# Large Language Models Badly Generalize across Option Length, Problem Types, and Irrelevant Noun Replacements

arXiv: https://arxiv.org/abs/2502.12459 · HTML: https://mlsanity.com/paper/2502.12459/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 21 of its 30 references that are in the map.
- First posted on arXiv 18 Feb 2025.

## Builds on (most cited first, 21 of 21)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [Sparks of Artificial General Intelligence: Early experiments with GPT-4](https://mlsanity.com/paper/2303.12712/) (2023)
- [MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark](https://mlsanity.com/paper/2406.01574/) (2024)
- [Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://mlsanity.com/paper/2005.04118/) (2020)
- [Large Language Models Are Not Robust Multiple Choice Selectors](https://mlsanity.com/paper/2309.03882/) (2023)
- [PAWS: Paraphrase Adversaries from Word Scrambling](https://mlsanity.com/paper/1904.01130/) (2019)
- [A Comprehensive Capability Analysis of GPT-3 and GPT-3.5 Series Models](https://mlsanity.com/paper/2303.10420/) (2023)
- [Stress Test Evaluation for Natural Language Inference](https://mlsanity.com/paper/1806.00692/) (2018)
- [Don't Make Your LLM an Evaluation Benchmark Cheater](https://mlsanity.com/paper/2311.01964/) (2023)
- [GSM-Plus: A Comprehensive Benchmark for Evaluating the Robustness of LLMs as Mathematical Problem Solvers](https://mlsanity.com/paper/2402.19255/) (2024)
- [When Benchmarks are Targets: Revealing the Sensitivity of Large Language Model Leaderboards](https://mlsanity.com/paper/2402.01781/) (2024)
- [DyVal: Dynamic Evaluation of Large Language Models for Reasoning Tasks](https://mlsanity.com/paper/2309.17167/) (2023)
- [Can multiple-choice questions really be useful in detecting the abilities of LLMs?](https://mlsanity.com/paper/2403.17752/) (2024)
- [How Robust is GPT-3.5 to Predecessors? A Comprehensive Study on Language Understanding Tasks](https://mlsanity.com/paper/2303.00293/) (2023)
- [People cannot distinguish GPT-4 from a human in a Turing test](https://mlsanity.com/paper/2405.08007/) (2024)
- [Changing Answer Order Can Decrease MMLU Accuracy](https://mlsanity.com/paper/2406.19470/) (2024)
- [KIEval: A Knowledge-grounded Interactive Evaluation Framework for Large Language Models](https://mlsanity.com/paper/2402.15043/) (2024)

## Built on by (most cited first, 4 of 4)

- [UnsafeChain: Enhancing Reasoning Model Safety via Hard Cases](https://mlsanity.com/paper/2507.21652/) (2025)
- [QualBench: Benchmarking Chinese LLMs with Localized Professional Qualifications for Vertical Domain Evaluation](https://mlsanity.com/paper/2505.05225/) (2025)
- [Evaluation is All You Need: Strategic Overclaiming of LLM Reasoning Capabilities Through Evaluation Design](https://mlsanity.com/paper/2506.04734/) (2025)
- [The Hitchhiker's Guide to Agentic AI: From Foundations to Systems](https://mlsanity.com/paper/2606.24937/) (2026)

## Abstract (arXiv)

In this paper, we propose a ``Generalization Stress Test" to assess Large Language Models' (LLMs) generalization ability under slight and controlled perturbations, including option length, problem types, and irrelevant noun replacements. We achieve novel and significant findings that, despite high benchmark scores, LLMs exhibit severe accuracy drops and unexpected biases (e.g., preference for longer distractors) when faced with these minor but content-preserving modifications. For example, Qwen 2.5 1.5B's MMLU score rises from 60 to 89 and drops from 89 to 36 when option lengths are changed without altering the question. Even GPT4o experiences a 25-point accuracy loss when problem types are changed, with a 6-point drop across all three modification categories. These analyses suggest that LLMs rely heavily on superficial cues rather than forming robust, abstract representations that generalize across formats, lexical variations, and irrelevant content shifts.