Language-Agnostic Representations Show a Shared Semantic Workspace
The standard story is that LLMs work in words. They predict the next token, so surely their internal reasoning is…
multilingual LLM interpretability
The study of how large language models process, represent, and explain behavior across multiple languages.