AI Paper Atlas

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

Evaluating SAE interpretability without generating explanations

Gonçalo Paulo, Nora Belrose

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Linguistic encoding in LLM representations

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
finetune the embedding model on highlighted-token contexts (ablation)example embedding scoring · Correlation with human intruder scores (112 latents)Pearson correlation: 0.42 -> 0.78Improved
sample activating examples from the highest activation decile instead of the lowestintruder detection · SmolLM2 SAE latentsaccuracy: up to 20% higher accuracyImproved
use a larger evaluator LLM (Llama 3.1 70b instead of 8b) (ablation)intruder detection · Correlation with human intruder scores (56 latents)Pearson correlation: 0.52 -> 0.76Improved
use Claude Sonnet 3.5 instead of Llama 3.1 8b as intruder judge (ablation)intruder detection · Correlation with human intruder scores (56 latents)Pearson correlation: 0.52 -> 0.83Improved
use intruder detection instead of detection scoringLLM scoring · Correlation with human intruder scores (105 latents)Pearson correlation: 0.48 -> 0.75Improved
use example embedding scoring instead of LLM intruder detectionscoring method · Correlation with human intruder scores (105 latents)Pearson correlation: 0.75 -> 0.78No clear effect
show an intruder from a different decile instead of non-activating examplesintruder detection · Deciles intruder taskaccuracy: near random for close decilesHurt
apply to other SAE families (Gemmascope, Pythia transcoders)example embedding scoring · Other sparse codersPearson correlation: 0.84 (Gemma 2 9b), 0.67 (Pythia-160m) with LLM intruderNo 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