Can Large Language Models Make Everyone Happy?
Usman Naseem, Gautam Siddharth Kashyap, Ebad Shabbir, Sushant Kumar Ray, Abdullah Mohammad, Rafiq Ali · 2026
Read Can Large Language Models Make Everyone Happy? by Usman Naseem, Gautam Siddharth Kashyap, Ebad Shabbir, Sushant Kumar Ray, Abdullah Mohammad, Rafiq Ali (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-2602-11091.txt.
Description
Safety, values, and culture rarely pull in the same direction. Usman Naseem and five co-authors ask whether one language model can satisfy all three at once. Most tests check each dimension alone, which hides the trade-offs between them.
Their answer is MisAlign-Profile, a single benchmark built on a data set spanning more than one hundred normative domains. Each prompt carries domain labels and a semantic type, with paired aligned and misaligned replies. Tests of general, tuned, and open-weight models reveal clear cross-dimension trade-offs. The work gives researchers a way to measure clashes that single-focus tests miss.
Catalog record
| CREATOR | Usman Naseem, Gautam Siddharth Kashyap, Ebad Shabbir, Sushant Kumar Ray, Abdullah Mohammad, Rafiq Ali |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-11091