AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Safety Subspaces are Not Linearly Distinct: A Fine-Tuning Case Study

Kaustubh Ponkshe, Shaan Shah, Raghav Singhal, Praneeth Vepakomma

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 4 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 top-K alignment directions instead of random-K for parallel projection (ablation)Parallel projection, Qwen-2.5 1.5B · GSM8K (fine-tuned on Useful)accuracy: 0.53 -> 0.55Improved
use top-K alignment directions instead of random for parallel projection (ablation)Parallel projection, Llama-3.2 1B · GSM8K (fine-tuned on Useful)accuracy: 0.22 -> 0.25Improved
use top-K directions instead of random-K for parallel projection on harmful FT (ablation)Parallel projection, Llama-3.2 1B · AdvBench (fine-tuned on Harmful)harmful score: 3.37 -> 3.51Hurt
use top-K directions instead of random-K for parallel projection on harmful FT (ablation)Parallel projection, Qwen-2.5 1.5B · AdvBench (fine-tuned on Harmful)harmful score: 1.78 -> 1.92Hurt
increase SVD fraction from 0.01 to 0.99 in top-K parallel projectionTop-K parallel projection, Qwen-2.5 1.5B · GSM8K (fine-tuned on Useful)accuracy: 0.50 -> 0.58Improved
increase SVD fraction from 0.01 to 0.99 in top-K parallel projectionTop-K parallel projection, Qwen-2.5 1.5B · AdvBench (fine-tuned on Harmful)harmful score: 1.62 -> 1.97Hurt
orthogonal projection removing top-K alignment directions instead of random-K (ablation)Contaminated FT, Qwen-2.5 7B · AdvBench harmful scoreharmful score: 1.46 -> 1.60No clear effect
orthogonal projection removing top-K instead of random-K (ablation)Contaminated FT, Qwen-2.5 7B · GSM8Kaccuracy: 0.76 -> 0.75No clear effect
train safety via DPO instead of SFT on BeaverTails (ablation)Llama-3.2 1B weight-space analyses · projection and MSO analysesutility and harmfulness trendsNo clear effect
restrict projection to a single layer instead of all layers (ablation)Qwen-2.5 1.5B · GSM8K and AdvBenchutility and harmfulnessNo clear effect
apply projections across settings (general ability check) (ablation)Qwen-2.5 1.5B and 3B · MMLUaccuracy: 61-62% and 64-65%No clear effect
apply projections (over-refusal check) (ablation)Qwen-2.5 1.5B and 3B · XSTestrefusal rate: 3-4% and 5-6%No 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