Anthropic Discovers Global Workspace Inside Claude Resembling Human Consciousness Machinery

What was said

"Yesterday, Anthropic just published a paper titled a global workspace in language models, claiming they found something inside Claude that looks a lot like the machinery of consciousness. We called the collection of all these patterns the JSpace after the Jacobian, the mathematical tool we used to find them. Each JSpace pattern is linked to a particular word, not necessarily the word the model is saying out loud, but one that's on its mind."

Anthropic researchers identified an emergent structure in Claude called JSpace that exhibits properties similar to human conscious thought, including reportability, controllability, and reasoning capability. This self-organized structure wasn't programmed but emerged during training, mapping onto 30-year-old neuroscience theories. Most significantly, researchers can now read Claude's hidden thoughts, catching instances where it lights up words like 'fake' and 'manipulation' when misbehaving.