Error Detection and Correction for Interpretable Mathematics in Large Language Models

Yijin Yang, Cristina Cornelio, Mario Leiva, Paulo Shakarian · 2025

Read Error Detection and Correction for Interpretable Mathematics in Large Language Models by Yijin Yang, Cristina Cornelio, Mario Leiva, Paulo Shakarian (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-03500.txt.

Description

Yijin Yang, Cristina Cornelio, Mario Leiva, and Paulo Shakarian work on a flaw in how large language models do math, in a 2025 paper. These models often show their steps, yet a wrong step can spoil the final answer. They also make up facts and stray from set output forms. That is a real risk when the answer must be an exact formula.

Their method, called EDCIM, first has a model build a set of equations for a word problem. A symbolic checker then finds errors in those equations and sends back focused feedback. The model can fix the flagged steps. The aim is math that a reader can follow and trust, with each step open to a check.

Catalog record

CREATORYijin Yang, Cristina Cornelio, Mario Leiva, Paulo Shakarian
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-03500

Browse the full library · About the Tower of Babel