Large Language Models Struggle with Unreasonability in Math Problems
National Key Laboratory for Multimedia Information Processing, School of Computer Science, Peking University(国家多媒体信息处理重点实验室,计算机学院,北京大学);Institute of Artificial Intelligence, Beihang University(人工智能学院,北航);Huawei Noah’s Ark Lab, China(华为诺亚实验室,中国)
Jingyuan Ma, Damai Dai, Zihang Yuan, Rui li, Weilin Luo, Bin Wang, Qun Liu, Lei Sha, Zhifang Sui
大型语言模型(LLMs)在广泛的数学和推理基准测试中展现了显著的成功。然而,我们观察到它们在面对不合理的数学问题时常常表现挣扎。模型未能识别这些问题,而是频繁地继续求解,仿佛问题是适定的,从而产生错误答案或陷入过度思考和冗长的自我纠正。为了系统地研究这一被忽视的脆弱性,我们提出了Unreasonable Math Problems(UMP)基准,旨在评估LLMs检测并应对不合理数学问题陈述的能力。基于涵盖19个LLMs的广泛实验,我们发现,即使是像GPT-4o这样最先进的通用模型在UMP上也仅获得0.6分。虽然像DeepSeek-R1这样的推理模型对不合理输入表现出更高的敏感度,但这通常以生成过于冗长且无意义、无法收敛的回复为代价。我们进一步探索了提示和微调方法,这些方法提供了部分改进,但也引入了权衡,揭示了LLMs在这一具有挑战性的环境中的潜力和局限性。
英文摘要:
Large Language Models (LLMs) have shown remarkable success on a wide range of math and reasoning benchmarks. However, we observe that they often struggle when faced with unreasonable math problems. Instead of recognizing these issues, models frequently proceed as if the problem is well-posed, producing incorrect answers or falling into overthinking and verbose self-correction. To systematically investigate this overlooked vulnerability, we propose the \textbf{Unreasonable Math Problems (UMP)} benchmark, designed to evaluate LLMs' ability to detect and respond to unreasonable math problem statements. Based on extensive experiments covering 19 LLMs, we find that even state-of-the-art general models like GPT-4o achieve only a score of 0.6 on UMP. While reasoning models such as DeepSeek-R1 demonstrate a higher sensitivity to unreasonable inputs, this often comes at the cost of generating overly long and meaningless responses that fail to converge. We further explore prompting and fine-tuning methods, which offer partial improvements but also introduce trade-offs, shedding light on both the potential and limitations of LLMs in this challenging setting.
Large language models struggle with ethnographic text annotation
Calleva Research Centre Oxford Internet Institute University of Oxford(牛津大学奥克斯福德互联网研究所卡列瓦研究中心);Cohn Institute for the History and Philosophy of Science and Ideas Tel Aviv University(特拉维夫大学科恩研究所);Birkbeck College University of London(伦敦大学伯克贝克学院);Centre for the Study of Social Cohesion University of Oxford(牛津大学社会凝聚力研究所以及牛津大学)
Why Do Large Language Models (LLMs) Struggle to Count Letters?
College of Mechanical & Electrical Engineering, Nanjing University of Aeronautics and Astronautics(南京航空航天大学机电学院);ETSI de Telecomunicación, Universidad Politécnica de Madrid(马德里理工大学电信高级工程学院)
Exploring Language Model Generalization in Low-Resource Extractive QA
Pennsylvania State University(宾夕法尼亚州立大学);Mohamed bin Zayed University of Artificial Intelligence(穆罕默德·本·扎耶德人工智能大学);Indian Institute of Technology (IIT) Bhubaneswar(印度理工学院布巴内斯瓦尔分校)
The Sound of Absence: Audio-Language Embedding Models Struggle with Negation
Graduate Institute of Communication Engineering, National Taiwan University, Taiwan(国家台湾大学通讯工程研究所);Artificial Intelligence Center of Research Excellence (AI-CoRE), National Taiwan University, Taiwan(国家台湾大学人工智能卓越研究中心)
Why Diffusion Language Models Struggle with Truly Parallel (Non-Autoregressive) Decoding?
