A Formal Comparison Between Chain-of-Thought and Latent Thought

Kevin Xu, Issei Sato · 2025

Read A Formal Comparison Between Chain-of-Thought and Latent Thought by Kevin Xu, Issei Sato (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-2509-25239.txt.

Description

Xu and Sato formally compare two ways a language model can think: spelling its steps out in words, or reasoning silently in a continuous inner space. Both styles reuse computation step after step, but their strengths were unmapped. Their 2025 analysis proves the silent style can run many threads in parallel more efficiently, since written steps must march one token at a time.

Written reasoning wins elsewhere: through random sampling it can count and draw samples approximately, tricks the silent style lacks. The split suggests matching the style to the job. Tasks needing parallel depth suit silent thought, while counting and sampling suit written steps. The authors offer the results as guidance for choosing reasoning designs.

Catalog record

CREATORKevin Xu, Issei Sato
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE107 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-25239

Browse the full library · About the Tower of Babel