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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add critical-thinking instruction to prompt (ablation) | Qwen2.5-7B-Instruct · UMP (GSM8K+MATH) | absolute score (%): 43.59 -> 66.88 | Improved |
| add critical-thinking instruction to prompt (ablation) | Qwen2.5-7B-Instruct · UMP (GSM8K+MATH) | original accuracy (%): 88.35 -> 86.96 | Hurt |
| add critical-thinking instruction to prompt (ablation) | R1-Distill-Qwen-7B · UMP (GSM8K+MATH) | absolute score (%): 63.06 -> 69.50 | Improved |
| add critical-thinking instruction to prompt (ablation) | R1-Distill-Qwen-7B · UMP (GSM8K+MATH) | original accuracy (%): 91.96 -> 89.88 | Hurt |
| fine-tune on unreasonableness-detection responses (ablation) | Qwen2.5-7B · UMP held-out test | absolute score (%): 41.11 -> 49.01 | Improved |
| fine-tune on unreasonableness-detection responses (ablation) | Qwen2.5-7B · UMP held-out test | original accuracy (%): 80.26 -> 79.93 | No clear effect |
| fine-tune on unreasonableness-detection responses (ablation) | Qwen2.5-3B · UMP held-out test | absolute score (%): 34.94 -> 46.30 | Improved |
| fine-tune on unreasonableness-detection responses (ablation) | Qwen2.5-3B · UMP held-out test | original accuracy (%): 70.90 -> 67.55 | Hurt |
| use reasoning model instead of general model | DeepSeek-R1 vs DeepSeek-V3 · UMP average | absolute score: 0.594 -> 0.815 | Improved |
| use math-specialized training instead of general model | Qwen2.5-Math-7B vs Qwen2.5-7B-Instruct · UMP average | absolute score: 0.429 -> 0.267 | Hurt |
| present unreasonable instead of original problem to reasoning models | reasoning models · UMP | reflection count | Hurt |
| present unreasonable instead of original problem | reasoning models · UMP | normalized token entropy: decrease of 0.1-0.15 | Hurt |
Closest papers by meaning
- A Careful Examination of Large Language Model Performance on Grade School Arithmetic (NeurIPS 2024 · similarity 0.91)
- GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models (ICLR 2025 · similarity 0.90)
- RealMath: A Continuous Benchmark for Evaluating Language Models on Research-Level Mathematics (NeurIPS 2025 · similarity 0.90)
- Putnam-AXIOM: A Functional & Static Benchmark for Measuring Higher Level Mathematical Reasoning in LLMs (ICML 2025 · similarity 0.90)
- UGMathBench: A Diverse and Dynamic Benchmark for Undergraduate-Level Mathematical Reasoning with Large Language Models (ICLR 2025 · similarity 0.90)
- Learning Theorem Rationale for Improving the Mathematical Reasoning Capability of Large Language Models (AAAI 2025 · similarity 0.90)
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.
