ReMind : Understanding Deductive Code Reasoning in LLMs DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06

Jun Gao, Yun Peng, Xiaoxue Ren · 2025

Read ReMind : Understanding Deductive Code Reasoning in LLMs DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 by Jun Gao, Yun Peng, Xiaoxue Ren (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-2511-00488.txt.

Description

Jun Gao, Yun Peng, and Xiaoxue Ren study why language models stumble at deductive code reasoning: following a program's execution step by step. Models write code well yet reason about it poorly, and the causes were murky. Their empirical study names three: a built-in gap between generation and reasoning skill, a steady bias toward certain code sources, and weak zero-shot generalization on hard benchmarks.

Their framework, ReMind, answers with three cooperating agents. A mutator creates code variants to break source bias. An executor traces variable states step by step to expose inconsistencies. An inspector flags faulty reasoning steps and refines control flow to close the reasoning gap. Across two benchmarks and five models, ReMind beat baselines and generalized to new problems without examples, a robust pattern for reasoning tools.

Catalog record

CREATORJun Gao, Yun Peng, Xiaoxue Ren
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE55 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-00488

Browse the full library · About the Tower of Babel