More Agents Improve Math Problem Solving but Adversarial Robustness Gap Persists

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Alavi, Khashayar, Yeltay, Zhastay, Flek, Lucie, Karimi, Akbar
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866917343697829888
author Alavi, Khashayar
Yeltay, Zhastay
Flek, Lucie
Karimi, Akbar
author_facet Alavi, Khashayar
Yeltay, Zhastay
Flek, Lucie
Karimi, Akbar
contents When LLM agents work together, they seem to be more powerful than a single LLM in mathematical question answering. However, are they also more robust to adversarial inputs? We investigate this question using adversarially perturbed math questions. These perturbations include punctuation noise with three intensities (10%, 30%, 50%), plus real-world and human-like typos (WikiTypo, R2ATA). Using a unified sampling-and-voting framework (Agent Forest), we evaluate six open-source models (Qwen3-4B/14B, Llama3.1-8B, Mistral-7B, Gemma3-4B/12B) across four benchmarks (GSM8K, MATH, MMLU-Math, MultiArith), with various numbers of agents n = {1,2,5,10,15,20,25}. Our findings show that 1) Noise type matters: punctuation noise harm scales with its severity, and the human typos remain the dominant bottleneck, yielding the largest gaps to Clean accuracy and the highest attack success rate (ASR) even with a large number of agents; 2) Collaboration reliably improves accuracy as the number of agents, n, increases, with the largest gains from n=1 to n=5 and diminishing returns beyond n$\approx$10. However, the adversarial robustness gap persists regardless of the agent count.
format Preprint
id arxiv_https___arxiv_org_abs_2511_07112
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle More Agents Improve Math Problem Solving but Adversarial Robustness Gap Persists
Alavi, Khashayar
Yeltay, Zhastay
Flek, Lucie
Karimi, Akbar
Computation and Language
Artificial Intelligence
When LLM agents work together, they seem to be more powerful than a single LLM in mathematical question answering. However, are they also more robust to adversarial inputs? We investigate this question using adversarially perturbed math questions. These perturbations include punctuation noise with three intensities (10%, 30%, 50%), plus real-world and human-like typos (WikiTypo, R2ATA). Using a unified sampling-and-voting framework (Agent Forest), we evaluate six open-source models (Qwen3-4B/14B, Llama3.1-8B, Mistral-7B, Gemma3-4B/12B) across four benchmarks (GSM8K, MATH, MMLU-Math, MultiArith), with various numbers of agents n = {1,2,5,10,15,20,25}. Our findings show that 1) Noise type matters: punctuation noise harm scales with its severity, and the human typos remain the dominant bottleneck, yielding the largest gaps to Clean accuracy and the highest attack success rate (ASR) even with a large number of agents; 2) Collaboration reliably improves accuracy as the number of agents, n, increases, with the largest gains from n=1 to n=5 and diminishing returns beyond n$\approx$10. However, the adversarial robustness gap persists regardless of the agent count.
title More Agents Improve Math Problem Solving but Adversarial Robustness Gap Persists
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2511.07112