AI Paper Atlas

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

Modular Pretraining Enables Access Control

Ethan Roland, Murat Cubuktepe, Erick Martinez, Stijn Servaes, Keenan Pepper, Michael Vaiana, Diogo S de Lucena, Judd Rosenblatt, Addie Foote, Cem Anil, Alex Cloud

ICML 2026 · Large language models › Language, safety and interpretability · approach: Data and evaluation methodology · 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
use GRAM instead of data filteringGRAM · Simple Storiesretain compute ratio: 0.96 -> 0.95No clear effect
add a shared always-active core MLP (GRAM vs DEMix) (ablation)GRAM · Simple Storiesretain compute ratio: 0.183 -> 0.952Improved
use GRAM instead of FT-LoRAGRAM · Simple Storiescore compute ratio: 0.841 -> 0.938Improved
use GRAM instead of MaxEnt unlearningGRAM · Simple Storieselicited forget compute ratio: 0.983 -> 0.855Improved
use GRAM instead of data filtering with 50% labeled dataGRAM · Dual-use 800M, 50% labeledforget compute ratio: 0.85 ± 0.01 -> 0.60 ± 0.01Improved
use GRAM instead of FT-LoRA with multiple modules activeGRAM · Dual-use 800M, all auxiliary activeretain compute ratioImproved
increase number of auxiliary categories from 4 to 20 (ablation)GRAM · Simple Storiesretain compute ratio: 4 -> 20 categoriesNo clear effect
increase number of auxiliary categories from 4 to 20 (ablation)GRAM · Simple Storiesforget compute ratioHurt
scale model from 50M to 5B parametersGRAM and data filtering · Dual-use scaling (Virology retained)forget compute ratioImproved
use GRAM instead of data filtering at 5BGRAM · Dual-use scaling 5Bretain (virology) compute ratioMixed
use MaxEnt post-hoc unlearning instead of GRAMMaxEnt · Simple Storiesforget compute ratio: 0.766 -> 0.840Hurt

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