Learning to Think from Multiple Thinkers

Nirmit Joshi, Roey Magen, Nathan Srebro, Nikolaos Tsilivis, Gal Vardi · 2026

Read Learning to Think from Multiple Thinkers by Nirmit Joshi, Roey Magen, Nathan Srebro, Nikolaos Tsilivis, Gal Vardi (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-2604-24737.txt.

Description

Joshi, Magen, Srebro, Tsilivis and Vardi study learning from chain-of-thought supervision by multiple thinkers, all correct but in set ways. Some classes are easy to learn from one thinker's chain of thought but hard from end results alone. Under cryptographic assumptions, chain-of-thought data can make learning hard by two or a few thinkers in passive settings. An active learning algorithm learns well with little chain-of-thought data per thinker, independent of target accuracy, plus a moderate number of thinkers and enough end-result data.

Catalog record

CREATORNirmit Joshi, Roey Magen, Nathan Srebro, Nikolaos Tsilivis, Gal Vardi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE262849
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2604.24737

Browse the full library · About the Tower of Babel