AI Paper Atlas

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

Narrow Finetuning Leaves Clearly Readable Traces in the Activation Differences

Julian Minder, Clément Dumas, Stewart Slocum, Helena Casademunt, Cameron Holmes, Robert West, Neel Nanda

ICLR 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling 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
give interpretability agent ADL results (Patchscope, Logit Lens, steering) versus blackbox onlygpt-5 interpretability agent · finetuning objective identification (grade >= 2)% organisms: 39 (blackbox i=50) -> 91 (ADL i=5)Improved
give ADL agent zero extra model interactions (ablation)ADL agent · grade >= 2% organisms: 91 -> 82No clear effect
increase interaction budget (ablation)interpretability agent · agent gradesgradeImproved
switch agent LLM from GPT5 to Gemini2.5 Pro (ablation)interpretability agent · agent gradesgradeNo clear effect
compare base model instead of chat model against finetuned chat model (ablation)ADL on SDF organisms · agent gradegradeNo clear effect
remove bias direction (causal ablation) on finetuning data (ablation)finetuned model · finetuning data CE losscross-entropyHurt
remove bias direction on pretraining data (ablation)finetuned model · pretraining data CE losscross-entropyMixed
mix unrelated pretraining data into SDF finetuning (ablation)SDF model organisms · bias traces (steering, token relevance)similarity / relevance: 1:0.1 ratio already significant reductionImproved
mix unrelated data at 1:0.1 ratio (ablation)Llama 3.2 1B SDF · False Fact AlignmentFFA scoreHurt
reduce number of finetuning samples (ablation)SDF organisms · bias and fact alignmentbias / FFAMixed
concept ablation during finetuning (CAFT) (ablation)SDF organisms · bias tracesbiasNo clear effect
apply ADL to broad chat finetuningchat-tuned models · trace detectabilitydetectabilityHurt

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