Unveiling and Causalizing CoT: A Causal Pespective

Jiarun Fu, Lizhong Ding, Hao Li, Pengqi Li, Qiuning Wei, Xu Chen · 2025

Read Unveiling and Causalizing CoT: A Causal Pespective by Jiarun Fu, Lizhong Ding, Hao Li, Pengqi Li, Qiuning Wei, Xu Chen (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-2502-18239.txt.

Description

Chain-of-thought prompting makes language models show their steps, but steps can sound smooth while being wrong or pointless. Jiarun Fu and five co-authors examine those steps with causal tools in 2025, asking which steps truly drive the answer.

They model a reasoning chain as a structural causal system and define a measure, the average causal effect of a step, to test whether later steps depend on earlier ones. Steps that fail the test are flagged as wrong or unintelligible. A role-playing query routine then rewrites those steps until the chain is causal throughout. In tests on open and closed models, the repaired chains corrected common step errors and lifted reasoning scores.

Catalog record

CREATORJiarun Fu, Lizhong Ding, Hao Li, Pengqi Li, Qiuning Wei, Xu Chen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-18239

Browse the full library · About the Tower of Babel