Testing LLM Arithmetic Reasoning Generalization with Automatic Numeric-Remapping Attacks

• URL: http://arxiv.org/abs/2606.03606v2
• Date: Wed, 03 Jun 2026 12:40:36 GMT
• Title: Testing LLM Arithmetic Reasoning Generalization with Automatic Numeric-Remapping Attacks
• Abstract summary: We introduce an automatic algorithm for generating numeric-remapping attacks on arithmetic word problems.<n>We evaluate DeepSeek-R1 (70B), Gemma4 (31B), and GPT-OSS (120B) on GSM8K, MAWPS, and MultiArith.
• Score: 2.6659012768342145
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language models achieve strong performance on arithmetic reasoning benchmarks, and one common response to arithmetic brittleness is to delegate computation to code. Yet models are still often used in settings where they must reason directly from natural language, and trustworthy models should solve small-number arithmetic word problems without external tools. Prior work shows that LLMs are sensitive to numerical variation: a model may solve an original problem but fail on structurally similar variants requiring the same reasoning procedure with different numbers. We ask whether this fragility persists under a stricter setting involving small, schema-preserving numeric changes that retain the original reasoning program and avoid large-number stress tests. We introduce an automatic algorithm for generating numeric-remapping attacks on arithmetic word problems. Unlike template-based perturbation methods requiring manual schemas or constraints, our approach derives problem-specific symbolic representations, generates constrained numeric remappings, recomputes gold answers, and realizes transformed questions through deterministic edits guided by LLM-generated edit plans. Stage-wise validation and a high-confidence audit retain reliable attacks, making the pipeline scalable with limited human intervention. We evaluate DeepSeek-R1 (70B), Gemma4 (31B), and GPT-OSS (120B) on GSM8K, MAWPS, and MultiArith. On GSM8K, completed runs show conditional accuracy drops of 12.16 to 25.82 percentage points. MAWPS and MultiArith are far more stable, with most attacked accuracies near or above 98%. These results show that numeric-remapping robustness depends strongly on dataset structure: GSM8K remains sensitive even when reasoning programs are preserved and answers are recomputed, while shorter, more regular datasets are more robust.

