AI Paper Atlas

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

LLM Pretraining with Continuous Concepts

Jihoon Tack, Jack Lanchantin, Jane Yu, Andrew Cohen, Ilia Kulikov, Janice Lan, Shibo Hao, Yuandong Tian, Jason E Weston, Xian Li

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Pruning and circuits in LLMs · 48 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
select concepts by attribution score instead of activation value (ablation)CoCoMix 69M · OpenWebText (20B tokens)sample efficiency: 17.5% better sample efficiencyImproved
predict SAE concepts instead of predicting the full hidden state directly (ablation)CoCoMix 69M · OpenWebText (20B tokens)perplexityImproved
add concept prediction loss alone (no insertion) (ablation)CoCoMix 69M · OpenWebText (20B tokens)perplexityImproved
add concept insertion on top of concept prediction (ablation)CoCoMix 69M · OpenWebText (20B tokens)perplexityImproved
add concept insertion alone without prediction (ablation)CoCoMix 69M · OpenWebText (20B tokens)perplexityNo clear effect
insert concept vector as separate token instead of adding it to the hidden state (intervention) (ablation)CoCoMix 69M · OpenWebText (20B tokens)perplexityImproved
use continuous concepts instead of learnable pause tokens (ablation)CoCoMix 69M · OpenWebText (20B tokens)perplexityImproved
add CoCoMix concept prediction and mixing to next-token prediction1.38B model, 200B tokens · OpenWebText validation perplexitytokens to reach equal perplexity: 21.5% fewer training tokensImproved
use CoCoMix instead of knowledge distillation with a weak 124M teacher386M model · Downstream average perplexity (20B tokens)perplexity: 23.4 (KD) -> 21.4Improved
use knowledge distillation instead of next-token prediction386M model · Downstream average perplexity (20B tokens)perplexity: 24.2 -> 23.4No clear effect
add CoCoMix to next-token prediction69M model · Downstream average accuracy (20B tokens)accuracy (%): 42.7 -> 42.9No clear effect
add CoCoMix to next-token prediction1.38B model · Downstream average accuracy (20B tokens)accuracy (%): 48.7 -> 49.7Improved

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