EXACT: Explicit Attribute-Guided Decoding-Time Personalization
Xin Yu, Hanwen Xing, Lingzhou Xue · 2026
Read EXACT: Explicit Attribute-Guided Decoding-Time Personalization by Xin Yu, Hanwen Xing, Lingzhou Xue (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-2602-17695.txt.
Description
Xin Yu and team introduce EXACT, a decoding-time personalization method for large language models. Personalized alignment must adapt to each user's shifting context, yet past decoding-time methods lean on implicit, hard-to-read taste encodings and rigid user profiles that ignore how tastes shift across prompts.
EXACT aligns output with limited pairwise taste feedback through a predefined set of plain attributes. An offline stage picks user-specific attribute subsets by maximizing the likelihood of preferred replies. At run time, it retrieves the most relevant attributes for each prompt and injects them into context to steer output. Theory backs the method under mild assumptions, and tests on human-annotated taste data show it beating strong baselines on taste modeling and personalized output quality.
Catalog record
| CREATOR | Xin Yu, Hanwen Xing, Lingzhou Xue |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 67 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-17695