OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes
Sepehr Dehdashtian, Gautam Sreekumar, Vishnu Naresh Boddeti · 2025
Read OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes by Sepehr Dehdashtian, Gautam Sreekumar, Vishnu Naresh Boddeti (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-2501-00962.txt.
Description
Sepehr Dehdashtian, Gautam Sreekumar, and Vishnu Naresh Boddeti measure stereotypes in pictures made by text-to-image models, in a paper published in two thousand twenty-five. Such models often bake in stock images of culture, jobs, and other ideas, but older measures blur the line between plain bias and a true stereotype.
The authors define a stereotype measure that matches how sociology uses the word. Their toolkit, OASIS, scores a set of generated images in two ways. One score checks whether stereotypical traits appear in a skewed spread. The other reads how much the images vary along such a trait. OASIS also includes methods that trace where inside the model these patterns come from. The title sums up the finding: the newest models make crisp pictures, yet the same old stereotypes persist.
Catalog record
| CREATOR | Sepehr Dehdashtian, Gautam Sreekumar, Vishnu Naresh Boddeti |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-00962