![[2602.20770] Pipeline for Verifying LLM-Generated Mathematical Solutions](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.20770] Pipeline for Verifying LLM-Generated Mathematical Solutions
Summary
This paper presents a pipeline for verifying mathematical solutions generated by Large Language Models (LLMs), emphasizing both automatic and interactive verification methods to enhance accuracy over traditional answer-checking approaches.
Why It Matters
As LLMs become increasingly prevalent in solving mathematical problems, ensuring the reliability of their outputs is crucial. This pipeline offers a structured method to verify solutions, potentially reducing errors and improving trust in AI-generated results. The open-source implementation further promotes accessibility and collaboration in the field.
Key Takeaways
- Introduces a verification pipeline for LLM-generated math solutions.
- Emphasizes the importance of both automatic and interactive verification.
- Demonstrates low probability of false positives through experiments.
- Includes a generator for correct solutions in formal and informal languages.
- Provides an open-source implementation for community use.
Computer Science > Artificial Intelligence arXiv:2602.20770 (cs) [Submitted on 24 Feb 2026] Title:Pipeline for Verifying LLM-Generated Mathematical Solutions Authors:Varvara Sazonova, Dmitri Shmelkin, Stanislav Kikot, Vasily Motolygin View a PDF of the paper titled Pipeline for Verifying LLM-Generated Mathematical Solutions, by Varvara Sazonova and 2 other authors View PDF HTML (experimental) Abstract:With the growing popularity of Large Reasoning Models and their results in solving mathematical problems, it becomes crucial to measure their capabilities. We introduce a pipeline for both automatic and interactive verification as a more accurate alternative to only checking the answer which is currently the most popular approach for benchmarks. The pipeline can also be used as a generator of correct solutions both in formal and informal languages. 3 AI agents, which can be chosen for the benchmark accordingly, are included in the structure. The key idea is the use of prompts to obtain the solution in the specific form which allows for easier verification using proof assistants and possible use of small models ($\le 8B$). Experiments on several datasets suggest low probability of False Positives. The open-source implementation with instructions on setting up a server is available at this https URL. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2602.20770 [cs.AI] (or arXiv:2602.20770v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2602.20770 Focus to ...