Superposition Yields Robust Neural Scaling

NeurIPSOral2025

Authors: Yizhou Liu, Ziming Liu, Jeff Gore
Affiliation: Massachusetts Institute of Technology
Venue: NeurIPS 2025
Track: Oral

TL;DR

Neural scaling law in LLMs is explained through representation interference due to superposition…

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

scaling law llm

← All NeurIPS 2025 Oral papers · Browse the whole archive