Reading Between the Tokens: Improving Preference Predictions through Mechanistic Forecasting
Sarah Ball, Simeon Allmendinger, Niklas Kühl, Frauke Kreuter · 2026
Read Reading Between the Tokens: Improving Preference Predictions through Mechanistic Forecasting by Sarah Ball, Simeon Allmendinger, Niklas Kühl, Frauke Kreuter (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-2602-02882.txt.
Description
Sarah Ball and 3 co-authors test a different way to predict human preferences with language models in a 2026 paper. Instead of only reading what a model outputs, they probe its internal representations. Election forecasting serves as their test case for the method, which they call mechanistic forecasting.
The team ran millions of configurations across seven models and six national elections. They varied persona traits and prompt wording. They tracked how demographic and ideological details activate internal components that encode political parties. Using that internal signal sometimes beat surface predictions. Effects differed by attribute type, party, country, and model. The findings suggest model internals hold usable information about human preferences.
Catalog record
| CREATOR | Sarah Ball, Simeon Allmendinger, Niklas Kühl, Frauke Kreuter |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-02882