Rethinking External Slow-Thinking: From Snowball Errors to Probability of Correct Reasoning

Zeyu Gan, Yun Liao, Yong Liu · 2025

Read Rethinking External Slow-Thinking: From Snowball Errors to Probability of Correct Reasoning by Zeyu Gan, Yun Liao, Yong Liu (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-2501-15602.txt.

Description

Slow, step-by-step thought in large language models is rethought by Zeyu Gan, Yun Liao, and Yong Liu in a 2025 arXiv paper. Giving a model more time at test can lift its work on hard tasks. Why that helps is still poorly grasped. The authors start from snowball errors, where one wrong step breeds more wrong steps down the chain.

They tie that growth to the chance of a fully right chain, using tools from data theory. Outside slow-thought tricks then read as ways to cut the chance of a wrong chain. A side-by-side study sorts known tricks from plain to rich. Gains, they argue, come less from the trick shape than from wider search.

Catalog record

CREATORZeyu Gan, Yun Liao, Yong Liu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE58 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-15602

Browse the full library · About the Tower of Babel