A global workspace in language models

Interpretability research on Claude's internal thoughts.

Read on anthropic.com