04:00
2026-08-26
arxiv.org
large-language-models
Discovering Cross-Language Reasoning Invariance in LLMs with Geometry-Invariant Sparse Autoencoders
A new arXiv preprint (2608.23809v1) investigates whether multilingual LLMs use shared or language-specific features for mathematical reasoning, introducing the Geometry-Invariant Sparse Autoencoder (G…