ANTHROPIC REVEALS 'GLOBAL WORKSPACE' ARCHITECTURE IN LANGUAGE MODELS

Anthropic researchers have identified a global workspace mechanism in language models that allows different neural components to share information during processing. The discovery provides new insights into how AI systems organize and communicate information internally.