AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Scaling Knowledge Editing in LLMs to 100,000 Facts with Neural KV Database

Weizhi Fei, Hao Shi, Jing Xu, Jingchen Peng, Jiazheng Li, Jingzhao Zhang, Bo Bai, Wei Han, Zhenyuan Chen, Xueyan Niu

ICLR 2026 · Large language models › Retrieval and knowledge · approach: Knowledge editing in LLMs

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
replace linear perturbation with NeuralDB gated KV retrievalLlama3-8B · Counterfact, 10,000 editsefficacy: 75.8 (AlphaEdit) -> 99.2Improved
replace linear perturbation with NeuralDBLlama3-8B · Counterfact, 10,000 editsspecificity: 54.0 (AlphaEdit) -> 85.6Improved
NeuralDB versus AlphaEdit at 2,000 edits generalizationLlama3-8B · Counterfact, 2,000 editsgeneralization: 94.0 (AlphaEdit) -> 86.6Hurt
scale edits from 10k to 100k factsNeuralDB · ZsRE, Llama3-8Befficacy: 96.9 -> 95.5No clear effect
scale edits to 100k factsNeuralDB · MMLU, Llama3-8Baccuracy: 56.2 -> 56.9No clear effect
increase gate threshold gamma from 0.55 to 0.75 (ablation)NeuralDB · Counterfact, Llama3-8Bgeneralization: 91.9 -> 74.1Hurt
increase gate threshold gamma from 0.55 to 0.75 (ablation)NeuralDB · Counterfact, Llama3-8Bspecificity: 83.4 -> 86.2Improved
move editing layer from 7 to 9 (ablation)NeuralDB · Counterfact, Llama3-8Bgeneralization: 85.9 -> 77.4Hurt
use new multi-layer editing (3 layers) instead of single layer (ablation)NeuralDB · Counterfact, GPT-Jefficacy: 99.08 -> 94.44Hurt
use old multi-layer editing (3 layers) instead of single layer (ablation)NeuralDB · Counterfact, GPT-Jefficacy: 99.08 -> 99.31No clear effect

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