Detect Before You Leap: Mirage Detection in Vision–Language Models

Sayeed Shafayet Chowdhury, Md. Shaown Miah, S. M. Taiabul Haque, Syed Ishtiaque Ahmed · 2026

Read Detect Before You Leap: Mirage Detection in Vision–Language Models by Sayeed Shafayet Chowdhury, Md. Shaown Miah, S. M. Taiabul Haque, Syed Ishtiaque Ahmed (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-2606-00435.txt.

Description

Miah and co-authors catch sight-word models that reply with no true sight proof at hand. Such sure-but-blind replies are called mirages in the field. The team study how to flag a mirage before a reply ships. Their aim is a go or no-go call on each reply.

Their tool, TC-LIA, tracks ask-image fit through the layers of a frozen CLIP tool. Four marks sum that track: end fit, late top fit, gain, and slope. No task tags train the tool at ship time. On near 19,000 cases from ten fields, mirage rates start high. TC-LIA cuts them to a small share on its own. A blend with blank checks and self-grades does yet better, and ports across tools.

Catalog record

CREATORSayeed Shafayet Chowdhury, Md. Shaown Miah, S. M. Taiabul Haque, Syed Ishtiaque Ahmed
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE106 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.00435

Browse the full library · About the Tower of Babel