Mirage: The Illusion of Visual Understanding
Mohammad Asadi, Jack W. O’Sullivan, Fang Cao, Tahoura Nedaee, Kamyar Rajabalifardi, Fei-Fei Li, Ehsan Adeli, Euan Ashley, [6pt] Department of Electrical Engineering, Stanford University, CA, USA, [4pt] Equal contributions · 2026
Read Mirage: The Illusion of Visual Understanding by Mohammad Asadi, Jack W. O’Sullivan, Fang Cao, Tahoura Nedaee, Kamyar Rajabalifardi, Fei-Fei Li, Ehsan Adeli, Euan Ashley, [6pt] Department of Electrical Engineering, Stanford University, CA, USA, [4pt] Equal contributions (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-2603-21687.txt.
Description
Asadi and co-authors report a failure mode in multimodal AI they call mirage reasoning. Frontier models happily describe images they were never shown, even inventing pathology-flavored findings. Without any image input, models still scored strikingly high on general and medical vision benchmarks. One model even topped a chest X-ray question benchmark with no images at all. When told plainly to guess, the same models turned cautious and scored far worse. The authors urge private, cue-free benchmarks and propose their B-Clean design for fair visual evaluation.
Catalog record
| CREATOR | Mohammad Asadi, Jack W. O’Sullivan, Fang Cao, Tahoura Nedaee, Kamyar Rajabalifardi, Fei-Fei Li, Ehsan Adeli, Euan Ashley, [6pt] Department of Electrical Engineering, Stanford University, CA, USA, [4pt] Equal contributions |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 71546 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.21687