AI Paper Atlas

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

Causality ≠ Invariance: Function and Concept Vectors in LLMs

Gustaw Opielka, Hannes Rosenbusch, Claire E Stevenson

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Reasoning mechanisms and failure modes

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
select heads by RSA (concept consistency) instead of activation patching to form vectorsFunction Vectors · Format invariance, 4 LLMsconcept vs format RSAImproved
extract vector from different input format (OE vs MC)Function Vectors · Llama 3.1 70B similarity matricescosine similarity: near orthogonal across formats; within-format 0.90Hurt
use RSA-selected CV heads vs AIE-selected FV headsConcept Vectors · Top-K head overlapnumber of overlapping heads: 0 overlap for K<=10 in all modelsNo clear effect
use CV instead of FV for OOD steering (MC, French)Concept Vectors · AmbiguousICLdelta P / KLImproved
use FV instead of CV for in-distribution steeringFunction Vectors · AmbiguousICL OE-ENGdelta P: salty->sweet: 49 -> 56 (+%)Improved
cross-format activation patching (ablation)Activation patching · Llama 3.1 70Bidentified heads: 0 overlap with CV top-5No clear effect
apply CV to zero-shot steeringConcept Vectors · Zero-shot steeringdelta PNo clear effect
extract FV from French/MC prompts and steerFunction Vectors · AmbiguousICLformat tokens: MC: '(' +53%; OE-FR: 'su' +31%Hurt
exclude Qwen 2.5 72B outlier datasets from AIE (ablation)Activation patching · Qwen 2.5 72Btop-5 head ranking: 100% overlapNo 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