy: decrease of 0.1-0.15Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy