Reading Between the Tokens: Improving Preference Predictions through Mechanistic Forecasting

2026-02-02

AI Summary

This work proposes a novel “mechanistic prediction” paradigm for forecasting human preferences, moving beyond reliance on surface-level model outputs to leverage structured information embedded within large language models’ internal representations. Using election prediction as a case study, the approach identifies and activates neural components encoding party affiliation, demographic attributes, and ideological signals. Through extensive experiments across seven prominent models, electoral data from six countries, and over 24 million configurations—combined with large-scale representation probing, cross-model and cross-national validation, and multidimensional prompt controls—the study systematically uncovers interpretable and exploitable preference signals within model internals. Results demonstrate that this method significantly outperforms conventional output-based prediction, with pronounced gains in accuracy for specific demographic groups, political parties, and national contexts.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Knowledge Representation and Reasoning: Preferences
• Humans and AI: Learning Human Values and Preferences

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Web Mining and Content Analysis: Large pretrained models with web data
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Large language models are increasingly used to predict human preferences in both scientific and business endeavors, yet current approaches rely exclusively on analyzing model outputs without considering the underlying mechanisms. Using election forecasting as a test case, we introduce mechanistic forecasting, a method that demonstrates that probing internal model representations offers a fundamentally different - and sometimes more effective - approach to preference prediction. Examining over 24 million configurations across 7 models, 6 national elections, multiple persona attributes, and prompt variations, we systematically analyze how demographic and ideological information activates latent party-encoding components within the respective models. We find that leveraging this internal knowledge via mechanistic forecasting (opposed to solely relying on surface-level predictions) can improve prediction accuracy. The effects vary across demographic versus opinion-based attributes, political parties, national contexts, and models. Our findings demonstrate that the latent representational structure of LLMs contains systematic, exploitable information about human preferences, establishing a new path for using language models in social science prediction tasks.

Problem

• preference prediction
• large language models
• mechanistic forecasting
• latent representations
• human preferences

Innovation

• mechanistic forecasting
• preference prediction
• latent representations
• large language models
• election forecasting