AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

KnowledgeSmith: Uncovering Knowledge Updating in LLMs with Model Editing and Unlearning

Yinyi Luo, Zhexian Zhou, Hao Chen, Kai Qiu, Marios Savvides, Sharon Li, Jindong Wang

ICLR 2026 · Large language models › Retrieval and knowledge · approach: Knowledge editing in LLMs · 2 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 editing instead of unlearningLLM knowledge update · KnowledgeSmith over-spreading (CCR)share of failures: 15 -> 35Hurt
use unlearning instead of editingLLM knowledge update · KnowledgeSmith under-forgetting (RR)share of failures: 20 -> 35Hurt
use editing instead of unlearningLLM knowledge update · KnowledgeSmith conflict emergenceshare of failures: 12 -> 30Hurt
use editing instead of LoRA at k=10 samplesLlama3-8B-Instruct · ID accuracyaccuracy (%): 16.7 -> 25Improved
increase LoRA training samples from 1 to 1000Llama3-8B-Instruct LoRA · OOD accuracy (MMLU)accuracy (%): 63.0 -> 61.6Hurt
increase LoRA training samples to 1000Llama3-8B-Instruct LoRA · ID accuracyaccuracy (%): 12.5Hurt
use editing instead of unlearningLLM knowledge update · training time on 1000 samples (H100)hours: 0.2 -> 6Hurt
increase training samples from 1 to 10000 for unlearningReLearn · KL representation similaritynormalized KL similarity: 0.014 -> 0.883Improved
increase training samples from 1 to 10000 for editingAlphaEdit · KL representation similaritynormalized KL similarity: 0.140 -> 0.652Improved
use larger models for updatesLLaMA3 family · plasticity under few-shot dataimmediate plasticityHurt
apply update at root rather than leaf nodeediting and unlearning · KnowledgeSmith hierarchyachievable in-domain accuracyHurt
update history-subject knowledge instead of other subjectsediting and unlearning · KnowledgeSmith subjectsupdate accuracyHurt

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