AI Paper Atlas

Research map / Large language models / Alignment and preferences

What Do Large Language Models Know About Opinions?

Erfan Jahanparast, Zhiqing Hong, Serina Chang

ICLR 2026 · Large language models › Alignment and preferences · approach: LLM-as-judge and feedback data

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 changedWhereMeasuredEffect
probe residual stream activations instead of reading next-token probabilitiesLlama-3.1-8B · OpinionQA/SubPOP 2-optionKL divergence: 0.223 -> 0.073Improved
probe residual stream activations instead of reading next-token probabilitiesLlama-3.1-8B · OpinionQA/SubPOP 3-optionKL divergence: 0.241 -> 0.113Improved
use a nonlinear MLP probe instead of logistic regression (ablation)linear probe · OpinionQA/SubPOP 2-optionKL divergence: 0.073 -> 0.068No clear effect
use a single shared probe for all groups instead of per-group probes (ablation)Llama-3.1-8B probe · OpinionQA/SubPOP 2-optionKL divergence: 0.071 -> 0.074No clear effect
switch prompt format from QA to BIO (ablation)Llama-3.1-8B probe · OpinionQA/SubPOP 2-optionKL divergence: 0.143 -> 0.067Improved
switch prompt format from QA to PORTRAY (ablation)Llama-3.1-8B probe · OpinionQA/SubPOP 3-optionKL divergence: 0.219 -> 0.119Improved
fine-tune only the unembeddings (ablation)Llama-3.1-8B · OpinionQA/SubPOP 2-optionKL divergence: 0.213 -> 0.082Improved
fine-tune only the unembeddings (ablation)Llama-3.1-8B · OpinionQA/SubPOP 3-optionKL divergence: 0.234 -> 0.126Improved
LoRA fine-tune the full model rather than probing (ablation)Llama-3.1-8B · OpinionQA/SubPOP 2-optionKL divergence: 0.213 -> 0.046Improved
probe on GlobalOpinionQA country opinionsLlama-3.1-8B · GlobalOpinionQA 2-optionKL divergence: 0.150 -> 0.092Improved
probe on GlobalOpinionQA country opinionsLlama-3.1-8B · GlobalOpinionQA 3-optionKL divergence: 0.193 -> 0.142Improved
intervene on all layers 2-30 instead of only layers 9-18 (ablation)SAE steering · Steering avg improvementimprovement ratio: 0.440 vs. 0.463No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy