AI Paper Atlas

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

The Power of Power Law: Asymmetry Enables Compositional Reasoning

Zixuan Wang, Xingyu Dang, Jason D. Lee, Kaifeng Lyu

ICML 2026 · Large language models › Language, safety and interpretability · approach: Faithfulness of LLM explanations · 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
sample skills from a power law (alpha=1.0) instead of uniform in training0.6B Qwen3-style model trained from scratch · multi-step arithmetic (k=4, no CoT), uniform testaccuracy / learning speedImproved
sample group elements from a power law (alpha=1.5) instead of uniformtransformer · S5 state tracking, no CoT or curriculumaccuracyImproved
use power-law instead of uniform skill samplingsmall GPT-like model · multi-hop QA (|E|=50, |R|=20, k=3)test accuracy vs training tokensImproved
use power-law instead of uniform number samplinglanguage model on synthetic grade-school math (iGSM, ops 2-8, mod 211) · synthetic GSMtest accuracy vs training tokensImproved
add power-law sampling on top of an explicit hop curriculum (k=1,2,3,4) (ablation)S5 state tracking with curriculum · S5 state trackingloss plateaus / training speedImproved
vary power-law exponent alpha (ablation)S5 state tracking · S5 state trackingtraining speedMixed
use a more fine-grained (more bins) power-law distribution (ablation)S5 state tracking · S5 state tracking, m in {5,10,20,40,60} binslearning speedImproved
order skills in power law lexicographically instead of random or reversed order (ablation)S5 state tracking · S5 state trackinglearning speedImproved
power-law distribution vs uniform in the final convergence stageS5 state tracking · S5 tail permutations (Bins 4, 5)test loss on tail skillsHurt

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