Study Maps How Large Language Models Encode Arithmetic in Their Activations
A new arXiv paper examines the geometry of arithmetic representations in large language models, tracing how linear probes can reveal structures such as helices and circles in model activations.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-11 05:00 · arXiv cs.LG
Linear Probes to Bayesian Manifolds: Geometry of Arithmetic in Large Language Models