CrispEdit: Low-Curvature Projections for Scalable Non-Destructive LLM Editing
2026-02-17 • Machine Learning
Machine LearningArtificial Intelligence
AI summaryⓘ
The authors address a problem in large language model (LLM) editing where improving a model's behavior can unintentionally harm its general abilities. They introduce CrispEdit, a new method that carefully changes the model while explicitly protecting its overall skills. Their approach uses a smart math technique called constrained optimization and a way to measure capability that avoids large errors. By efficiently applying these methods to large models, they achieve precise edits with minimal negative effects, outperforming earlier techniques in tests.
large language modelsmodel editingcapability preservationconstrained optimizationBregman divergenceGauss-Newton HessianKronecker-factored approximate curvatureprojection methodssecond-order optimization
Authors
Zarif Ikram, Arad Firouzkouhi, Stephen Tu, Mahdi Soltanolkotabi, Paria Rashidinejad
Abstract
A central challenge in large language model (LLM) editing is capability preservation: methods that successfully change targeted behavior can quietly game the editing proxy and corrupt general capabilities, producing degenerate behaviors reminiscent of proxy/reward hacking. We present CrispEdit, a scalable and principled second-order editing algorithm that treats capability preservation as an explicit constraint, unifying and generalizing several existing editing approaches. CrispEdit formulates editing as constrained optimization and enforces the constraint by projecting edit updates onto the low-curvature subspace of the capability-loss landscape. At the crux of CrispEdit is expressing capability constraint via Bregman divergence, whose quadratic form yields the Gauss-Newton Hessian exactly and even when the base model is not trained to convergence. We make this second-order procedure efficient at the LLM scale using Kronecker-factored approximate curvature (K-FAC) and a novel matrix-free projector that exploits Kronecker structure to avoid constructing massive projection matrices. Across standard model-editing benchmarks, CrispEdit achieves high edit success while keeping capability degradation below 1% on average across datasets, significantly improving over prior editors.