See, Think, Learn: A Self-Taught Multimodal Reasoner

Sourabh Sharma, Sonam Gupta, Sadbhawna[4pt] [email protected] [email protected] [email protected], [4pt] Malaviya National Institute of Technology Jaipur · 2025

Read See, Think, Learn: A Self-Taught Multimodal Reasoner by Sourabh Sharma, Sonam Gupta, Sadbhawna[4pt] [email protected] [email protected] [email protected], [4pt] Malaviya National Institute of Technology Jaipur (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-2512-02456.txt.

Description

A vision-language model must both see accurately and reason soundly, and weakness in either caps the whole. Sourabh Sharma, Sonam Gupta, and Sadbhawna present in 2025 a self-taught multimodal reasoner, built to improve its own perception and reasoning together rather than importing the skill from outside.

Earlier efforts strengthened reasoning while leaving perception to chance, so models reasoned elegantly about things they had mis-seen. A self-taught loop promises gains without proportionally more labeled data. Multimodal systems that check their own seeing before thinking could become markedly more reliable on visual questions.

Catalog record

CREATORSourabh Sharma, Sonam Gupta, Sadbhawna[4pt] [email protected] [email protected] [email protected], [4pt] Malaviya National Institute of Technology Jaipur
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-02456

Browse the full library · About the Tower of Babel