AI Paper Atlas

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

Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainers

Adam Karvonen, James Chua, Clément Dumas, Kit Fraser-Taliente, Subhash Kantamneni, Julian Minder, Euan Ong, Arnab Sen Sharma, Daniel Wen, Owain Evans, Samuel Marks

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 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
use a generalist Activation Oracle instead of best white-box baselineActivation Oracle · Taboo (Gemma-2-9B-IT)accuracy: 0.041 -> 0.205Improved
use Activation Oracle instead of best white-box baselineActivation Oracle · Secret Side Constraint (Llama-3.3-70B)accuracy: 0.522 -> 0.590Improved
use Activation Oracle instead of best black-box prefill methodActivation Oracle · Secret Side Constraint (Llama-3.3-70B)accuracy: 0.968 -> 0.590Hurt
replace ADL with Activation Oracle for model diffingemergent misalignment audit · emergent misalignment auditgrade (1-5): 2.03 -> 2.00No clear effect
add classification and context prediction tasks to the AO training mixture (ablation)Activation Oracle · 12 model/evaluation combinationsaccuracy: 10 of 12 best with full mixtureImproved
train on diverse mixture instead of SPQA only (ablation)Activation Oracle · PersonaQA and held-out classificationaccuracy: better in 7 of 8 settingsImproved
inject activations at layer 1 instead of layer 0 (ablation)Activation Oracle · Secret Side Constraintaccuracy: +11%Improved
use activations from layer 75% instead of 50% at evaluation (ablation)Activation Oracle · Taboo (Gemma-2-9B-IT)accuracy: 0.205 -> 0.056Hurt
use full-sequence activations instead of single-token input (ablation)Activation Oracle · secret-keeping evaluationsaccuracyImproved
filter predictions by majority-vote consensus >=0.9Activation Oracle · Taboo (Qwen3-8B)accuracy: 47% -> 97%Improved
scale training mixture diversity and quantity (ablation)Activation Oracle (Qwen3-8B) · controlled data ablationaccuracyImproved
use layers 0% and 10% as input depth (ablation)Activation Oracle · classificationaccuracyHurt

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