AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Precise and Interpretable Editing of Code Knowledge in Large Language Models

Min Xue, Nikolai Bolik, Lennart Stöpler, Erik Imgrund, Janik Schmid, Artur Andrzejak

ICLR 2026 · Large language models › Retrieval and knowledge · approach: Knowledge-augmented question answering

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 MLP with sparse TransCoder and update only active neurons (TCPE LTC4) instead of ROMEROME · G4GD multi-error editingScore: 109.60 -> 171.45Improved
TCPE with TransCoder instead of ROME on MLPROME · G4GD multi-error editingpost-edit accuracy: 57.33 -> 64.00Improved
use TCPE instead of MEMIT/PMET multi-layer editsMEMIT · G4GD multi-error editingScore: 14.96 -> 171.45Improved
widen TransCoder from 4x to 8x (ablation)TCPE · G4GD multi-error editingScore: 171.45 -> 174.82Improved
use TCPE instead of ROME, HumanEval reliabilityROME · HumanEval after G4GD editsreliability (RE): 93.61 -> 100.00Improved
use TCPE instead of MEMIT, number of updated neuronsMEMIT · HumanEval after G4GD editsupdated neurons: 55,040 -> 69Improved
replace MLP layer with TransCoder (layer 19, LTC4) (ablation)CodeLlama · G4GD overall accuracyaccuracy: 57.50% -> 58.33%No clear effect
update only highly active neurons (acv>0.2) instead of all active neurons (ablation)TCPE LTC4 · G4GD multi-error editingGeneralization (GN): 24.29 -> 32.86No clear effect
restrict update to low-activation or inactive neurons only (acv<=0.1) (ablation)TCPE LTC4 · G4GD multi-error editingGeneralization (GN): 32.86 -> 7.14Hurt
use TransCoder features instead of MLP neuronsCodeLlama · Interpretable features out of 50count interpretable: 4 -> 33Improved
lower activation threshold to include all neurons (acv>0 vs acv>0.08) (ablation)TCPE LTC4 C8 single-error · G4GD single-error editing C8Generalization (GN): 34.62 -> 28.85Hurt

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