Get Started
Home
Topics
Search
Library
Research questionWhat determines whether an LLM transfers arithmetic reasoning across numeric and verbal prompts?LLMs may solve an arithmetic problem in numeric form yet fail when the same operation is expressed with words. The difficulty is identifying what governs this failure when only the surface format changes.
AI
Mechanistic Interpretability
Natural Language Processing
Reasoning
Research Paper
Latest papersRecent research connected to this question, newest first.Shared circuits predict whether LLMs generalize across formats in arithmetic reasoningThe evidence concerns arithmetic prompts in numeric form and verbal renditions in English, Spanish, and Italian. Models’ internal circuits are analyzed with attribution patching, and circuit overlap is related to format difficulty, model-level generalization, and item-level correctness without labeled data.research paper · Sep 3, 2026
Related questions
How can we tell whether LLMs follow coherent, human-like prerequisite relationships in mathematical reasoning?Can multilingual LLMs maintain mathematical reasoning when equivalent inputs use different word order or voice?How can mathematical-reasoning LLMs learn to construct counterexamples that reveal conceptual understanding?How can we evaluate LLM reasoning quality beyond final-answer accuracy across deployment contexts?