Do Reasoning Models Ask Better Questions? A Formal Information-Theoretic Analysis on Multi-Turn LLM Games

Daniel M. Pedrozo, Telma W. de L. Soares, Bryan L. M. de Oliveira · 2026

Read Do Reasoning Models Ask Better Questions? A Formal Information-Theoretic Analysis on Multi-Turn LLM Games by Daniel M. Pedrozo, Telma W. de L. Soares, Bryan L. M. de Oliveira (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-2601-17716.txt.

Description

Daniel M. Pedrozo, Telma W. de L. Soares, and Bryan L. M. de Oliveira ask if reasoning models ask better questions. Good questions resolve ambiguity in user requests. Language models still struggle with that skill. Past work used word games to test question quality. Those tests lack full scoring frames. The authors give a formal information-theoretic analysis. It scores both final and intermediate steps.

Information gain is their yardstick for each question. Multi-turn games let that gain build over time. A fair frame can rank models on curiosity, not just answers. Agents that ask well waste less user time. They also reach right answers with fewer turns. This work gives builders a way to train and test that skill.

Catalog record

CREATORDaniel M. Pedrozo, Telma W. de L. Soares, Bryan L. M. de Oliveira
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-17716

Browse the full library · About the Tower of Babel