AI Paper Atlas

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

CrispEdit: Low-Curvature Projections for Scalable Non-Destructive LLM Editing

Zarif Ikram, Arad Firouzkouhi, Stephen Tu, Mahdi Soltanolkotabi, Paria Rashidinejad

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling data · 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
add low-curvature (K-FAC nullspace) projection to fine-tuning (ablation)LLaMA-3-8B-Instruct fine-tuning · ZsRE 3000 edits (WILD)reliability and base capabilityImproved
use CrispEdit instead of AlphaEditLLaMA-3-8B-Instruct · ZsRE 3000 editsMMLU: 52.7 -> 69.5Improved
use CrispEdit instead of AlphaEditLLaMA-3-8B-Instruct · ZsRE 3000 editsQA-context reliability: 70.1 -> 80.5Improved
scale edits from 3000 to 10000 with CrispEdit (ablation)CrispEdit · ZsREQA-context reliability: 80.5 -> 77.4No clear effect
scale edits from 3000 to 10000 with LocBF-FT (ablation)LocBF-FT · ZsREQA-context reliability: 69.5 -> 53.5Hurt
scale edits from 3000 to 10000 with AlphaEdit (ablation)AlphaEdit · ZsREQA-context reliability: 70.1 -> 0.3Hurt
vary energy threshold gamma from 0.5 to 0.99 (ablation)CrispEdit · LLaMA-3-8B-Instructbase capabilityNo clear effect
vary capability dataset size n from 10 to 100,000 (ablation)CrispEdit · LLaMA-3-8B-Instructbase capabilityNo clear effect
apply CrispEdit to Qwen-2.5-1.5B-Instruct (ablation)CrispEdit · ZsRE 3000 editsedit performance and base capabilityImproved
use sequential editing variant CrispEdit-Seq instead of one-shot (ablation)CrispEdit · ZsRE 3000 edits (WILD, no context)reliability: 57.4 -> 72.8Improved
use CrispEdit-Seq instead of one-shot CrispEdit (ablation)CrispEdit · CounterFact 3000 editsQA-context reliability: 79.4 -> 66.5Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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