AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Spherical Steering: Geometry-Aware Activation Rotation for Language Models

Zejia You, Chunyuan Deng, Hanjie Chen

ICML 2026 · Large language models › Language, safety and interpretability · approach: Pruning and circuits in LLMs · 0 citations (OpenAlex)

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
replace additive steering (CAA) with norm-preserving rotationLLaMA-3.1-8B-Instruct · TruthfulQA MC averageaccuracy: 40.54 -> 53.17Improved
replace additive steering (ITI) with rotationLLaMA-3.1-8B-Instruct · TruthfulQA open-endedTRUE x INFO: 40.31 -> 54.63Improved
add ITI additive steering to base modelLLaMA-3.1-8B-Instruct · TruthfulQA open-endedTRUE x INFO: 48.24 -> 40.31Hurt
apply Spherical SteeringLLaMA-3.1-8B-Instruct · COPAaccuracy: 83.00 -> 95.00Improved
apply Spherical SteeringLLaMA-3.1-8B-Instruct · MMLUaccuracy: 60.60 -> 62.05Improved
add vMF confidence gate to rotation (ablation)Spherical Steering · TruthfulQA (Qwen2.5-7B-Instruct) at high strengthTRUE x INFOImproved
add vMF confidence gate to rotation (ablation)Spherical Steering · TruthfulQA (Qwen2.5-7B-Instruct) MCMC accuracy: peak at alpha 0.7 -> 0.8Improved
steer 2-3 layers instead of 1 (ablation)Spherical Steering · TruthfulQA (LLaMA-3.1-8B-Instruct)INFO: 62.92 -> 92.70Improved
steer 4 or more layers (ablation)Spherical Steering · TruthfulQA (LLaMA-3.1-8B-Instruct)MC1: 47.62 -> 41.37Hurt
increase contrastive data from 25 to 300 questions (ablation)Spherical Steering · TruthfulQA (LLaMA-3.1-8B-Instruct)MC1: 51.52 -> 53.92Improved
increase rotation strength alpha from 0.6 to 1.0 (ablation)Spherical Steering · TruthfulQA open-ended (LLaMA-3.1-8B, layer 24)TRUE x INFO: 53.98 -> 11.37Hurt
increase rotation strength alpha from 0.4 to 1.0 (ablation)Spherical Steering · TruthfulQA MC1 (LLaMA-3.1-8B, layer 24)MC1: 32.44 -> 46.15Improved

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