The Hong Kong Polytechnic University, Hong Kong, China(香港理工大学);ELLIS Institute Tübingen, Tübingen, Germany(图宾根ELLIS研究所);Max Planck Institute for Intelligent Systems, Tübingen, Germany(智能系统马克斯·普朗克研究所);Tübingen AI Center, Tübingen, Germany(图宾根人工智能中心);University of Surrey, Guildford, United Kingdom(萨里大学);The University of North Carolina at Chapel Hill, Chapel Hill, NC, USA(北卡罗来纳大学教堂山分校)
Camellia: Benchmarking Cultural Biases in LLMs for Asian Languages
Georgia Institute of Technology(佐治亚理工学院);Samsung R&D Institute Philippines(三星菲律宾研发院);Sungkyunkwan University(成均馆大学);Tohoku University(东北大学);National University of Singapore(新加坡国立大学);University of Copenhagen(哥本哈根大学);University of Michigan(密歇根大学);Indian Institute of Technology Delhi(印度理工学院德里);Institute of Science Tokyo(东京科学研究院)
Evaluating Large Language Models on Controlled Generation Tasks
University of Southern California(南加州大学);University of California, Los Angeles(加州大学洛杉矶分校);Amazon(亚马逊);Google DeepMind(谷歌DeepMind);ETH Zurich(苏黎世联邦理工学院)
Multilingual LLMs Struggle to Link Orthography and Semantics in Bilingual Word Processing
Department of Electrical Engineering, Indian Institute of Technology Delhi(印度理工学院德里电气工程系);Yardi School of Artificial Intelligence, Indian Institute of Technology Delhi(印度理工学院德里人工智能学院)
CoCoLM: COmplex COmmonsense Enhanced Language Model with Discourse Relations
HKUST(香港科技大学);Tencent AI Lab(腾讯人工智能实验室)
大规模的预训练语言模型已经表现出强大的知识表示能力。然而,最近的研究表明,即使这些巨大的模型包含丰富的简单常识知识(例如,鸟会飞,鱼会游。他们经常与涉及多种可能性的复杂常识知识(以动词为中心的短语,例如,识别"Jim yells at Bob“和" Bob is upset”之间的关系)。为了解决这个问题,在本文中,我们提出帮助预训练的语言模型更好地结合复杂的常识知识。不同于现有的微调方法,我们不专注于一个特定的任务,并提出了一个通用的语言模型命名为CoCoLM。通过对大规模可能性知识图谱ASER的仔细训练,我们成功地教授了预先训练的语言模型(即,BERT和ROBERTa)在可能性中丰富了复杂的常识性知识。在多个下游常识任务上的实验表明了CoCoLM的有效性。
Training Data Size Sensitivity in Unsupervised Rhyme Recognition
Institute of Czech Literature, Czech Academy of Sciences(捷克科学院捷克文学研究所);University of Tartu(塔尔图大学);Charles University(查理大学);Tilburg University(蒂尔堡大学);University of Basel(巴塞尔大学);University of Ljubljana(卢布尔雅那大学);Institute of Polish Language, Polish Academy of Sciences(波兰科学院波兰语言研究所);University of Potsdam(波茨坦大学)
To Know or Not To Know? Analyzing Self-Consistency of Large Language Models under Ambiguity
LMU Munich(慕尼黑大学);Faculty of Philological and Cultural Studies, University of Vienna(维也纳大学语言与文化学院);Munich Center for Machine Learning (MCML)(慕尼黑机器学习中心);UniVie Doctoral School Computer Science(维也纳大学计算机博士学院)
RealMath-Eval: Why SOTA Judges Struggle with Real Human Reasoning
University of Wisconsin–Madison(威斯康星大学麦迪逊分校);East China Normal University(华东师范大学);New York University(纽约大学);Tongji University(同济大学);The Chinese University of Hong Kong, Shenzhen(香港中文大学(深圳))
The Validation Gap: A Mechanistic Analysis of How Language Models Compute Arithmetic but Fail to Validate It
DISI, University of Trento(特伦托大学DISI中心);MaiNLP, Center for Information and Language Processing, LMU Munich(慕尼黑大学信息与语言处理中心);Munich Center for Machine Learning (MCML), Munich, Germany(慕尼黑机器学习中心);Free University of Bozen-Bolzano, Italy(博兹纳自由大学)
A Closer Look at the Self-Verification Abilities of Large Language Models in Logical Reasoning
Institute for Artificial Intelligence, Tsinghua University(清华大学人工智能研究院);Beijing National Research Center for Information Science and Technology(北京信息科学与技术国家研究中心);Tencent AI Lab(腾讯AI实验室)