Related papers

• Representation Robustness Under Executable Reasoning Constraints in Large Language Models for Mathematical Problem Solving [3.0618862102164996]
  This paper investigates representation robustness in large language models (LLMs)<n>We find substantial representational sensitivity, with nontrivial flip rates across story, symbolic, and word-equation variants.<n>We argue that representation should be treated as a first-class interface design variable in LLM evaluation and deployment.
  arXiv  Detail & Related papers  (2026-07-08T17:21:28Z)
• Constrained Decoding for Diffusion Language Models via Efficient Inference over Finite Automata [57.27430779838529]
  We present an exact and tractable algorithm for sampling from the constrained mean-field posterior under any constraint expressible as a finite automaton.<n>The approach guarantees constraint satisfaction by construction, supports both greedy and sampling-based decoding, and is compatible with parallel and block-wise decoding.<n> Empirical evaluations on Dream-7B and LLaDA-8 show substantial accuracy gains across various tasks.
  arXiv  Detail & Related papers  (2026-07-08T05:48:57Z)
• Multiplication in Multimodal LLMs: Computation with Text, Image, and Audio Inputs [0.8129517423858409]
  Multimodal LLMs can accurately perceive numerical content across modalities yet fail to perform exact multi-digit multiplication.<n>We introduce a controlled multimodal multiplication benchmark that factorially varies digit length, digit sparsity, representation.<n>We also define arithmetic load, C, as the product of the total and non-zero digit count.
  arXiv  Detail & Related papers  (2026-04-20T12:52:07Z)
• SciML Agents: Write the Solver, Not the Solution [69.5021018644143]
  We introduce two new datasets: a diagnostic dataset of adversarial "misleading" problems; and a large-scale benchmark of 1,000 diverse ODE tasks.<n>We evaluate open- and closed-source LLM models along two axes: (i) unguided versus guided prompting with domain-specific knowledge; and (ii) off-the-shelf versus fine-tuned variants.<n>Preliminary results indicate that careful prompting and fine-tuning can yield a specialized LLM agent capable of reliably solving simple ODE problems.
  arXiv  Detail & Related papers  (2025-09-12T02:53:57Z)
• Understanding and Mitigating Numerical Sources of Nondeterminism in LLM Inference [31.2331188304598]
  Changing system configuration, such as evaluation batch size, GPU count, and GPU version, can introduce significant differences in the generated responses.<n>We trace the root cause of this variability to the non-associative nature of floating-point arithmetic under limited numerical precision.<n>Inspired by this, we develop a lightweight inference pipeline, dubbed LayerCast, that stores weights in 16-bit precision but performs all computations in FP32.
  arXiv  Detail & Related papers  (2025-06-11T08:23:53Z)
• Fast Controlled Generation from Language Models with Adaptive Weighted Rejection Sampling [90.86991492288487]
  evaluating constraint on every token can be prohibitively expensive.<n> LCD can distort the global distribution over strings, sampling tokens based only on local information.<n>We show that our approach is superior to state-of-the-art baselines.
  arXiv  Detail & Related papers  (2025-04-07T18:30:18Z)
• Do Large Language Models Truly Grasp Addition? A Rule-Focused Diagnostic Using Two-Integer Arithmetic [21.014229380679975]
  Large language models (LLMs) achieve impressive results on advanced mathematics benchmarks but sometimes fail on basic arithmetic tasks.<n>We investigate whether they have truly grasped fundamental arithmetic rules or are merely relying on pattern matching.<n>Our evaluation of 12 leading LLMs reveals a stark disconnect: while models achieve high numeric accuracy, they systematically fail these diagnostics.
  arXiv  Detail & Related papers  (2025-04-07T16:57:10Z)
• Sample, Don't Search: Rethinking Test-Time Alignment for Language Models [55.2480439325792]
  We introduce QAlign, a new test-time alignment approach.<n>As we scale test-time compute, QAlign converges to sampling from the optimal aligned distribution for each individual prompt.<n>By adopting recent advances in Markov chain Monte Carlo for text generation, our method enables better-aligned outputs without modifying the underlying model or even requiring logit access.
  arXiv  Detail & Related papers  (2025-04-04T00:41:40Z)
• Mathematical Reasoning in Large Language Models: Assessing Logical and Arithmetic Errors across Wide Numerical Ranges [0.0]
  We introduce GSM-Ranges, a dataset generator that systematically perturbs numerical values in math problems to assess model robustness across varying numerical scales.<n>We also propose a novel grading methodology that distinguishes between logical and non-logical errors, offering a more precise evaluation of reasoning processes beyond computational accuracy.
  arXiv  Detail & Related papers  (2025-02-12T09:53:10Z)
• DCR: Divide-and-Conquer Reasoning for Multi-choice Question Answering with LLMs [9.561022942046279]
  We propose Divide and Conquer Reasoning (DCR) to enhance the reasoning capability of large language models (LLMs) We first categorize questions into two subsets based on confidence score ($mathcalCS$), which is estimated by statistical frequency of generated answers. In particular, we first categorize questions into two subsets based on confidence score ($mathcalCS$), which is estimated by statistical frequency of generated answers.
  arXiv  Detail & Related papers  (2024-01-10T14:38:46Z)
• The Consensus Game: Language Model Generation via Equilibrium Search [73.51411916625032]
  We introduce a new, a training-free, game-theoretic procedure for language model decoding. Our approach casts language model decoding as a regularized imperfect-information sequential signaling game. Applying EQUILIBRIUM-RANKING to LLaMA-7B outperforms the much larger LLaMA-65B and PaLM-540B models.
  arXiv  Detail & Related papers  (2023-10-13T14:27:21Z)
• SatLM: Satisfiability-Aided Language Models Using Declarative Prompting [68.40726892904286]
  We propose a new satisfiability-aided language modeling (SatLM) approach for improving the reasoning capabilities of large language models (LLMs) We use an LLM to generate a declarative task specification rather than an imperative program and leverage an off-the-shelf automated theorem prover to derive the final answer. We evaluate SATLM on 8 different datasets and show that it consistently outperforms program-aided LMs in the imperative paradigm.
  arXiv  Detail & Related papers  (2023-05-16T17:55:51Z)
• MathPrompter: Mathematical Reasoning using Large Language Models [7.953723258038284]
  Large Language Models (LLMs) have limited performance when solving arithmetic reasoning tasks. MathPrompter uses the Zero-shot chain-of-thought prompting technique to generate multiple Algebraic expressions or Python functions to solve the same math problem in different ways.
  arXiv  Detail & Related papers  (2023-03-04T04:43:49Z)
• NAPG: Non-Autoregressive Program Generation for Hybrid Tabular-Textual Question Answering [52.10214317661547]
  Current numerical reasoning methods autoregressively decode program sequences. The accuracy of program generation drops sharply as the decoding steps unfold due to error propagation. In this paper, we propose a non-autoregressive program generation framework.
  arXiv  Detail & Related papers  (2022-11-07T11:25:21Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.