Capabilities and Fundamental Limits of Latent Chain-of-Thought

Jiaxuan Zou, Yaozhong Xiong, Yong Liu · 2026

Read Capabilities and Fundamental Limits of Latent Chain-of-Thought by Jiaxuan Zou, Yaozhong Xiong, Yong Liu (2026) 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-2602-01148.txt.

Description

Jiaxuan Zou and co-authors map the bounds of latent chain-of-thought, where models think in smooth vectors, not words. Such models shine at search yet flop at sums, a puzzle the team traces to choice sureness. High sureness drives exact runs but chokes search, while low sureness frees search but piles up slips. A symbolic index gauges that sureness and steers the trade, and the team proves staged learning is a must, since direct training fails on data drift.

Catalog record

CREATORJiaxuan Zou, Yaozhong Xiong, Yong Liu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE89093
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.01148

Browse the full library · About the Tower of Babel