Research map / Large language models / Alignment and preferences
Reading Between the Tokens: Improving Preference Predictions through Mechanistic Forecasting
Sarah Ball, Simeon Allmendinger, Frauke Kreuter, Niklas Kühl
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use value probe instead of generic regression readout on activations (ablation) | party probes · Party probes (held-out split) | macro-F1: 0.93 ± 0.05 -> 0.99 ± 0.01 | Improved |
| train probes on meaningful party labels rather than shuffled labels (ablation) | party probes · Party probes (held-out split) | macro-F1: 0.33 -> 0.99 | Improved |
| use trained probe weights rather than random weights (ablation) | party probes · Party probes (held-out split) | macro-F1: 0.24 -> 0.99 | Improved |
| use mechanistic forecasting (latent activations) instead of next-token probabilities | LLM persona election forecasting · 6 national elections, 7 models | attribute-level win-rate vs survey | Mixed |
| apply mechanistic forecasting to base instead of instruction-tuned models (ablation) | mechanistic forecasting · 6 national elections | win-rate | Improved |
| add in-context party-position data to probability-based baseline (ablation) | probability-based estimation · 6 national elections | closeness to survey | No clear effect |
| use latent estimators for demographic attributes | mechanistic forecasting · US and UK, age/education/employment | win-rate | Improved |
| use latent estimators for opinion-based attributes | mechanistic forecasting · political orientation, immigration, inequality across countries | win-rate | Mixed |
| gate mechanistic forecasting by normalized entropy above 0.85 (ablation) | mechanistic forecasting · high-entropy attributes | median error improvement | Improved |
| use latent estimators for low-entropy attributes | mechanistic forecasting · low-entropy attributes | error vs survey | No clear effect |
| restrict probe layers to 0.5L-0.9L window (ablation) | party probes · selectivity gap across seeds, models, countries | selectivity gap lower whisker | Improved |
| add sign-inversion gating of value vectors (ablation) | value vector selection · intervention experiments | intervention effect consistency | Improved |
Closest papers by meaning
- Examining Alignment of Large Language Models through Representative Heuristics: the case of political stereotypes (ICLR 2025 · similarity 0.90)
- Prompts to Proxies: Emulating Human Preferences via a Compact LLM Ensemble (NeurIPS 2026 · similarity 0.86)
- Latent Preference Coding: Aligning Large Language Models via Discrete Latent Codes (ICML 2025 · similarity 0.86)
- Generative Social Choice: The Next Generation (ICML 2025 · similarity 0.86)
- Embeddings for Preferences, Not Semantics (NeurIPS 2026 · similarity 0.86)
- Linear Representations of Political Perspective Emerge in Large Language Models (ICLR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
